Remove the unwraps, asserts and return actual io errors

Move to a two bytes header
Introduce a first working version of the DeBitmapCodec
2025-11-22 12:46:53 +00:00 · 2025-11-10 18:37:29 +01:00 · 2025-11-10 17:56:37 +01:00 · 2025-11-10 16:56:19 +01:00 · 2025-11-10 09:45:30 +00:00 · 2025-11-10 09:33:31 +01:00
43 changed files with 2276 additions and 203 deletions
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -1128,9 +1128,9 @@ dependencies = [

 [[package]]
 name = "charabia"
-version = "0.9.7"
+version = "0.9.8"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7c2f456825b7f15eac01a1cae40c12c3f55e931d4327e6e4fa59508d664e9568"
+checksum = "bbdc8cd8f999e8b8b13ed71d30962bbf98cf39e2f2a9f1ae1ba354199239d66e"
 dependencies = [
 "aho-corasick",
 "csv",
@@ -1139,7 +1139,6 @@ dependencies = [
 "irg-kvariants",
 "jieba-rs",
 "lindera",
- "once_cell",
 "pinyin",
 "serde",
 "slice-group-by",
@@ -1824,7 +1823,7 @@ version = "0.4.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "415b6ec780d34dcf624666747194393603d0373b7141eef01d12ee58881507d9"
 dependencies = [
- "phf",
+ "phf 0.11.3",
 ]

 [[package]]
@@ -2009,6 +2008,16 @@ dependencies = [
 "syn 2.0.106",
 ]

+[[package]]
+name = "env_logger"
+version = "0.8.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "a19187fea3ac7e84da7dacf48de0c45d63c6a76f9490dae389aead16c243fce3"
+dependencies = [
+ "log",
+ "regex",
+]
+
 [[package]]
 name = "equivalent"
 version = "1.0.2"
@@ -2838,9 +2847,9 @@ checksum = "2304e00983f87ffb38b55b444b5e3b60a884b5d30c0fca7d82fe33449bbe55ea"

 [[package]]
 name = "heed"
-version = "0.22.1-nested-rtxns"
+version = "0.22.1-nested-rtxns-6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "0ff115ba5712b1f1fc7617b195f5c2f139e29c397ff79da040cd19db75ccc240"
+checksum = "c69e07cd539834bedcfa938f3d7d8520cce1ad2b0776c122b5ccdf8fd5bafe12"
 dependencies = [
 "bitflags 2.9.4",
 "byteorder",
@@ -3242,6 +3251,7 @@ dependencies = [
 "bumpalo",
 "bumparaw-collections",
 "byte-unit",
+ "bytes",
 "convert_case 0.8.0",
 "crossbeam-channel",
 "csv",
@@ -3259,13 +3269,17 @@ dependencies = [
 "memmap2",
 "page_size",
 "rayon",
+ "reqwest",
 "roaring 0.10.12",
+ "rusty-s3",
 "serde",
 "serde_json",
 "synchronoise",
+ "tar",
 "tempfile",
 "thiserror 2.0.16",
 "time",
+ "tokio",
 "tracing",
 "ureq",
 "uuid",
@@ -3443,26 +3457,49 @@ checksum = "4a5f13b858c8d314ee3e8f639011f7ccefe71f97f96e50151fb991f267928e2c"

 [[package]]
 name = "jieba-macros"
-version = "0.7.1"
+version = "0.8.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "7c676b32a471d3cfae8dac2ad2f8334cd52e53377733cca8c1fb0a5062fec192"
+checksum = "348294e44ee7e3c42685da656490f8febc7359632544019621588902216da95c"
 dependencies = [
 "phf_codegen",
 ]

 [[package]]
 name = "jieba-rs"
-version = "0.7.4"
+version = "0.8.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f5dd552bbb95d578520ee68403bf8aaf0dbbb2ce55b0854d019f9350ad61040a"
+checksum = "766bd7012aa5ba49411ebdf4e93bddd59b182d2918e085d58dec5bb9b54b7105"
 dependencies = [
 "cedarwood",
- "fxhash",
 "include-flate",
 "jieba-macros",
- "lazy_static",
- "phf",
+ "phf 0.13.1",
 "regex",
+ "rustc-hash 2.1.1",
+]
+
+[[package]]
+name = "jiff"
+version = "0.2.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "be1f93b8b1eb69c77f24bbb0afdf66f54b632ee39af40ca21c4365a1d7347e49"
+dependencies = [
+ "jiff-static",
+ "log",
+ "portable-atomic",
+ "portable-atomic-util",
+ "serde",
+]
+
+[[package]]
+name = "jiff-static"
+version = "0.2.15"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "03343451ff899767262ec32146f6d559dd759fdadf42ff0e227c7c48f72594b4"
+dependencies = [
+ "proc-macro2",
+ "quote",
+ "syn 2.0.106",
 ]

 [[package]]
@@ -3888,9 +3925,9 @@ checksum = "241eaef5fd12c88705a01fc1066c48c4b36e0dd4377dcdc7ec3942cea7a69956"

 [[package]]
 name = "lmdb-master-sys"
-version = "0.2.6-nested-rtxns"
+version = "0.2.6-nested-rtxns-6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f4ff85130e3c994b36877045fbbb138d521dea7197bfc19dc3d5d95101a8e20a"
+checksum = "e113d9bf240f974fbe7fd516cbfd8c422e925c0655495501c7237548425493d0"
 dependencies = [
 "cc",
 "doxygen-rs",
@@ -3988,6 +4025,16 @@ version = "1.0.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3e2e65a1a2e43cfcb47a895c4c8b10d1f4a61097f9f254f183aee60cad9c651d"

+[[package]]
+name = "md-5"
+version = "0.10.6"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d89e7ee0cfbedfc4da3340218492196241d89eefb6dab27de5df917a6d2e78cf"
+dependencies = [
+ "cfg-if",
+ "digest",
+]
+
 [[package]]
 name = "md5"
 version = "0.7.0"
@@ -4200,6 +4247,7 @@ dependencies = [
 "big_s",
 "bimap",
 "bincode 1.3.3",
+ "bitpacking",
 "bstr",
 "bumpalo",
 "bumparaw-collections",
@@ -4246,6 +4294,7 @@ dependencies = [
 "obkv",
 "once_cell",
 "ordered-float 5.0.0",
+ "quickcheck",
 "rand 0.8.5",
 "rayon",
 "rhai",
@@ -4830,17 +4879,27 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "1fd6780a80ae0c52cc120a26a1a42c1ae51b247a253e4e06113d23d2c2edd078"
 dependencies = [
 "phf_macros",
- "phf_shared",
+ "phf_shared 0.11.3",
+]
+
+[[package]]
+name = "phf"
+version = "0.13.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c1562dc717473dbaa4c1f85a36410e03c047b2e7df7f45ee938fbef64ae7fadf"
+dependencies = [
+ "phf_shared 0.13.1",
+ "serde",
 ]

 [[package]]
 name = "phf_codegen"
-version = "0.11.3"
+version = "0.13.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "aef8048c789fa5e851558d709946d6d79a8ff88c0440c587967f8e94bfb1216a"
+checksum = "49aa7f9d80421bca176ca8dbfebe668cc7a2684708594ec9f3c0db0805d5d6e1"
 dependencies = [
- "phf_generator",
- "phf_shared",
+ "phf_generator 0.13.1",
+ "phf_shared 0.13.1",
 ]

 [[package]]
@@ -4849,18 +4908,28 @@ version = "0.11.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3c80231409c20246a13fddb31776fb942c38553c51e871f8cbd687a4cfb5843d"
 dependencies = [
- "phf_shared",
+ "phf_shared 0.11.3",
 "rand 0.8.5",
 ]

+[[package]]
+name = "phf_generator"
+version = "0.13.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "135ace3a761e564ec88c03a77317a7c6b80bb7f7135ef2544dbe054243b89737"
+dependencies = [
+ "fastrand",
+ "phf_shared 0.13.1",
+]
+
 [[package]]
 name = "phf_macros"
 version = "0.11.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f84ac04429c13a7ff43785d75ad27569f2951ce0ffd30a3321230db2fc727216"
 dependencies = [
- "phf_generator",
- "phf_shared",
+ "phf_generator 0.11.3",
+ "phf_shared 0.11.3",
 "proc-macro2",
 "quote",
 "syn 2.0.106",
@@ -4875,6 +4944,15 @@ dependencies = [
 "siphasher",
 ]

+[[package]]
+name = "phf_shared"
+version = "0.13.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e57fef6bc5981e38c2ce2d63bfa546861309f875b8a75f092d1d54ae2d64f266"
+dependencies = [
+ "siphasher",
+]
+
 [[package]]
 name = "pin-project"
 version = "1.1.10"
@@ -4962,6 +5040,15 @@ version = "1.11.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f84267b20a16ea918e43c6a88433c2d54fa145c92a811b5b047ccbe153674483"

+[[package]]
+name = "portable-atomic-util"
+version = "0.2.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "d8a2f0d8d040d7848a709caf78912debcc3f33ee4b3cac47d73d1e1069e83507"
+dependencies = [
+ "portable-atomic",
+]
+
 [[package]]
 name = "potential_utf"
 version = "0.1.3"
@@ -5139,6 +5226,27 @@ dependencies = [
 "version_check",
 ]

+[[package]]
+name = "quick-xml"
+version = "0.38.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "42a232e7487fc2ef313d96dde7948e7a3c05101870d8985e4fd8d26aedd27b89"
+dependencies = [
+ "memchr",
+ "serde",
+]
+
+[[package]]
+name = "quickcheck"
+version = "1.0.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "588f6378e4dd99458b60ec275b4477add41ce4fa9f64dcba6f15adccb19b50d6"
+dependencies = [
+ "env_logger",
+ "log",
+ "rand 0.8.5",
+]
+
 [[package]]
 name = "quinn"
 version = "0.11.9"
@@ -5414,6 +5522,7 @@ dependencies = [
 "futures-channel",
 "futures-core",
 "futures-util",
+ "h2 0.4.12",
 "http 1.3.1",
 "http-body",
 "http-body-util",
@@ -5704,6 +5813,25 @@ version = "1.0.22"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b39cdef0fa800fc44525c84ccb54a029961a8215f9619753635a9c0d2538d46d"

+[[package]]
+name = "rusty-s3"
+version = "0.8.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "fac2edd2f0b56bd79a7343f49afc01c2d41010df480538a510e0abc56044f66c"
+dependencies = [
+ "base64 0.22.1",
+ "hmac",
+ "jiff",
+ "md-5",
+ "percent-encoding",
+ "quick-xml",
+ "serde",
+ "serde_json",
+ "sha2",
+ "url",
+ "zeroize",
+]
+
 [[package]]
 name = "ryu"
 version = "1.0.20"
--- a/README.md
+++ b/README.md
@@ -39,6 +39,7 @@
 ## 🖥 Examples

 - [**Movies**](https://where2watch.meilisearch.com/?utm_campaign=oss&utm_source=github&utm_medium=organization) — An application to help you find streaming platforms to watch movies using [hybrid search](https://www.meilisearch.com/solutions/hybrid-search?utm_campaign=oss&utm_source=github&utm_medium=meilisearch&utm_content=demos).
+- [**Flickr**](https://flickr.meilisearch.com/?utm_campaign=oss&utm_source=github&utm_medium=organization) — Search and explore one hundred million Flickr images with semantic search.
 - [**Ecommerce**](https://ecommerce.meilisearch.com/?utm_campaign=oss&utm_source=github&utm_medium=meilisearch&utm_content=demos) — Ecommerce website using disjunctive [facets](https://www.meilisearch.com/docs/learn/fine_tuning_results/faceted_search?utm_campaign=oss&utm_source=github&utm_medium=meilisearch&utm_content=demos), range and rating filtering, and pagination.
 - [**Songs**](https://music.meilisearch.com/?utm_campaign=oss&utm_source=github&utm_medium=meilisearch&utm_content=demos) — Search through 47 million of songs.
 - [**SaaS**](https://saas.meilisearch.com/?utm_campaign=oss&utm_source=github&utm_medium=meilisearch&utm_content=demos) — Search for contacts, deals, and companies in this [multi-tenant](https://www.meilisearch.com/docs/learn/security/multitenancy_tenant_tokens?utm_campaign=oss&utm_source=github&utm_medium=meilisearch&utm_content=demos) CRM application.
--- a/crates/file-store/src/lib.rs
+++ b/crates/file-store/src/lib.rs
@@ -60,7 +60,7 @@ impl FileStore {

    /// Returns the file corresponding to the requested uuid.
    pub fn get_update(&self, uuid: Uuid) -> Result<StdFile> {
-        let path = self.get_update_path(uuid);
+        let path = self.update_path(uuid);
        let file = match StdFile::open(path) {
            Ok(file) => file,
            Err(e) => {
@@ -72,7 +72,7 @@ impl FileStore {
    }

    /// Returns the path that correspond to this uuid, the path could not exists.
-    pub fn get_update_path(&self, uuid: Uuid) -> PathBuf {
+    pub fn update_path(&self, uuid: Uuid) -> PathBuf {
        self.path.join(uuid.to_string())
    }

--- a/crates/index-scheduler/Cargo.toml
+++ b/crates/index-scheduler/Cargo.toml
@@ -14,6 +14,7 @@ license.workspace = true
 anyhow = "1.0.98"
 bincode = "1.3.3"
 byte-unit = "5.1.6"
+bytes = "1.10.1"
 bumpalo = "3.18.1"
 bumparaw-collections = "0.1.4"
 convert_case = "0.8.0"
@@ -32,6 +33,7 @@ rayon = "1.10.0"
 roaring = { version = "0.10.12", features = ["serde"] }
 serde = { version = "1.0.219", features = ["derive"] }
 serde_json = { version = "1.0.140", features = ["preserve_order"] }
+tar = "0.4.44"
 synchronoise = "1.0.1"
 tempfile = "3.20.0"
 thiserror = "2.0.12"
@@ -45,6 +47,9 @@ tracing = "0.1.41"
 ureq = "2.12.1"
 uuid = { version = "1.17.0", features = ["serde", "v4"] }
 backoff = "0.4.0"
+reqwest = { version = "0.12.23", features = ["rustls-tls", "http2"], default-features = false }
+rusty-s3 = "0.8.1"
+tokio = { version = "1.47.1", features = ["full"] }

 [dev-dependencies]
 big_s = "1.0.2"
--- a/crates/index-scheduler/src/error.rs
+++ b/crates/index-scheduler/src/error.rs
@@ -5,6 +5,7 @@ use meilisearch_types::error::{Code, ErrorCode};
 use meilisearch_types::milli::index::RollbackOutcome;
 use meilisearch_types::tasks::{Kind, Status};
 use meilisearch_types::{heed, milli};
+use reqwest::StatusCode;
 use thiserror::Error;

 use crate::TaskId;
@@ -127,6 +128,14 @@ pub enum Error {
    #[error("Aborted task")]
    AbortedTask,

+    #[error("S3 error: status: {status}, body: {body}")]
+    S3Error { status: StatusCode, body: String },
+    #[error("S3 HTTP error: {0}")]
+    S3HttpError(reqwest::Error),
+    #[error("S3 XML error: {0}")]
+    S3XmlError(Box<dyn std::error::Error + Send + Sync>),
+    #[error("S3 bucket error: {0}")]
+    S3BucketError(rusty_s3::BucketError),
    #[error(transparent)]
    Dump(#[from] dump::Error),
    #[error(transparent)]
@@ -226,6 +235,10 @@ impl Error {
            | Error::TaskCancelationWithEmptyQuery
            | Error::FromRemoteWhenExporting { .. }
            | Error::AbortedTask
+            | Error::S3Error { .. }
+            | Error::S3HttpError(_)
+            | Error::S3XmlError(_)
+            | Error::S3BucketError(_)
            | Error::Dump(_)
            | Error::Heed(_)
            | Error::Milli { .. }
@@ -293,8 +306,14 @@ impl ErrorCode for Error {
            Error::BatchNotFound(_) => Code::BatchNotFound,
            Error::TaskDeletionWithEmptyQuery => Code::MissingTaskFilters,
            Error::TaskCancelationWithEmptyQuery => Code::MissingTaskFilters,
-            // TODO: not sure of the Code to use
            Error::NoSpaceLeftInTaskQueue => Code::NoSpaceLeftOnDevice,
+            Error::S3Error { status, .. } if status.is_client_error() => {
+                Code::InvalidS3SnapshotRequest
+            }
+            Error::S3Error { .. } => Code::S3SnapshotServerError,
+            Error::S3HttpError(_) => Code::S3SnapshotServerError,
+            Error::S3XmlError(_) => Code::S3SnapshotServerError,
+            Error::S3BucketError(_) => Code::InvalidS3SnapshotParameters,
            Error::Dump(e) => e.error_code(),
            Error::Milli { error, .. } => error.error_code(),
            Error::ProcessBatchPanicked(_) => Code::Internal,
--- a/crates/index-scheduler/src/insta_snapshot.rs
+++ b/crates/index-scheduler/src/insta_snapshot.rs
@@ -36,6 +36,7 @@ pub fn snapshot_index_scheduler(scheduler: &IndexScheduler) -> String {
        run_loop_iteration: _,
        embedders: _,
        chat_settings: _,
+        runtime: _,
    } = scheduler;

    let rtxn = env.read_txn().unwrap();
--- a/crates/index-scheduler/src/lib.rs
+++ b/crates/index-scheduler/src/lib.rs
@@ -216,6 +216,9 @@ pub struct IndexScheduler {
    /// A counter that is incremented before every call to [`tick`](IndexScheduler::tick)
    #[cfg(test)]
    run_loop_iteration: Arc<RwLock<usize>>,
+
+    /// The tokio runtime used for asynchronous tasks.
+    runtime: Option<tokio::runtime::Handle>,
 }

 impl IndexScheduler {
@@ -242,6 +245,7 @@ impl IndexScheduler {
            run_loop_iteration: self.run_loop_iteration.clone(),
            features: self.features.clone(),
            chat_settings: self.chat_settings,
+            runtime: self.runtime.clone(),
        }
    }

@@ -255,13 +259,23 @@ impl IndexScheduler {
    }

    /// Create an index scheduler and start its run loop.
-    #[allow(private_interfaces)] // because test_utils is private
    pub fn new(
        options: IndexSchedulerOptions,
        auth_env: Env<WithoutTls>,
        from_db_version: (u32, u32, u32),
-        #[cfg(test)] test_breakpoint_sdr: crossbeam_channel::Sender<(test_utils::Breakpoint, bool)>,
-        #[cfg(test)] planned_failures: Vec<(usize, test_utils::FailureLocation)>,
+        runtime: Option<tokio::runtime::Handle>,
+    ) -> Result<Self> {
+        let this = Self::new_without_run(options, auth_env, from_db_version, runtime)?;
+
+        this.run();
+        Ok(this)
+    }
+
+    fn new_without_run(
+        options: IndexSchedulerOptions,
+        auth_env: Env<WithoutTls>,
+        from_db_version: (u32, u32, u32),
+        runtime: Option<tokio::runtime::Handle>,
    ) -> Result<Self> {
        std::fs::create_dir_all(&options.tasks_path)?;
        std::fs::create_dir_all(&options.update_file_path)?;
@@ -316,8 +330,7 @@ impl IndexScheduler {

        wtxn.commit()?;

-        // allow unreachable_code to get rids of the warning in the case of a test build.
-        let this = Self {
+        Ok(Self {
            processing_tasks: Arc::new(RwLock::new(ProcessingTasks::new())),
            version,
            queue,
@@ -333,15 +346,32 @@ impl IndexScheduler {
            webhooks: Arc::new(webhooks),
            embedders: Default::default(),

-            #[cfg(test)]
-            test_breakpoint_sdr,
-            #[cfg(test)]
-            planned_failures,
+            #[cfg(test)] // Will be replaced in `new_tests` in test environments
+            test_breakpoint_sdr: crossbeam_channel::bounded(0).0,
+            #[cfg(test)] // Will be replaced in `new_tests` in test environments
+            planned_failures: Default::default(),
            #[cfg(test)]
            run_loop_iteration: Arc::new(RwLock::new(0)),
            features,
            chat_settings,
-        };
+            runtime,
+        })
+    }
+
+    /// Create an index scheduler and start its run loop.
+    #[cfg(test)]
+    fn new_test(
+        options: IndexSchedulerOptions,
+        auth_env: Env<WithoutTls>,
+        from_db_version: (u32, u32, u32),
+        runtime: Option<tokio::runtime::Handle>,
+        test_breakpoint_sdr: crossbeam_channel::Sender<(test_utils::Breakpoint, bool)>,
+        planned_failures: Vec<(usize, test_utils::FailureLocation)>,
+    ) -> Result<Self> {
+        let mut this = Self::new_without_run(options, auth_env, from_db_version, runtime)?;
+
+        this.test_breakpoint_sdr = test_breakpoint_sdr;
+        this.planned_failures = planned_failures;

        this.run();
        Ok(this)
--- a/crates/index-scheduler/src/scheduler/mod.rs
+++ b/crates/index-scheduler/src/scheduler/mod.rs
@@ -25,6 +25,7 @@ use convert_case::{Case, Casing as _};
 use meilisearch_types::error::ResponseError;
 use meilisearch_types::heed::{Env, WithoutTls};
 use meilisearch_types::milli;
+use meilisearch_types::milli::update::S3SnapshotOptions;
 use meilisearch_types::tasks::Status;
 use process_batch::ProcessBatchInfo;
 use rayon::current_num_threads;
@@ -87,11 +88,14 @@ pub struct Scheduler {

    /// Snapshot compaction status.
    pub(crate) experimental_no_snapshot_compaction: bool,
+
+    /// S3 Snapshot options.
+    pub(crate) s3_snapshot_options: Option<S3SnapshotOptions>,
 }

 impl Scheduler {
-    pub(crate) fn private_clone(&self) -> Scheduler {
-        Scheduler {
+    pub(crate) fn private_clone(&self) -> Self {
+        Self {
            must_stop_processing: self.must_stop_processing.clone(),
            wake_up: self.wake_up.clone(),
            autobatching_enabled: self.autobatching_enabled,
@@ -103,23 +107,52 @@ impl Scheduler {
            version_file_path: self.version_file_path.clone(),
            embedding_cache_cap: self.embedding_cache_cap,
            experimental_no_snapshot_compaction: self.experimental_no_snapshot_compaction,
+            s3_snapshot_options: self.s3_snapshot_options.clone(),
        }
    }

    pub fn new(options: &IndexSchedulerOptions, auth_env: Env<WithoutTls>) -> Scheduler {
+        let IndexSchedulerOptions {
+            version_file_path,
+            auth_path: _,
+            tasks_path: _,
+            update_file_path: _,
+            indexes_path: _,
+            snapshots_path,
+            dumps_path,
+            cli_webhook_url: _,
+            cli_webhook_authorization: _,
+            task_db_size: _,
+            index_base_map_size: _,
+            enable_mdb_writemap: _,
+            index_growth_amount: _,
+            index_count: _,
+            indexer_config,
+            autobatching_enabled,
+            cleanup_enabled: _,
+            max_number_of_tasks: _,
+            max_number_of_batched_tasks,
+            batched_tasks_size_limit,
+            instance_features: _,
+            auto_upgrade: _,
+            embedding_cache_cap,
+            experimental_no_snapshot_compaction,
+        } = options;
+
        Scheduler {
            must_stop_processing: MustStopProcessing::default(),
            // we want to start the loop right away in case meilisearch was ctrl+Ced while processing things
            wake_up: Arc::new(SignalEvent::auto(true)),
-            autobatching_enabled: options.autobatching_enabled,
-            max_number_of_batched_tasks: options.max_number_of_batched_tasks,
-            batched_tasks_size_limit: options.batched_tasks_size_limit,
-            dumps_path: options.dumps_path.clone(),
-            snapshots_path: options.snapshots_path.clone(),
+            autobatching_enabled: *autobatching_enabled,
+            max_number_of_batched_tasks: *max_number_of_batched_tasks,
+            batched_tasks_size_limit: *batched_tasks_size_limit,
+            dumps_path: dumps_path.clone(),
+            snapshots_path: snapshots_path.clone(),
            auth_env,
-            version_file_path: options.version_file_path.clone(),
-            embedding_cache_cap: options.embedding_cache_cap,
-            experimental_no_snapshot_compaction: options.experimental_no_snapshot_compaction,
+            version_file_path: version_file_path.clone(),
+            embedding_cache_cap: *embedding_cache_cap,
+            experimental_no_snapshot_compaction: *experimental_no_snapshot_compaction,
+            s3_snapshot_options: indexer_config.s3_snapshot_options.clone(),
        }
    }
 }
--- a/crates/index-scheduler/src/scheduler/process_snapshot_creation.rs
+++ b/crates/index-scheduler/src/scheduler/process_snapshot_creation.rs
@@ -12,6 +12,8 @@ use crate::processing::{AtomicUpdateFileStep, SnapshotCreationProgress};
 use crate::queue::TaskQueue;
 use crate::{Error, IndexScheduler, Result};

+const UPDATE_FILES_DIR_NAME: &str = "update_files";
+
 /// # Safety
 ///
 /// See [`EnvOpenOptions::open`].
@@ -78,10 +80,32 @@ impl IndexScheduler {
    pub(super) fn process_snapshot(
        &self,
        progress: Progress,
-        mut tasks: Vec<Task>,
+        tasks: Vec<Task>,
    ) -> Result<Vec<Task>> {
        progress.update_progress(SnapshotCreationProgress::StartTheSnapshotCreation);

+        match self.scheduler.s3_snapshot_options.clone() {
+            Some(options) => {
+                #[cfg(not(unix))]
+                {
+                    let _ = options;
+                    panic!("Non-unix platform does not support S3 snapshotting");
+                }
+                #[cfg(unix)]
+                self.runtime
+                    .as_ref()
+                    .expect("Runtime not initialized")
+                    .block_on(self.process_snapshot_to_s3(progress, options, tasks))
+            }
+            None => self.process_snapshots_to_disk(progress, tasks),
+        }
+    }
+
+    fn process_snapshots_to_disk(
+        &self,
+        progress: Progress,
+        mut tasks: Vec<Task>,
+    ) -> Result<Vec<Task>, Error> {
        fs::create_dir_all(&self.scheduler.snapshots_path)?;
        let temp_snapshot_dir = tempfile::tempdir()?;

@@ -128,7 +152,7 @@ impl IndexScheduler {
        let rtxn = self.env.read_txn()?;

        // 2.4 Create the update files directory
-        let update_files_dir = temp_snapshot_dir.path().join("update_files");
+        let update_files_dir = temp_snapshot_dir.path().join(UPDATE_FILES_DIR_NAME);
        fs::create_dir_all(&update_files_dir)?;

        // 2.5 Only copy the update files of the enqueued tasks
@@ -140,7 +164,7 @@ impl IndexScheduler {
            let task =
                self.queue.tasks.get_task(&rtxn, task_id)?.ok_or(Error::CorruptedTaskQueue)?;
            if let Some(content_uuid) = task.content_uuid() {
-                let src = self.queue.file_store.get_update_path(content_uuid);
+                let src = self.queue.file_store.update_path(content_uuid);
                let dst = update_files_dir.join(content_uuid.to_string());
                fs::copy(src, dst)?;
            }
@@ -206,4 +230,403 @@ impl IndexScheduler {

        Ok(tasks)
    }
+
+    #[cfg(unix)]
+    pub(super) async fn process_snapshot_to_s3(
+        &self,
+        progress: Progress,
+        opts: meilisearch_types::milli::update::S3SnapshotOptions,
+        mut tasks: Vec<Task>,
+    ) -> Result<Vec<Task>> {
+        use meilisearch_types::milli::update::S3SnapshotOptions;
+
+        let S3SnapshotOptions {
+            s3_bucket_url,
+            s3_bucket_region,
+            s3_bucket_name,
+            s3_snapshot_prefix,
+            s3_access_key,
+            s3_secret_key,
+            s3_max_in_flight_parts,
+            s3_compression_level: level,
+            s3_signature_duration,
+            s3_multipart_part_size,
+        } = opts;
+
+        let must_stop_processing = self.scheduler.must_stop_processing.clone();
+        let retry_backoff = backoff::ExponentialBackoff::default();
+        let db_name = {
+            let mut base_path = self.env.path().to_owned();
+            base_path.pop();
+            base_path.file_name().and_then(OsStr::to_str).unwrap_or("data.ms").to_string()
+        };
+
+        let (reader, writer) = std::io::pipe()?;
+        let uploader_task = tokio::spawn(multipart_stream_to_s3(
+            s3_bucket_url,
+            s3_bucket_region,
+            s3_bucket_name,
+            s3_snapshot_prefix,
+            s3_access_key,
+            s3_secret_key,
+            s3_max_in_flight_parts,
+            s3_signature_duration,
+            s3_multipart_part_size,
+            must_stop_processing,
+            retry_backoff,
+            db_name,
+            reader,
+        ));
+
+        let index_scheduler = IndexScheduler::private_clone(self);
+        let builder_task = tokio::task::spawn_blocking(move || {
+            stream_tarball_into_pipe(progress, level, writer, index_scheduler)
+        });
+
+        let (uploader_result, builder_result) = tokio::join!(uploader_task, builder_task);
+
+        // Check uploader result first to early return on task abortion.
+        // safety: JoinHandle can return an error if the task was aborted, cancelled, or panicked.
+        uploader_result.unwrap()?;
+        builder_result.unwrap()?;
+
+        for task in &mut tasks {
+            task.status = Status::Succeeded;
+        }
+
+        Ok(tasks)
+    }
+}
+
+/// Streams a tarball of the database content into a pipe.
+#[cfg(unix)]
+fn stream_tarball_into_pipe(
+    progress: Progress,
+    level: u32,
+    writer: std::io::PipeWriter,
+    index_scheduler: IndexScheduler,
+) -> std::result::Result<(), Error> {
+    use std::io::Write as _;
+    use std::path::Path;
+
+    let writer = flate2::write::GzEncoder::new(writer, flate2::Compression::new(level));
+    let mut tarball = tar::Builder::new(writer);
+
+    // 1. Snapshot the version file
+    tarball
+        .append_path_with_name(&index_scheduler.scheduler.version_file_path, VERSION_FILE_NAME)?;
+
+    // 2. Snapshot the index scheduler LMDB env
+    progress.update_progress(SnapshotCreationProgress::SnapshotTheIndexScheduler);
+    let tasks_env_file = index_scheduler.env.try_clone_inner_file()?;
+    let path = Path::new("tasks").join("data.mdb");
+    append_file_to_tarball(&mut tarball, path, tasks_env_file)?;
+
+    // 2.3 Create a read transaction on the index-scheduler
+    let rtxn = index_scheduler.env.read_txn()?;
+
+    // 2.4 Create the update files directory
+    //     And only copy the update files of the enqueued tasks
+    progress.update_progress(SnapshotCreationProgress::SnapshotTheUpdateFiles);
+    let enqueued = index_scheduler.queue.tasks.get_status(&rtxn, Status::Enqueued)?;
+    let (atomic, update_file_progress) = AtomicUpdateFileStep::new(enqueued.len() as u32);
+    progress.update_progress(update_file_progress);
+
+    // We create the update_files directory so that it
+    // always exists even if there are no update files
+    let update_files_dir = Path::new(UPDATE_FILES_DIR_NAME);
+    let src_update_files_dir = {
+        let mut path = index_scheduler.env.path().to_path_buf();
+        path.pop();
+        path.join(UPDATE_FILES_DIR_NAME)
+    };
+    tarball.append_dir(update_files_dir, src_update_files_dir)?;
+
+    for task_id in enqueued {
+        let task = index_scheduler
+            .queue
+            .tasks
+            .get_task(&rtxn, task_id)?
+            .ok_or(Error::CorruptedTaskQueue)?;
+        if let Some(content_uuid) = task.content_uuid() {
+            use std::fs::File;
+
+            let src = index_scheduler.queue.file_store.update_path(content_uuid);
+            let mut update_file = File::open(src)?;
+            let path = update_files_dir.join(content_uuid.to_string());
+            tarball.append_file(path, &mut update_file)?;
+        }
+        atomic.fetch_add(1, Ordering::Relaxed);
+    }
+
+    // 3. Snapshot every indexes
+    progress.update_progress(SnapshotCreationProgress::SnapshotTheIndexes);
+    let index_mapping = index_scheduler.index_mapper.index_mapping;
+    let nb_indexes = index_mapping.len(&rtxn)? as u32;
+    let indexes_dir = Path::new("indexes");
+    let indexes_references: Vec<_> = index_scheduler
+        .index_mapper
+        .index_mapping
+        .iter(&rtxn)?
+        .map(|res| res.map_err(Error::from).map(|(name, uuid)| (name.to_string(), uuid)))
+        .collect::<Result<_, Error>>()?;
+
+    // It's prettier to use a for loop instead of the IndexMapper::try_for_each_index
+    // method, especially when we need to access the UUID, local path and index number.
+    for (i, (name, uuid)) in indexes_references.into_iter().enumerate() {
+        progress.update_progress(VariableNameStep::<SnapshotCreationProgress>::new(
+            &name, i as u32, nb_indexes,
+        ));
+        let path = indexes_dir.join(uuid.to_string()).join("data.mdb");
+        let index = index_scheduler.index_mapper.index(&rtxn, &name)?;
+        let index_file = index.try_clone_inner_file()?;
+        tracing::trace!("Appending index file for {name} in {}", path.display());
+        append_file_to_tarball(&mut tarball, path, index_file)?;
+    }
+
+    drop(rtxn);
+
+    // 4. Snapshot the auth LMDB env
+    progress.update_progress(SnapshotCreationProgress::SnapshotTheApiKeys);
+    let auth_env_file = index_scheduler.scheduler.auth_env.try_clone_inner_file()?;
+    let path = Path::new("auth").join("data.mdb");
+    append_file_to_tarball(&mut tarball, path, auth_env_file)?;
+
+    let mut gzencoder = tarball.into_inner()?;
+    gzencoder.flush()?;
+    gzencoder.try_finish()?;
+    let mut writer = gzencoder.finish()?;
+    writer.flush()?;
+
+    Result::<_, Error>::Ok(())
+}
+
+#[cfg(unix)]
+fn append_file_to_tarball<W, P>(
+    tarball: &mut tar::Builder<W>,
+    path: P,
+    mut auth_env_file: fs::File,
+) -> Result<(), Error>
+where
+    W: std::io::Write,
+    P: AsRef<std::path::Path>,
+{
+    use std::io::{Seek as _, SeekFrom};
+
+    // Note: A previous snapshot operation may have left the cursor
+    //       at the end of the file so we need to seek to the start.
+    auth_env_file.seek(SeekFrom::Start(0))?;
+    tarball.append_file(path, &mut auth_env_file)?;
+    Ok(())
+}
+
+/// Streams the content read from the given reader to S3.
+#[cfg(unix)]
+#[allow(clippy::too_many_arguments)]
+async fn multipart_stream_to_s3(
+    s3_bucket_url: String,
+    s3_bucket_region: String,
+    s3_bucket_name: String,
+    s3_snapshot_prefix: String,
+    s3_access_key: String,
+    s3_secret_key: String,
+    s3_max_in_flight_parts: std::num::NonZero<usize>,
+    s3_signature_duration: std::time::Duration,
+    s3_multipart_part_size: u64,
+    must_stop_processing: super::MustStopProcessing,
+    retry_backoff: backoff::exponential::ExponentialBackoff<backoff::SystemClock>,
+    db_name: String,
+    reader: std::io::PipeReader,
+) -> Result<(), Error> {
+    use std::{collections::VecDeque, os::fd::OwnedFd, path::PathBuf};
+
+    use bytes::{Bytes, BytesMut};
+    use reqwest::{Client, Response};
+    use rusty_s3::S3Action as _;
+    use rusty_s3::{actions::CreateMultipartUpload, Bucket, BucketError, Credentials, UrlStyle};
+    use tokio::task::JoinHandle;
+
+    let reader = OwnedFd::from(reader);
+    let reader = tokio::net::unix::pipe::Receiver::from_owned_fd(reader)?;
+    let s3_snapshot_prefix = PathBuf::from(s3_snapshot_prefix);
+    let url =
+        s3_bucket_url.parse().map_err(BucketError::ParseError).map_err(Error::S3BucketError)?;
+    let bucket = Bucket::new(url, UrlStyle::Path, s3_bucket_name, s3_bucket_region)
+        .map_err(Error::S3BucketError)?;
+    let credential = Credentials::new(s3_access_key, s3_secret_key);
+
+    // Note for the future (rust 1.91+): use with_added_extension, it's prettier
+    let object_path = s3_snapshot_prefix.join(format!("{db_name}.snapshot"));
+    // Note: It doesn't work on Windows and if a port to this platform is needed,
+    //       use the slash-path crate or similar to get the correct path separator.
+    let object = object_path.display().to_string();
+
+    let action = bucket.create_multipart_upload(Some(&credential), &object);
+    let url = action.sign(s3_signature_duration);
+
+    let client = Client::new();
+    let resp = client.post(url).send().await.map_err(Error::S3HttpError)?;
+    let status = resp.status();
+
+    let body = match resp.error_for_status_ref() {
+        Ok(_) => resp.text().await.map_err(Error::S3HttpError)?,
+        Err(_) => {
+            return Err(Error::S3Error { status, body: resp.text().await.unwrap_or_default() })
+        }
+    };
+
+    let multipart =
+        CreateMultipartUpload::parse_response(&body).map_err(|e| Error::S3XmlError(Box::new(e)))?;
+    tracing::debug!("Starting the upload of the snapshot to {object}");
+
+    // We use this bumpalo for etags strings.
+    let bump = bumpalo::Bump::new();
+    let mut etags = Vec::<&str>::new();
+    let mut in_flight = VecDeque::<(JoinHandle<reqwest::Result<Response>>, Bytes)>::with_capacity(
+        s3_max_in_flight_parts.get(),
+    );
+
+    // Part numbers start at 1 and cannot be larger than 10k
+    for part_number in 1u16.. {
+        if must_stop_processing.get() {
+            return Err(Error::AbortedTask);
+        }
+
+        let part_upload =
+            bucket.upload_part(Some(&credential), &object, part_number, multipart.upload_id());
+        let url = part_upload.sign(s3_signature_duration);
+
+        // Wait for a buffer to be ready if there are in-flight parts that landed
+        let mut buffer = if in_flight.len() >= s3_max_in_flight_parts.get() {
+            let (handle, buffer) = in_flight.pop_front().expect("At least one in flight request");
+            let resp = join_and_map_error(handle).await?;
+            extract_and_append_etag(&bump, &mut etags, resp.headers())?;
+
+            let mut buffer = match buffer.try_into_mut() {
+                Ok(buffer) => buffer,
+                Err(_) => unreachable!("All bytes references were consumed in the task"),
+            };
+            buffer.clear();
+            buffer
+        } else {
+            BytesMut::with_capacity(s3_multipart_part_size as usize)
+        };
+
+        // If we successfully read enough bytes,
+        // we can continue and send the buffer/part
+        while buffer.len() < (s3_multipart_part_size as usize / 2) {
+            // Wait for the pipe to be readable
+
+            use std::io;
+            reader.readable().await?;
+
+            match reader.try_read_buf(&mut buffer) {
+                Ok(0) => break,
+                // We read some bytes but maybe not enough
+                Ok(_) => continue,
+                // The readiness event is a false positive.
+                Err(ref e) if e.kind() == io::ErrorKind::WouldBlock => continue,
+                Err(e) => return Err(e.into()),
+            }
+        }
+
+        if buffer.is_empty() {
+            // Break the loop if the buffer is
+            // empty after we tried to read bytes
+            break;
+        }
+
+        let body = buffer.freeze();
+        tracing::trace!("Sending part {part_number}");
+        let task = tokio::spawn({
+            let client = client.clone();
+            let body = body.clone();
+            backoff::future::retry(retry_backoff.clone(), move || {
+                let client = client.clone();
+                let url = url.clone();
+                let body = body.clone();
+                async move {
+                    match client.put(url).body(body).send().await {
+                        Ok(resp) if resp.status().is_client_error() => {
+                            resp.error_for_status().map_err(backoff::Error::Permanent)
+                        }
+                        Ok(resp) => Ok(resp),
+                        Err(e) => Err(backoff::Error::transient(e)),
+                    }
+                }
+            })
+        });
+        in_flight.push_back((task, body));
+    }
+
+    for (handle, _buffer) in in_flight {
+        let resp = join_and_map_error(handle).await?;
+        extract_and_append_etag(&bump, &mut etags, resp.headers())?;
+    }
+
+    tracing::debug!("Finalizing the multipart upload");
+
+    let action = bucket.complete_multipart_upload(
+        Some(&credential),
+        &object,
+        multipart.upload_id(),
+        etags.iter().map(AsRef::as_ref),
+    );
+    let url = action.sign(s3_signature_duration);
+    let body = action.body();
+    let resp = backoff::future::retry(retry_backoff, move || {
+        let client = client.clone();
+        let url = url.clone();
+        let body = body.clone();
+        async move {
+            match client.post(url).body(body).send().await {
+                Ok(resp) if resp.status().is_client_error() => {
+                    resp.error_for_status().map_err(backoff::Error::Permanent)
+                }
+                Ok(resp) => Ok(resp),
+                Err(e) => Err(backoff::Error::transient(e)),
+            }
+        }
+    })
+    .await
+    .map_err(Error::S3HttpError)?;
+
+    let status = resp.status();
+    let body = resp.text().await.map_err(|e| Error::S3Error { status, body: e.to_string() })?;
+    if status.is_success() {
+        Ok(())
+    } else {
+        Err(Error::S3Error { status, body })
+    }
+}
+
+#[cfg(unix)]
+async fn join_and_map_error(
+    join_handle: tokio::task::JoinHandle<Result<reqwest::Response, reqwest::Error>>,
+) -> Result<reqwest::Response> {
+    // safety: Panic happens if the task (JoinHandle) was aborted, cancelled, or panicked
+    let request = join_handle.await.unwrap();
+    let resp = request.map_err(Error::S3HttpError)?;
+    match resp.error_for_status_ref() {
+        Ok(_) => Ok(resp),
+        Err(_) => Err(Error::S3Error {
+            status: resp.status(),
+            body: resp.text().await.unwrap_or_default(),
+        }),
+    }
+}
+
+#[cfg(unix)]
+fn extract_and_append_etag<'b>(
+    bump: &'b bumpalo::Bump,
+    etags: &mut Vec<&'b str>,
+    headers: &reqwest::header::HeaderMap,
+) -> Result<()> {
+    use reqwest::header::ETAG;
+
+    let etag = headers.get(ETAG).ok_or_else(|| Error::S3XmlError("Missing ETag header".into()))?;
+    let etag = etag.to_str().map_err(|e| Error::S3XmlError(Box::new(e)))?;
+    etags.push(bump.alloc_str(etag));
+
+    Ok(())
 }
--- a/crates/index-scheduler/src/test_utils.rs
+++ b/crates/index-scheduler/src/test_utils.rs
@@ -126,7 +126,7 @@ impl IndexScheduler {
        std::fs::create_dir_all(&options.auth_path).unwrap();
        let auth_env = open_auth_store_env(&options.auth_path).unwrap();
        let index_scheduler =
-            Self::new(options, auth_env, version, sender, planned_failures).unwrap();
+            Self::new_test(options, auth_env, version, None, sender, planned_failures).unwrap();

        // To be 100% consistent between all test we're going to start the scheduler right now
        // and ensure it's in the expected starting state.
--- a/crates/meilisearch-types/src/error.rs
+++ b/crates/meilisearch-types/src/error.rs
@@ -258,6 +258,7 @@ InvalidIndexUid                                , InvalidRequest       , BAD_REQU
 InvalidMultiSearchFacets                       , InvalidRequest       , BAD_REQUEST ;
 InvalidMultiSearchFacetsByIndex                , InvalidRequest       , BAD_REQUEST ;
 InvalidMultiSearchFacetOrder                   , InvalidRequest       , BAD_REQUEST ;
+InvalidMultiSearchQueryPersonalization         , InvalidRequest       , BAD_REQUEST ;
 InvalidMultiSearchFederated                    , InvalidRequest       , BAD_REQUEST ;
 InvalidMultiSearchFederationOptions            , InvalidRequest       , BAD_REQUEST ;
 InvalidMultiSearchMaxValuesPerFacet            , InvalidRequest       , BAD_REQUEST ;
@@ -315,6 +316,8 @@ InvalidSearchShowRankingScoreDetails           , InvalidRequest       , BAD_REQU
 InvalidSimilarShowRankingScoreDetails          , InvalidRequest       , BAD_REQUEST ;
 InvalidSearchSort                              , InvalidRequest       , BAD_REQUEST ;
 InvalidSearchDistinct                          , InvalidRequest       , BAD_REQUEST ;
+InvalidSearchPersonalize                       , InvalidRequest       , BAD_REQUEST ;
+InvalidSearchPersonalizeUserContext            , InvalidRequest       , BAD_REQUEST ;
 InvalidSearchMediaAndVector                    , InvalidRequest       , BAD_REQUEST ;
 InvalidSettingsDisplayedAttributes             , InvalidRequest       , BAD_REQUEST ;
 InvalidSettingsDistinctAttribute               , InvalidRequest       , BAD_REQUEST ;
@@ -390,6 +393,9 @@ TooManyVectors                                 , InvalidRequest       , BAD_REQU
 UnretrievableDocument                          , Internal             , BAD_REQUEST ;
 UnretrievableErrorCode                         , InvalidRequest       , BAD_REQUEST ;
 UnsupportedMediaType                           , InvalidRequest       , UNSUPPORTED_MEDIA_TYPE ;
+InvalidS3SnapshotRequest                       , Internal             , BAD_REQUEST ;
+InvalidS3SnapshotParameters                    , Internal             , BAD_REQUEST ;
+S3SnapshotServerError                          , Internal             , BAD_GATEWAY ;

 // Experimental features
 VectorEmbeddingError                           , InvalidRequest       , BAD_REQUEST ;
@@ -679,6 +685,18 @@ impl fmt::Display for deserr_codes::InvalidNetworkSearchApiKey {
    }
 }

+impl fmt::Display for deserr_codes::InvalidSearchPersonalize {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "the value of `personalize` is invalid, expected a JSON object with `userContext` string.")
+    }
+}
+
+impl fmt::Display for deserr_codes::InvalidSearchPersonalizeUserContext {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        write!(f, "the value of `userContext` is invalid, expected a string.")
+    }
+}
+
 #[macro_export]
 macro_rules! internal_error {
    ($target:ty : $($other:path), *) => {
--- a/crates/meilisearch-types/src/settings.rs
+++ b/crates/meilisearch-types/src/settings.rs
@@ -346,24 +346,26 @@ impl<T> Settings<T> {
                continue;
            };

-            Self::hide_secret(api_key);
+            hide_secret(api_key, 0);
        }
    }
+}

-    fn hide_secret(secret: &mut String) {
-        match secret.len() {
-            x if x < 10 => {
-                secret.replace_range(.., "XXX...");
-            }
-            x if x < 20 => {
-                secret.replace_range(2.., "XXXX...");
-            }
-            x if x < 30 => {
-                secret.replace_range(3.., "XXXXX...");
-            }
-            _x => {
-                secret.replace_range(5.., "XXXXXX...");
-            }
+/// Redact a secret string, starting from the `secret_offset`th byte.
+pub fn hide_secret(secret: &mut String, secret_offset: usize) {
+    match secret.len().checked_sub(secret_offset) {
+        None => (),
+        Some(x) if x < 10 => {
+            secret.replace_range(secret_offset.., "XXX...");
+        }
+        Some(x) if x < 20 => {
+            secret.replace_range((secret_offset + 2).., "XXXX...");
+        }
+        Some(x) if x < 30 => {
+            secret.replace_range((secret_offset + 3).., "XXXXX...");
+        }
+        Some(_x) => {
+            secret.replace_range((secret_offset + 5).., "XXXXXX...");
        }
    }
 }
--- a/crates/meilisearch-types/src/webhooks.rs
+++ b/crates/meilisearch-types/src/webhooks.rs
@@ -11,6 +11,24 @@ pub struct Webhook {
    pub headers: BTreeMap<String, String>,
 }

+impl Webhook {
+    pub fn redact_authorization_header(&mut self) {
+        // headers are case insensitive, so to make the redaction robust we iterate over qualifying headers
+        // rather than getting one canonical `Authorization` header.
+        for value in self
+            .headers
+            .iter_mut()
+            .filter_map(|(name, value)| name.eq_ignore_ascii_case("authorization").then_some(value))
+        {
+            if value.starts_with("Bearer ") {
+                crate::settings::hide_secret(value, "Bearer ".len());
+            } else {
+                crate::settings::hide_secret(value, 0);
+            }
+        }
+    }
+}
+
 #[derive(Debug, Serialize, Default, Clone, PartialEq)]
 #[serde(rename_all = "camelCase")]
 pub struct WebhooksView {
--- a/crates/meilisearch/src/analytics/segment_analytics.rs
+++ b/crates/meilisearch/src/analytics/segment_analytics.rs
@@ -208,6 +208,7 @@ struct Infos {
    experimental_no_edition_2024_for_prefix_post_processing: bool,
    experimental_no_edition_2024_for_facet_post_processing: bool,
    experimental_vector_store_setting: bool,
+    experimental_personalization: bool,
    gpu_enabled: bool,
    db_path: bool,
    import_dump: bool,
@@ -217,6 +218,7 @@ struct Infos {
    import_snapshot: bool,
    schedule_snapshot: Option<u64>,
    snapshot_dir: bool,
+    uses_s3_snapshots: bool,
    ignore_missing_snapshot: bool,
    ignore_snapshot_if_db_exists: bool,
    http_addr: bool,
@@ -285,6 +287,8 @@ impl Infos {
            indexer_options,
            config_file_path,
            no_analytics: _,
+            experimental_personalization_api_key,
+            s3_snapshot_options,
        } = options;

        let schedule_snapshot = match schedule_snapshot {
@@ -348,6 +352,7 @@ impl Infos {
            import_snapshot: import_snapshot.is_some(),
            schedule_snapshot,
            snapshot_dir: snapshot_dir != PathBuf::from("snapshots/"),
+            uses_s3_snapshots: s3_snapshot_options.is_some(),
            ignore_missing_snapshot,
            ignore_snapshot_if_db_exists,
            http_addr: http_addr != default_http_addr(),
@@ -371,6 +376,7 @@ impl Infos {
            experimental_no_edition_2024_for_settings,
            experimental_no_edition_2024_for_prefix_post_processing,
            experimental_no_edition_2024_for_facet_post_processing,
+            experimental_personalization: experimental_personalization_api_key.is_some(),
        }
    }
 }
--- a/crates/meilisearch/src/error.rs
+++ b/crates/meilisearch/src/error.rs
@@ -38,6 +38,8 @@ pub enum MeilisearchHttpError {
    PaginationInFederatedQuery(usize, &'static str),
    #[error("Inside `.queries[{0}]`: Using facet options is not allowed in federated queries.\n - Hint: remove `facets` from query #{0} or remove `federation` from the request\n - Hint: pass `federation.facetsByIndex.{1}: {2:?}` for facets in federated search")]
    FacetsInFederatedQuery(usize, String, Vec<String>),
+    #[error("Inside `.queries[{0}]`: Using `.personalize` is not allowed in federated queries.\n - Hint: remove `personalize` from query #{0} or remove `federation` from the request")]
+    PersonalizationInFederatedQuery(usize),
    #[error("Inconsistent order for values in facet `{facet}`: index `{previous_uid}` orders {previous_facet_order}, but index `{current_uid}` orders {index_facet_order}.\n - Hint: Remove `federation.mergeFacets` or change `faceting.sortFacetValuesBy` to be consistent in settings.")]
    InconsistentFacetOrder {
        facet: String,
@@ -137,6 +139,9 @@ impl ErrorCode for MeilisearchHttpError {
            MeilisearchHttpError::InconsistentFacetOrder { .. } => {
                Code::InvalidMultiSearchFacetOrder
            }
+            MeilisearchHttpError::PersonalizationInFederatedQuery(_) => {
+                Code::InvalidMultiSearchQueryPersonalization
+            }
            MeilisearchHttpError::InconsistentOriginHeaders { .. } => {
                Code::InconsistentDocumentChangeHeaders
            }
--- a/crates/meilisearch/src/lib.rs
+++ b/crates/meilisearch/src/lib.rs
@@ -11,6 +11,7 @@ pub mod middleware;
 pub mod option;
 #[cfg(test)]
 mod option_test;
+pub mod personalization;
 pub mod routes;
 pub mod search;
 pub mod search_queue;
@@ -58,6 +59,7 @@ use tracing::{error, info_span};
 use tracing_subscriber::filter::Targets;

 use crate::error::MeilisearchHttpError;
+use crate::personalization::PersonalizationService;

 /// Default number of simultaneously opened indexes.
 ///
@@ -128,12 +130,8 @@ pub type LogStderrType = tracing_subscriber::filter::Filtered<
 >;

 pub fn create_app(
-    index_scheduler: Data<IndexScheduler>,
-    auth_controller: Data<AuthController>,
-    search_queue: Data<SearchQueue>,
+    services: ServicesData,
    opt: Opt,
-    logs: (LogRouteHandle, LogStderrHandle),
-    analytics: Data<Analytics>,
    enable_dashboard: bool,
 ) -> actix_web::App<
    impl ServiceFactory<
@@ -145,17 +143,7 @@ pub fn create_app(
    >,
 > {
    let app = actix_web::App::new()
-        .configure(|s| {
-            configure_data(
-                s,
-                index_scheduler.clone(),
-                auth_controller.clone(),
-                search_queue.clone(),
-                &opt,
-                logs,
-                analytics.clone(),
-            )
-        })
+        .configure(|s| configure_data(s, services, &opt))
        .configure(routes::configure)
        .configure(|s| dashboard(s, enable_dashboard));

@@ -216,7 +204,10 @@ enum OnFailure {
    KeepDb,
 }

-pub fn setup_meilisearch(opt: &Opt) -> anyhow::Result<(Arc<IndexScheduler>, Arc<AuthController>)> {
+pub fn setup_meilisearch(
+    opt: &Opt,
+    handle: tokio::runtime::Handle,
+) -> anyhow::Result<(Arc<IndexScheduler>, Arc<AuthController>)> {
    let index_scheduler_opt = IndexSchedulerOptions {
        version_file_path: opt.db_path.join(VERSION_FILE_NAME),
        auth_path: opt.db_path.join("auth"),
@@ -230,7 +221,11 @@ pub fn setup_meilisearch(opt: &Opt) -> anyhow::Result<(Arc<IndexScheduler>, Arc<
        task_db_size: opt.max_task_db_size.as_u64() as usize,
        index_base_map_size: opt.max_index_size.as_u64() as usize,
        enable_mdb_writemap: opt.experimental_reduce_indexing_memory_usage,
-        indexer_config: Arc::new((&opt.indexer_options).try_into()?),
+        indexer_config: Arc::new({
+            let s3_snapshot_options =
+                opt.s3_snapshot_options.clone().map(|opt| opt.try_into()).transpose()?;
+            IndexerConfig { s3_snapshot_options, ..(&opt.indexer_options).try_into()? }
+        }),
        autobatching_enabled: true,
        cleanup_enabled: !opt.experimental_replication_parameters,
        max_number_of_tasks: 1_000_000,
@@ -256,6 +251,7 @@ pub fn setup_meilisearch(opt: &Opt) -> anyhow::Result<(Arc<IndexScheduler>, Arc<
                    index_scheduler_opt,
                    OnFailure::RemoveDb,
                    binary_version, // the db is empty
+                    handle,
                )?,
                Err(e) => {
                    std::fs::remove_dir_all(&opt.db_path)?;
@@ -273,7 +269,7 @@ pub fn setup_meilisearch(opt: &Opt) -> anyhow::Result<(Arc<IndexScheduler>, Arc<
            bail!("snapshot doesn't exist at {}", snapshot_path.display())
        // the snapshot and the db exist, and we can ignore the snapshot because of the ignore_snapshot_if_db_exists flag
        } else {
-            open_or_create_database(opt, index_scheduler_opt, empty_db, binary_version)?
+            open_or_create_database(opt, index_scheduler_opt, empty_db, binary_version, handle)?
        }
    } else if let Some(ref path) = opt.import_dump {
        let src_path_exists = path.exists();
@@ -284,6 +280,7 @@ pub fn setup_meilisearch(opt: &Opt) -> anyhow::Result<(Arc<IndexScheduler>, Arc<
                index_scheduler_opt,
                OnFailure::RemoveDb,
                binary_version, // the db is empty
+                handle,
            )?;
            match import_dump(&opt.db_path, path, &mut index_scheduler, &mut auth_controller) {
                Ok(()) => (index_scheduler, auth_controller),
@@ -304,10 +301,10 @@ pub fn setup_meilisearch(opt: &Opt) -> anyhow::Result<(Arc<IndexScheduler>, Arc<
        // the dump and the db exist and we can ignore the dump because of the ignore_dump_if_db_exists flag
        // or, the dump is missing but we can ignore that because of the ignore_missing_dump flag
        } else {
-            open_or_create_database(opt, index_scheduler_opt, empty_db, binary_version)?
+            open_or_create_database(opt, index_scheduler_opt, empty_db, binary_version, handle)?
        }
    } else {
-        open_or_create_database(opt, index_scheduler_opt, empty_db, binary_version)?
+        open_or_create_database(opt, index_scheduler_opt, empty_db, binary_version, handle)?
    };

    // We create a loop in a thread that registers snapshotCreation tasks
@@ -338,6 +335,7 @@ fn open_or_create_database_unchecked(
    index_scheduler_opt: IndexSchedulerOptions,
    on_failure: OnFailure,
    version: (u32, u32, u32),
+    handle: tokio::runtime::Handle,
 ) -> anyhow::Result<(IndexScheduler, AuthController)> {
    // we don't want to create anything in the data.ms yet, thus we
    // wrap our two builders in a closure that'll be executed later.
@@ -345,7 +343,7 @@ fn open_or_create_database_unchecked(
    let auth_env = open_auth_store_env(&index_scheduler_opt.auth_path).unwrap();
    let auth_controller = AuthController::new(auth_env.clone(), &opt.master_key);
    let index_scheduler_builder = || -> anyhow::Result<_> {
-        Ok(IndexScheduler::new(index_scheduler_opt, auth_env, version)?)
+        Ok(IndexScheduler::new(index_scheduler_opt, auth_env, version, Some(handle))?)
    };

    match (
@@ -452,6 +450,7 @@ fn open_or_create_database(
    index_scheduler_opt: IndexSchedulerOptions,
    empty_db: bool,
    binary_version: (u32, u32, u32),
+    handle: tokio::runtime::Handle,
 ) -> anyhow::Result<(IndexScheduler, AuthController)> {
    let version = if !empty_db {
        check_version(opt, &index_scheduler_opt, binary_version)?
@@ -459,7 +458,7 @@ fn open_or_create_database(
        binary_version
    };

-    open_or_create_database_unchecked(opt, index_scheduler_opt, OnFailure::KeepDb, version)
+    open_or_create_database_unchecked(opt, index_scheduler_opt, OnFailure::KeepDb, version, handle)
 }

 fn import_dump(
@@ -527,7 +526,11 @@ fn import_dump(
    let indexer_config = if base_config.max_threads.is_none() {
        let (thread_pool, _) = default_thread_pool_and_threads();

-        let _config = IndexerConfig { thread_pool, ..*base_config };
+        let _config = IndexerConfig {
+            thread_pool,
+            s3_snapshot_options: base_config.s3_snapshot_options.clone(),
+            ..*base_config
+        };
        backup_config = _config;
        &backup_config
    } else {
@@ -675,23 +678,26 @@ fn import_dump(
    Ok(index_scheduler_dump.finish()?)
 }

-pub fn configure_data(
-    config: &mut web::ServiceConfig,
-    index_scheduler: Data<IndexScheduler>,
-    auth: Data<AuthController>,
-    search_queue: Data<SearchQueue>,
-    opt: &Opt,
-    (logs_route, logs_stderr): (LogRouteHandle, LogStderrHandle),
-    analytics: Data<Analytics>,
-) {
+pub fn configure_data(config: &mut web::ServiceConfig, services: ServicesData, opt: &Opt) {
+    let ServicesData {
+        index_scheduler,
+        auth,
+        search_queue,
+        personalization_service,
+        logs_route_handle,
+        logs_stderr_handle,
+        analytics,
+    } = services;
+
    let http_payload_size_limit = opt.http_payload_size_limit.as_u64() as usize;
    config
        .app_data(index_scheduler)
        .app_data(auth)
        .app_data(search_queue)
        .app_data(analytics)
-        .app_data(web::Data::new(logs_route))
-        .app_data(web::Data::new(logs_stderr))
+        .app_data(personalization_service)
+        .app_data(logs_route_handle)
+        .app_data(logs_stderr_handle)
        .app_data(web::Data::new(opt.clone()))
        .app_data(
            web::JsonConfig::default()
@@ -752,3 +758,14 @@ pub fn dashboard(config: &mut web::ServiceConfig, enable_frontend: bool) {
 pub fn dashboard(config: &mut web::ServiceConfig, _enable_frontend: bool) {
    config.service(web::resource("/").route(web::get().to(routes::running)));
 }
+
+#[derive(Clone)]
+pub struct ServicesData {
+    pub index_scheduler: Data<IndexScheduler>,
+    pub auth: Data<AuthController>,
+    pub search_queue: Data<SearchQueue>,
+    pub personalization_service: Data<PersonalizationService>,
+    pub logs_route_handle: Data<LogRouteHandle>,
+    pub logs_stderr_handle: Data<LogStderrHandle>,
+    pub analytics: Data<Analytics>,
+}
--- a/crates/meilisearch/src/main.rs
+++ b/crates/meilisearch/src/main.rs
@@ -14,10 +14,11 @@ use index_scheduler::IndexScheduler;
 use is_terminal::IsTerminal;
 use meilisearch::analytics::Analytics;
 use meilisearch::option::LogMode;
+use meilisearch::personalization::PersonalizationService;
 use meilisearch::search_queue::SearchQueue;
 use meilisearch::{
    analytics, create_app, setup_meilisearch, LogRouteHandle, LogRouteType, LogStderrHandle,
-    LogStderrType, Opt, SubscriberForSecondLayer,
+    LogStderrType, Opt, ServicesData, SubscriberForSecondLayer,
 };
 use meilisearch_auth::{generate_master_key, AuthController, MASTER_KEY_MIN_SIZE};
 use termcolor::{Color, ColorChoice, ColorSpec, StandardStream, WriteColor};
@@ -76,7 +77,10 @@ fn on_panic(info: &std::panic::PanicHookInfo) {

 #[actix_web::main]
 async fn main() -> anyhow::Result<()> {
-    try_main().await.inspect_err(|error| {
+    // won't panic inside of tokio::main
+    let runtime = tokio::runtime::Handle::current();
+
+    try_main(runtime).await.inspect_err(|error| {
        tracing::error!(%error);
        let mut current = error.source();
        let mut depth = 0;
@@ -88,7 +92,7 @@ async fn main() -> anyhow::Result<()> {
    })
 }

-async fn try_main() -> anyhow::Result<()> {
+async fn try_main(runtime: tokio::runtime::Handle) -> anyhow::Result<()> {
    let (opt, config_read_from) = Opt::try_build()?;

    std::panic::set_hook(Box::new(on_panic));
@@ -122,7 +126,7 @@ async fn try_main() -> anyhow::Result<()> {
        _ => (),
    }

-    let (index_scheduler, auth_controller) = setup_meilisearch(&opt)?;
+    let (index_scheduler, auth_controller) = setup_meilisearch(&opt, runtime)?;

    let analytics =
        analytics::Analytics::new(&opt, index_scheduler.clone(), auth_controller.clone()).await;
@@ -149,8 +153,15 @@ async fn run_http(
    let enable_dashboard = &opt.env == "development";
    let opt_clone = opt.clone();
    let index_scheduler = Data::from(index_scheduler);
-    let auth_controller = Data::from(auth_controller);
+    let auth = Data::from(auth_controller);
    let analytics = Data::from(analytics);
+    // Create personalization service with API key from options
+    let personalization_service = Data::new(
+        opt.experimental_personalization_api_key
+            .clone()
+            .map(PersonalizationService::cohere)
+            .unwrap_or_else(PersonalizationService::disabled),
+    );
    let search_queue = SearchQueue::new(
        opt.experimental_search_queue_size,
        available_parallelism()
@@ -162,21 +173,25 @@ async fn run_http(
        usize::from(opt.experimental_drop_search_after) as u64
    ));
    let search_queue = Data::new(search_queue);
+    let (logs_route_handle, logs_stderr_handle) = logs;
+    let logs_route_handle = Data::new(logs_route_handle);
+    let logs_stderr_handle = Data::new(logs_stderr_handle);

-    let http_server = HttpServer::new(move || {
-        create_app(
-            index_scheduler.clone(),
-            auth_controller.clone(),
-            search_queue.clone(),
-            opt.clone(),
-            logs.clone(),
-            analytics.clone(),
-            enable_dashboard,
-        )
-    })
-    // Disable signals allows the server to terminate immediately when a user enter CTRL-C
-    .disable_signals()
-    .keep_alive(KeepAlive::Os);
+    let services = ServicesData {
+        index_scheduler,
+        auth,
+        search_queue,
+        personalization_service,
+        logs_route_handle,
+        logs_stderr_handle,
+        analytics,
+    };
+
+    let http_server =
+        HttpServer::new(move || create_app(services.clone(), opt.clone(), enable_dashboard))
+            // Disable signals allows the server to terminate immediately when a user enter CTRL-C
+            .disable_signals()
+            .keep_alive(KeepAlive::Os);

    if let Some(config) = opt_clone.get_ssl_config()? {
        http_server.bind_rustls_0_23(opt_clone.http_addr, config)?.run().await?;
--- a/crates/meilisearch/src/metrics.rs
+++ b/crates/meilisearch/src/metrics.rs
@@ -114,4 +114,9 @@ lazy_static! {
            "Meilisearch Task Queue Size Until Stop Registering",
        ))
        .expect("Can't create a metric");
+    pub static ref MEILISEARCH_PERSONALIZED_SEARCH_REQUESTS: IntGauge = register_int_gauge!(opts!(
+        "meilisearch_personalized_search_requests",
+        "Meilisearch number of search requests with personalization"
+    ))
+    .expect("Can't create a metric");
 }
--- a/crates/meilisearch/src/option.rs
+++ b/crates/meilisearch/src/option.rs
@@ -7,12 +7,13 @@ use std::ops::Deref;
 use std::path::PathBuf;
 use std::str::FromStr;
 use std::sync::Arc;
+use std::time::Duration;
 use std::{env, fmt, fs};

 use byte_unit::{Byte, ParseError, UnitType};
 use clap::Parser;
 use meilisearch_types::features::InstanceTogglableFeatures;
-use meilisearch_types::milli::update::IndexerConfig;
+use meilisearch_types::milli::update::{IndexerConfig, S3SnapshotOptions};
 use meilisearch_types::milli::ThreadPoolNoAbortBuilder;
 use rustls::server::{ServerSessionMemoryCache, WebPkiClientVerifier};
 use rustls::RootCertStore;
@@ -74,6 +75,22 @@ const MEILI_EXPERIMENTAL_EMBEDDING_CACHE_ENTRIES: &str =
 const MEILI_EXPERIMENTAL_NO_SNAPSHOT_COMPACTION: &str = "MEILI_EXPERIMENTAL_NO_SNAPSHOT_COMPACTION";
 const MEILI_EXPERIMENTAL_NO_EDITION_2024_FOR_DUMPS: &str =
    "MEILI_EXPERIMENTAL_NO_EDITION_2024_FOR_DUMPS";
+const MEILI_EXPERIMENTAL_PERSONALIZATION_API_KEY: &str =
+    "MEILI_EXPERIMENTAL_PERSONALIZATION_API_KEY";
+
+// Related to S3 snapshots
+const MEILI_S3_BUCKET_URL: &str = "MEILI_S3_BUCKET_URL";
+const MEILI_S3_BUCKET_REGION: &str = "MEILI_S3_BUCKET_REGION";
+const MEILI_S3_BUCKET_NAME: &str = "MEILI_S3_BUCKET_NAME";
+const MEILI_S3_SNAPSHOT_PREFIX: &str = "MEILI_S3_SNAPSHOT_PREFIX";
+const MEILI_S3_ACCESS_KEY: &str = "MEILI_S3_ACCESS_KEY";
+const MEILI_S3_SECRET_KEY: &str = "MEILI_S3_SECRET_KEY";
+const MEILI_EXPERIMENTAL_S3_MAX_IN_FLIGHT_PARTS: &str = "MEILI_EXPERIMENTAL_S3_MAX_IN_FLIGHT_PARTS";
+const MEILI_EXPERIMENTAL_S3_COMPRESSION_LEVEL: &str = "MEILI_EXPERIMENTAL_S3_COMPRESSION_LEVEL";
+const MEILI_EXPERIMENTAL_S3_SIGNATURE_DURATION_SECONDS: &str =
+    "MEILI_EXPERIMENTAL_S3_SIGNATURE_DURATION_SECONDS";
+const MEILI_EXPERIMENTAL_S3_MULTIPART_PART_SIZE: &str = "MEILI_EXPERIMENTAL_S3_MULTIPART_PART_SIZE";
+
 const DEFAULT_CONFIG_FILE_PATH: &str = "./config.toml";
 const DEFAULT_DB_PATH: &str = "./data.ms";
 const DEFAULT_HTTP_ADDR: &str = "localhost:7700";
@@ -83,6 +100,10 @@ const DEFAULT_SNAPSHOT_DIR: &str = "snapshots/";
 const DEFAULT_SNAPSHOT_INTERVAL_SEC: u64 = 86400;
 const DEFAULT_SNAPSHOT_INTERVAL_SEC_STR: &str = "86400";
 const DEFAULT_DUMP_DIR: &str = "dumps/";
+const DEFAULT_S3_SNAPSHOT_MAX_IN_FLIGHT_PARTS: NonZeroUsize = NonZeroUsize::new(10).unwrap();
+const DEFAULT_S3_SNAPSHOT_COMPRESSION_LEVEL: u32 = 0;
+const DEFAULT_S3_SNAPSHOT_SIGNATURE_DURATION_SECONDS: u64 = 8 * 3600; // 8 hours
+const DEFAULT_S3_SNAPSHOT_MULTIPART_PART_SIZE: Byte = Byte::from_u64(375 * 1024 * 1024); // 375 MiB

 const MEILI_MAX_INDEXING_MEMORY: &str = "MEILI_MAX_INDEXING_MEMORY";
 const MEILI_MAX_INDEXING_THREADS: &str = "MEILI_MAX_INDEXING_THREADS";
@@ -475,10 +496,20 @@ pub struct Opt {
    #[serde(default)]
    pub experimental_no_snapshot_compaction: bool,

+    /// Experimental personalization API key feature.
+    ///
+    /// Sets the API key for personalization features.
+    #[clap(long, env = MEILI_EXPERIMENTAL_PERSONALIZATION_API_KEY)]
+    pub experimental_personalization_api_key: Option<String>,
+
    #[serde(flatten)]
    #[clap(flatten)]
    pub indexer_options: IndexerOpts,

+    #[serde(flatten)]
+    #[clap(flatten)]
+    pub s3_snapshot_options: Option<S3SnapshotOpts>,
+
    /// Set the path to a configuration file that should be used to setup the engine.
    /// Format must be TOML.
    #[clap(long)]
@@ -580,6 +611,8 @@ impl Opt {
            experimental_limit_batched_tasks_total_size,
            experimental_embedding_cache_entries,
            experimental_no_snapshot_compaction,
+            experimental_personalization_api_key,
+            s3_snapshot_options,
        } = self;
        export_to_env_if_not_present(MEILI_DB_PATH, db_path);
        export_to_env_if_not_present(MEILI_HTTP_ADDR, http_addr);
@@ -680,7 +713,22 @@ impl Opt {
            MEILI_EXPERIMENTAL_NO_SNAPSHOT_COMPACTION,
            experimental_no_snapshot_compaction.to_string(),
        );
+        if let Some(experimental_personalization_api_key) = experimental_personalization_api_key {
+            export_to_env_if_not_present(
+                MEILI_EXPERIMENTAL_PERSONALIZATION_API_KEY,
+                experimental_personalization_api_key,
+            );
+        }
        indexer_options.export_to_env();
+        if let Some(s3_snapshot_options) = s3_snapshot_options {
+            #[cfg(not(unix))]
+            {
+                let _ = s3_snapshot_options;
+                panic!("S3 snapshot options are not supported on Windows");
+            }
+            #[cfg(unix)]
+            s3_snapshot_options.export_to_env();
+        }
    }

    pub fn get_ssl_config(&self) -> anyhow::Result<Option<rustls::ServerConfig>> {
@@ -849,6 +897,16 @@ impl TryFrom<&IndexerOpts> for IndexerConfig {
    type Error = anyhow::Error;

    fn try_from(other: &IndexerOpts) -> Result<Self, Self::Error> {
+        let IndexerOpts {
+            max_indexing_memory,
+            max_indexing_threads,
+            skip_index_budget,
+            experimental_no_edition_2024_for_settings,
+            experimental_no_edition_2024_for_dumps,
+            experimental_no_edition_2024_for_prefix_post_processing,
+            experimental_no_edition_2024_for_facet_post_processing,
+        } = other;
+
        let thread_pool = ThreadPoolNoAbortBuilder::new_for_indexing()
            .num_threads(other.max_indexing_threads.unwrap_or_else(|| num_cpus::get() / 2))
            .build()?;
@@ -856,21 +914,163 @@ impl TryFrom<&IndexerOpts> for IndexerConfig {
        Ok(Self {
            thread_pool,
            log_every_n: Some(DEFAULT_LOG_EVERY_N),
-            max_memory: other.max_indexing_memory.map(|b| b.as_u64() as usize),
-            max_threads: *other.max_indexing_threads,
+            max_memory: max_indexing_memory.map(|b| b.as_u64() as usize),
+            max_threads: max_indexing_threads.0,
            max_positions_per_attributes: None,
-            skip_index_budget: other.skip_index_budget,
-            experimental_no_edition_2024_for_settings: other
-                .experimental_no_edition_2024_for_settings,
-            experimental_no_edition_2024_for_dumps: other.experimental_no_edition_2024_for_dumps,
+            skip_index_budget: *skip_index_budget,
+            experimental_no_edition_2024_for_settings: *experimental_no_edition_2024_for_settings,
+            experimental_no_edition_2024_for_dumps: *experimental_no_edition_2024_for_dumps,
            chunk_compression_type: Default::default(),
            chunk_compression_level: Default::default(),
            documents_chunk_size: Default::default(),
            max_nb_chunks: Default::default(),
-            experimental_no_edition_2024_for_prefix_post_processing: other
-                .experimental_no_edition_2024_for_prefix_post_processing,
-            experimental_no_edition_2024_for_facet_post_processing: other
-                .experimental_no_edition_2024_for_facet_post_processing,
+            experimental_no_edition_2024_for_prefix_post_processing:
+                *experimental_no_edition_2024_for_prefix_post_processing,
+            experimental_no_edition_2024_for_facet_post_processing:
+                *experimental_no_edition_2024_for_facet_post_processing,
+            s3_snapshot_options: None,
+        })
+    }
+}
+
+#[derive(Debug, Clone, Parser, Deserialize)]
+// This group is a bit tricky but makes it possible to require all listed fields if one of them
+// is specified. It lets us keep an Option for the S3SnapshotOpts configuration.
+// <https://github.com/clap-rs/clap/issues/5092#issuecomment-2616986075>
+#[group(requires_all = ["s3_bucket_url", "s3_bucket_region", "s3_bucket_name", "s3_snapshot_prefix", "s3_access_key", "s3_secret_key"])]
+pub struct S3SnapshotOpts {
+    /// The S3 bucket URL in the format https://s3.<region>.amazonaws.com.
+    #[clap(long, env = MEILI_S3_BUCKET_URL, required = false)]
+    #[serde(default)]
+    pub s3_bucket_url: String,
+
+    /// The region in the format us-east-1.
+    #[clap(long, env = MEILI_S3_BUCKET_REGION, required = false)]
+    #[serde(default)]
+    pub s3_bucket_region: String,
+
+    /// The bucket name.
+    #[clap(long, env = MEILI_S3_BUCKET_NAME, required = false)]
+    #[serde(default)]
+    pub s3_bucket_name: String,
+
+    /// The prefix path where to put the snapshot, uses normal slashes (/).
+    #[clap(long, env = MEILI_S3_SNAPSHOT_PREFIX, required = false)]
+    #[serde(default)]
+    pub s3_snapshot_prefix: String,
+
+    /// The S3 access key.
+    #[clap(long, env = MEILI_S3_ACCESS_KEY, required = false)]
+    #[serde(default)]
+    pub s3_access_key: String,
+
+    /// The S3 secret key.
+    #[clap(long, env = MEILI_S3_SECRET_KEY, required = false)]
+    #[serde(default)]
+    pub s3_secret_key: String,
+
+    /// The maximum number of parts that can be uploaded in parallel.
+    ///
+    /// For more information, see <https://github.com/orgs/meilisearch/discussions/869>.
+    #[clap(long, env = MEILI_EXPERIMENTAL_S3_MAX_IN_FLIGHT_PARTS, default_value_t = default_experimental_s3_snapshot_max_in_flight_parts())]
+    #[serde(default = "default_experimental_s3_snapshot_max_in_flight_parts")]
+    pub experimental_s3_max_in_flight_parts: NonZeroUsize,
+
+    /// The compression level. Defaults to no compression (0).
+    ///
+    /// For more information, see <https://github.com/orgs/meilisearch/discussions/869>.
+    #[clap(long, env = MEILI_EXPERIMENTAL_S3_COMPRESSION_LEVEL, default_value_t = default_experimental_s3_snapshot_compression_level())]
+    #[serde(default = "default_experimental_s3_snapshot_compression_level")]
+    pub experimental_s3_compression_level: u32,
+
+    /// The signature duration for the multipart upload.
+    ///
+    /// For more information, see <https://github.com/orgs/meilisearch/discussions/869>.
+    #[clap(long, env = MEILI_EXPERIMENTAL_S3_SIGNATURE_DURATION_SECONDS, default_value_t = default_experimental_s3_snapshot_signature_duration_seconds())]
+    #[serde(default = "default_experimental_s3_snapshot_signature_duration_seconds")]
+    pub experimental_s3_signature_duration_seconds: u64,
+
+    /// The size of the the multipart parts.
+    ///
+    /// Must not be less than 10MiB and larger than 8GiB. Yes,
+    /// twice the boundaries of the AWS S3 multipart upload
+    /// because we use it a bit differently internally.
+    ///
+    /// For more information, see <https://github.com/orgs/meilisearch/discussions/869>.
+    #[clap(long, env = MEILI_EXPERIMENTAL_S3_MULTIPART_PART_SIZE, default_value_t = default_experimental_s3_snapshot_multipart_part_size())]
+    #[serde(default = "default_experimental_s3_snapshot_multipart_part_size")]
+    pub experimental_s3_multipart_part_size: Byte,
+}
+
+impl S3SnapshotOpts {
+    /// Exports the values to their corresponding env vars if they are not set.
+    pub fn export_to_env(self) {
+        let S3SnapshotOpts {
+            s3_bucket_url,
+            s3_bucket_region,
+            s3_bucket_name,
+            s3_snapshot_prefix,
+            s3_access_key,
+            s3_secret_key,
+            experimental_s3_max_in_flight_parts,
+            experimental_s3_compression_level,
+            experimental_s3_signature_duration_seconds,
+            experimental_s3_multipart_part_size,
+        } = self;
+
+        export_to_env_if_not_present(MEILI_S3_BUCKET_URL, s3_bucket_url);
+        export_to_env_if_not_present(MEILI_S3_BUCKET_REGION, s3_bucket_region);
+        export_to_env_if_not_present(MEILI_S3_BUCKET_NAME, s3_bucket_name);
+        export_to_env_if_not_present(MEILI_S3_SNAPSHOT_PREFIX, s3_snapshot_prefix);
+        export_to_env_if_not_present(MEILI_S3_ACCESS_KEY, s3_access_key);
+        export_to_env_if_not_present(MEILI_S3_SECRET_KEY, s3_secret_key);
+        export_to_env_if_not_present(
+            MEILI_EXPERIMENTAL_S3_MAX_IN_FLIGHT_PARTS,
+            experimental_s3_max_in_flight_parts.to_string(),
+        );
+        export_to_env_if_not_present(
+            MEILI_EXPERIMENTAL_S3_COMPRESSION_LEVEL,
+            experimental_s3_compression_level.to_string(),
+        );
+        export_to_env_if_not_present(
+            MEILI_EXPERIMENTAL_S3_SIGNATURE_DURATION_SECONDS,
+            experimental_s3_signature_duration_seconds.to_string(),
+        );
+        export_to_env_if_not_present(
+            MEILI_EXPERIMENTAL_S3_MULTIPART_PART_SIZE,
+            experimental_s3_multipart_part_size.to_string(),
+        );
+    }
+}
+
+impl TryFrom<S3SnapshotOpts> for S3SnapshotOptions {
+    type Error = anyhow::Error;
+
+    fn try_from(other: S3SnapshotOpts) -> Result<Self, Self::Error> {
+        let S3SnapshotOpts {
+            s3_bucket_url,
+            s3_bucket_region,
+            s3_bucket_name,
+            s3_snapshot_prefix,
+            s3_access_key,
+            s3_secret_key,
+            experimental_s3_max_in_flight_parts,
+            experimental_s3_compression_level,
+            experimental_s3_signature_duration_seconds,
+            experimental_s3_multipart_part_size,
+        } = other;
+
+        Ok(S3SnapshotOptions {
+            s3_bucket_url,
+            s3_bucket_region,
+            s3_bucket_name,
+            s3_snapshot_prefix,
+            s3_access_key,
+            s3_secret_key,
+            s3_max_in_flight_parts: experimental_s3_max_in_flight_parts,
+            s3_compression_level: experimental_s3_compression_level,
+            s3_signature_duration: Duration::from_secs(experimental_s3_signature_duration_seconds),
+            s3_multipart_part_size: experimental_s3_multipart_part_size.as_u64(),
        })
    }
 }
@@ -1089,6 +1289,22 @@ fn default_snapshot_interval_sec() -> &'static str {
    DEFAULT_SNAPSHOT_INTERVAL_SEC_STR
 }

+fn default_experimental_s3_snapshot_max_in_flight_parts() -> NonZeroUsize {
+    DEFAULT_S3_SNAPSHOT_MAX_IN_FLIGHT_PARTS
+}
+
+fn default_experimental_s3_snapshot_compression_level() -> u32 {
+    DEFAULT_S3_SNAPSHOT_COMPRESSION_LEVEL
+}
+
+fn default_experimental_s3_snapshot_signature_duration_seconds() -> u64 {
+    DEFAULT_S3_SNAPSHOT_SIGNATURE_DURATION_SECONDS
+}
+
+fn default_experimental_s3_snapshot_multipart_part_size() -> Byte {
+    DEFAULT_S3_SNAPSHOT_MULTIPART_PART_SIZE
+}
+
 fn default_dump_dir() -> PathBuf {
    PathBuf::from(DEFAULT_DUMP_DIR)
 }
--- a/crates/meilisearch/src/personalization/mod.rs
+++ b/crates/meilisearch/src/personalization/mod.rs
@@ -0,0 +1,366 @@
+use crate::search::{Personalize, SearchResult};
+use meilisearch_types::{
+    error::{Code, ErrorCode, ResponseError},
+    milli::TimeBudget,
+};
+use rand::Rng;
+use reqwest::Client;
+use serde::{Deserialize, Serialize};
+use std::time::Duration;
+use tracing::{debug, info, warn};
+
+const COHERE_API_URL: &str = "https://api.cohere.ai/v1/rerank";
+const MAX_RETRIES: u32 = 10;
+
+#[derive(Debug, thiserror::Error)]
+enum PersonalizationError {
+    #[error("Personalization service: HTTP request failed: {0}")]
+    Request(#[from] reqwest::Error),
+    #[error("Personalization service: Failed to parse response: {0}")]
+    Parse(String),
+    #[error("Personalization service: Cohere API error: {0}")]
+    Api(String),
+    #[error("Personalization service: Unauthorized: invalid API key")]
+    Unauthorized,
+    #[error("Personalization service: Rate limited: too many requests")]
+    RateLimited,
+    #[error("Personalization service: Bad request: {0}")]
+    BadRequest(String),
+    #[error("Personalization service: Internal server error: {0}")]
+    InternalServerError(String),
+    #[error("Personalization service: Network error: {0}")]
+    Network(String),
+    #[error("Personalization service: Deadline exceeded")]
+    DeadlineExceeded,
+    #[error(transparent)]
+    FeatureNotEnabled(#[from] index_scheduler::error::FeatureNotEnabledError),
+}
+
+impl ErrorCode for PersonalizationError {
+    fn error_code(&self) -> Code {
+        match self {
+            PersonalizationError::FeatureNotEnabled { .. } => Code::FeatureNotEnabled,
+            PersonalizationError::Unauthorized => Code::RemoteInvalidApiKey,
+            PersonalizationError::RateLimited => Code::TooManySearchRequests,
+            PersonalizationError::BadRequest(_) => Code::RemoteBadRequest,
+            PersonalizationError::InternalServerError(_) => Code::RemoteRemoteError,
+            PersonalizationError::Network(_) | PersonalizationError::Request(_) => {
+                Code::RemoteCouldNotSendRequest
+            }
+            PersonalizationError::Parse(_) | PersonalizationError::Api(_) => {
+                Code::RemoteBadResponse
+            }
+            PersonalizationError::DeadlineExceeded => Code::Internal, // should not be returned to the client
+        }
+    }
+}
+
+pub struct CohereService {
+    client: Client,
+    api_key: String,
+}
+
+impl CohereService {
+    pub fn new(api_key: String) -> Self {
+        info!("Personalization service initialized with Cohere API");
+        let client = Client::builder()
+            .timeout(Duration::from_secs(30))
+            .build()
+            .expect("Failed to create HTTP client");
+        Self { client, api_key }
+    }
+
+    pub async fn rerank_search_results(
+        &self,
+        search_result: SearchResult,
+        personalize: &Personalize,
+        query: Option<&str>,
+        time_budget: TimeBudget,
+    ) -> Result<SearchResult, ResponseError> {
+        if time_budget.exceeded() {
+            warn!("Could not rerank due to deadline");
+            // If the deadline is exceeded, return the original search result instead of an error
+            return Ok(search_result);
+        }
+
+        // Extract user context from personalization
+        let user_context = personalize.user_context.as_str();
+
+        // Build the prompt by merging query and user context
+        let prompt = match query {
+            Some(q) => format!("User Context: {user_context}\nQuery: {q}"),
+            None => format!("User Context: {user_context}"),
+        };
+
+        // Extract documents for reranking
+        let documents: Vec<String> = search_result
+            .hits
+            .iter()
+            .map(|hit| {
+                // Convert the document to a string representation for reranking
+                serde_json::to_string(&hit.document).unwrap_or_else(|_| "{}".to_string())
+            })
+            .collect();
+
+        if documents.is_empty() {
+            return Ok(search_result);
+        }
+
+        // Call Cohere's rerank API with retry logic
+        let reranked_indices =
+            match self.call_rerank_with_retry(&prompt, &documents, time_budget).await {
+                Ok(indices) => indices,
+                Err(PersonalizationError::DeadlineExceeded) => {
+                    // If the deadline is exceeded, return the original search result instead of an error
+                    return Ok(search_result);
+                }
+                Err(e) => return Err(e.into()),
+            };
+
+        debug!("Cohere rerank successful, reordering {} results", search_result.hits.len());
+
+        // Reorder the hits based on Cohere's reranking
+        let mut reranked_hits = Vec::new();
+        for index in reranked_indices.iter() {
+            if let Some(hit) = search_result.hits.get(*index) {
+                reranked_hits.push(hit.clone());
+            }
+        }
+
+        Ok(SearchResult { hits: reranked_hits, ..search_result })
+    }
+
+    async fn call_rerank_with_retry(
+        &self,
+        query: &str,
+        documents: &[String],
+        time_budget: TimeBudget,
+    ) -> Result<Vec<usize>, PersonalizationError> {
+        let request_body = CohereRerankRequest {
+            query: query.to_string(),
+            documents: documents.to_vec(),
+            model: "rerank-english-v3.0".to_string(),
+        };
+
+        // Retry loop similar to vector extraction
+        for attempt in 0..MAX_RETRIES {
+            let response_result = self.send_rerank_request(&request_body).await;
+
+            let retry_duration = match self.handle_response(response_result).await {
+                Ok(indices) => return Ok(indices),
+                Err(retry) => {
+                    warn!("Cohere rerank attempt #{} failed: {}", attempt, retry.error);
+
+                    if time_budget.exceeded() {
+                        warn!("Could not rerank due to deadline");
+                        return Err(PersonalizationError::DeadlineExceeded);
+                    } else {
+                        match retry.into_duration(attempt) {
+                            Ok(d) => d,
+                            Err(error) => return Err(error),
+                        }
+                    }
+                }
+            };
+
+            // randomly up to double the retry duration
+            let retry_duration = retry_duration
+                + rand::thread_rng().gen_range(std::time::Duration::ZERO..retry_duration);
+
+            warn!("Retrying after {}ms", retry_duration.as_millis());
+            tokio::time::sleep(retry_duration).await;
+        }
+
+        // Final attempt without retry
+        let response_result = self.send_rerank_request(&request_body).await;
+
+        match self.handle_response(response_result).await {
+            Ok(indices) => Ok(indices),
+            Err(retry) => Err(retry.into_error()),
+        }
+    }
+
+    async fn send_rerank_request(
+        &self,
+        request_body: &CohereRerankRequest,
+    ) -> Result<reqwest::Response, reqwest::Error> {
+        self.client
+            .post(COHERE_API_URL)
+            .header("Authorization", format!("Bearer {}", self.api_key))
+            .header("Content-Type", "application/json")
+            .json(request_body)
+            .send()
+            .await
+    }
+
+    async fn handle_response(
+        &self,
+        response_result: Result<reqwest::Response, reqwest::Error>,
+    ) -> Result<Vec<usize>, Retry> {
+        let response = match response_result {
+            Ok(r) => r,
+            Err(e) if e.is_timeout() => {
+                return Err(Retry::retry_later(PersonalizationError::Network(format!(
+                    "Request timeout: {}",
+                    e
+                ))));
+            }
+            Err(e) => {
+                return Err(Retry::retry_later(PersonalizationError::Network(format!(
+                    "Network error: {}",
+                    e
+                ))));
+            }
+        };
+
+        let status = response.status();
+        let status_code = status.as_u16();
+
+        if status.is_success() {
+            let rerank_response: CohereRerankResponse = match response.json().await {
+                Ok(r) => r,
+                Err(e) => {
+                    return Err(Retry::retry_later(PersonalizationError::Parse(format!(
+                        "Failed to parse response: {}",
+                        e
+                    ))));
+                }
+            };
+
+            // Extract indices from rerank results
+            let indices: Vec<usize> =
+                rerank_response.results.iter().map(|result| result.index as usize).collect();
+
+            return Ok(indices);
+        }
+
+        // Handle error status codes
+        let error_body = response.text().await.unwrap_or_else(|_| "Unknown error".to_string());
+
+        let retry = match status_code {
+            401 => Retry::give_up(PersonalizationError::Unauthorized),
+            429 => Retry::rate_limited(PersonalizationError::RateLimited),
+            400 => Retry::give_up(PersonalizationError::BadRequest(error_body)),
+            500..=599 => Retry::retry_later(PersonalizationError::InternalServerError(format!(
+                "Status {}: {}",
+                status_code, error_body
+            ))),
+            402..=499 => Retry::give_up(PersonalizationError::Api(format!(
+                "Status {}: {}",
+                status_code, error_body
+            ))),
+            _ => Retry::retry_later(PersonalizationError::Api(format!(
+                "Unexpected status {}: {}",
+                status_code, error_body
+            ))),
+        };
+
+        Err(retry)
+    }
+}
+
+#[derive(Serialize)]
+struct CohereRerankRequest {
+    query: String,
+    documents: Vec<String>,
+    model: String,
+}
+
+#[derive(Deserialize)]
+struct CohereRerankResponse {
+    results: Vec<CohereRerankResult>,
+}
+
+#[derive(Deserialize)]
+struct CohereRerankResult {
+    index: u32,
+}
+
+// Retry strategy similar to vector extraction
+struct Retry {
+    error: PersonalizationError,
+    strategy: RetryStrategy,
+}
+
+enum RetryStrategy {
+    GiveUp,
+    Retry,
+    RetryAfterRateLimit,
+}
+
+impl Retry {
+    fn give_up(error: PersonalizationError) -> Self {
+        Self { error, strategy: RetryStrategy::GiveUp }
+    }
+
+    fn retry_later(error: PersonalizationError) -> Self {
+        Self { error, strategy: RetryStrategy::Retry }
+    }
+
+    fn rate_limited(error: PersonalizationError) -> Self {
+        Self { error, strategy: RetryStrategy::RetryAfterRateLimit }
+    }
+
+    fn into_duration(self, attempt: u32) -> Result<Duration, PersonalizationError> {
+        match self.strategy {
+            RetryStrategy::GiveUp => Err(self.error),
+            RetryStrategy::Retry => {
+                // Exponential backoff: 10^attempt milliseconds
+                Ok(Duration::from_millis((10u64).pow(attempt)))
+            }
+            RetryStrategy::RetryAfterRateLimit => {
+                // Longer backoff for rate limits: 100ms + exponential
+                Ok(Duration::from_millis(100 + (10u64).pow(attempt)))
+            }
+        }
+    }
+
+    fn into_error(self) -> PersonalizationError {
+        self.error
+    }
+}
+
+pub enum PersonalizationService {
+    Cohere(CohereService),
+    Disabled,
+}
+
+impl PersonalizationService {
+    pub fn cohere(api_key: String) -> Self {
+        // If the API key is empty, consider the personalization service as disabled
+        if api_key.trim().is_empty() {
+            Self::disabled()
+        } else {
+            Self::Cohere(CohereService::new(api_key))
+        }
+    }
+
+    pub fn disabled() -> Self {
+        debug!("Personalization service disabled");
+        Self::Disabled
+    }
+
+    pub async fn rerank_search_results(
+        &self,
+        search_result: SearchResult,
+        personalize: &Personalize,
+        query: Option<&str>,
+        time_budget: TimeBudget,
+    ) -> Result<SearchResult, ResponseError> {
+        match self {
+            Self::Cohere(cohere_service) => {
+                cohere_service
+                    .rerank_search_results(search_result, personalize, query, time_budget)
+                    .await
+            }
+            Self::Disabled => Err(PersonalizationError::FeatureNotEnabled(
+                index_scheduler::error::FeatureNotEnabledError {
+                    disabled_action: "reranking search results",
+                    feature: "personalization",
+                    issue_link: "https://github.com/orgs/meilisearch/discussions/866",
+                },
+            )
+            .into()),
+        }
+    }
+}
--- a/crates/meilisearch/src/routes/indexes/facet_search.rs
+++ b/crates/meilisearch/src/routes/indexes/facet_search.rs
@@ -343,6 +343,7 @@ impl From<FacetSearchQuery> for SearchQuery {
            hybrid,
            ranking_score_threshold,
            locales,
+            personalize: None,
        }
    }
 }
--- a/crates/meilisearch/src/routes/indexes/search.rs
+++ b/crates/meilisearch/src/routes/indexes/search.rs
@@ -24,9 +24,9 @@ use crate::metrics::MEILISEARCH_DEGRADED_SEARCH_REQUESTS;
 use crate::routes::indexes::search_analytics::{SearchAggregator, SearchGET, SearchPOST};
 use crate::routes::parse_include_metadata_header;
 use crate::search::{
-    add_search_rules, perform_search, HybridQuery, MatchingStrategy, RankingScoreThreshold,
-    RetrieveVectors, SearchKind, SearchParams, SearchQuery, SearchResult, SemanticRatio,
-    DEFAULT_CROP_LENGTH, DEFAULT_CROP_MARKER, DEFAULT_HIGHLIGHT_POST_TAG,
+    add_search_rules, perform_search, HybridQuery, MatchingStrategy, Personalize,
+    RankingScoreThreshold, RetrieveVectors, SearchKind, SearchParams, SearchQuery, SearchResult,
+    SemanticRatio, DEFAULT_CROP_LENGTH, DEFAULT_CROP_MARKER, DEFAULT_HIGHLIGHT_POST_TAG,
    DEFAULT_HIGHLIGHT_PRE_TAG, DEFAULT_SEARCH_LIMIT, DEFAULT_SEARCH_OFFSET, DEFAULT_SEMANTIC_RATIO,
 };
 use crate::search_queue::SearchQueue;
@@ -134,6 +134,8 @@ pub struct SearchQueryGet {
    #[deserr(default, error = DeserrQueryParamError<InvalidSearchLocales>)]
    #[param(value_type = Vec<Locale>, explode = false)]
    pub locales: Option<CS<Locale>>,
+    #[deserr(default, error = DeserrQueryParamError<InvalidSearchPersonalizeUserContext>)]
+    pub personalize_user_context: Option<String>,
 }

 #[derive(Debug, Clone, Copy, PartialEq, deserr::Deserr)]
@@ -205,6 +207,9 @@ impl TryFrom<SearchQueryGet> for SearchQuery {
            ));
        }

+        let personalize =
+            other.personalize_user_context.map(|user_context| Personalize { user_context });
+
        Ok(Self {
            q: other.q,
            // `media` not supported for `GET`
@@ -234,6 +239,7 @@ impl TryFrom<SearchQueryGet> for SearchQuery {
            hybrid,
            ranking_score_threshold: other.ranking_score_threshold.map(|o| o.0),
            locales: other.locales.map(|o| o.into_iter().collect()),
+            personalize,
        })
    }
 }
@@ -322,6 +328,7 @@ pub fn fix_sort_query_parameters(sort_query: &str) -> Vec<String> {
 pub async fn search_with_url_query(
    index_scheduler: GuardedData<ActionPolicy<{ actions::SEARCH }>, Data<IndexScheduler>>,
    search_queue: web::Data<SearchQueue>,
+    personalization_service: web::Data<crate::personalization::PersonalizationService>,
    index_uid: web::Path<String>,
    params: AwebQueryParameter<SearchQueryGet, DeserrQueryParamError>,
    req: HttpRequest,
@@ -342,9 +349,16 @@ pub async fn search_with_url_query(

    let index = index_scheduler.index(&index_uid)?;

+    // Extract personalization and query string before moving query
+    let personalize = query.personalize.take();
+
    let search_kind =
        search_kind(&query, index_scheduler.get_ref(), index_uid.to_string(), &index)?;
    let retrieve_vector = RetrieveVectors::new(query.retrieve_vectors);
+
+    // Save the query string for personalization if requested
+    let personalize_query = personalize.is_some().then(|| query.q.clone()).flatten();
+
    let permit = search_queue.try_get_search_permit().await?;
    let include_metadata = parse_include_metadata_header(&req);

@@ -365,12 +379,24 @@ pub async fn search_with_url_query(
    .await;
    permit.drop().await;
    let search_result = search_result?;
-    if let Ok(ref search_result) = search_result {
+    if let Ok((search_result, _)) = search_result.as_ref() {
        aggregate.succeed(search_result);
    }
    analytics.publish(aggregate, &req);

-    let search_result = search_result?;
+    let (mut search_result, time_budget) = search_result?;
+
+    // Apply personalization if requested
+    if let Some(personalize) = personalize.as_ref() {
+        search_result = personalization_service
+            .rerank_search_results(
+                search_result,
+                personalize,
+                personalize_query.as_deref(),
+                time_budget,
+            )
+            .await?;
+    }

    debug!(request_uid = ?request_uid, returns = ?search_result, "Search get");
    Ok(HttpResponse::Ok().json(search_result))
@@ -435,6 +461,7 @@ pub async fn search_with_url_query(
 pub async fn search_with_post(
    index_scheduler: GuardedData<ActionPolicy<{ actions::SEARCH }>, Data<IndexScheduler>>,
    search_queue: web::Data<SearchQueue>,
+    personalization_service: web::Data<crate::personalization::PersonalizationService>,
    index_uid: web::Path<String>,
    params: AwebJson<SearchQuery, DeserrJsonError>,
    req: HttpRequest,
@@ -455,12 +482,18 @@ pub async fn search_with_post(

    let index = index_scheduler.index(&index_uid)?;

+    // Extract personalization and query string before moving query
+    let personalize = query.personalize.take();
+
    let search_kind =
        search_kind(&query, index_scheduler.get_ref(), index_uid.to_string(), &index)?;
    let retrieve_vectors = RetrieveVectors::new(query.retrieve_vectors);

    let include_metadata = parse_include_metadata_header(&req);

+    // Save the query string for personalization if requested
+    let personalize_query = personalize.is_some().then(|| query.q.clone()).flatten();
+
    let permit = search_queue.try_get_search_permit().await?;
    let search_result = tokio::task::spawn_blocking(move || {
        perform_search(
@@ -479,7 +512,7 @@ pub async fn search_with_post(
    .await;
    permit.drop().await;
    let search_result = search_result?;
-    if let Ok(ref search_result) = search_result {
+    if let Ok((ref search_result, _)) = search_result {
        aggregate.succeed(search_result);
        if search_result.degraded {
            MEILISEARCH_DEGRADED_SEARCH_REQUESTS.inc();
@@ -487,7 +520,19 @@ pub async fn search_with_post(
    }
    analytics.publish(aggregate, &req);

-    let search_result = search_result?;
+    let (mut search_result, time_budget) = search_result?;
+
+    // Apply personalization if requested
+    if let Some(personalize) = personalize.as_ref() {
+        search_result = personalization_service
+            .rerank_search_results(
+                search_result,
+                personalize,
+                personalize_query.as_deref(),
+                time_budget,
+            )
+            .await?;
+    }

    debug!(request_uid = ?request_uid, returns = ?search_result, "Search post");
    Ok(HttpResponse::Ok().json(search_result))
--- a/crates/meilisearch/src/routes/indexes/search_analytics.rs
+++ b/crates/meilisearch/src/routes/indexes/search_analytics.rs
@@ -7,6 +7,7 @@ use serde_json::{json, Value};

 use crate::aggregate_methods;
 use crate::analytics::{Aggregate, AggregateMethod};
+use crate::metrics::MEILISEARCH_PERSONALIZED_SEARCH_REQUESTS;
 use crate::search::{
    SearchQuery, SearchResult, DEFAULT_CROP_LENGTH, DEFAULT_CROP_MARKER,
    DEFAULT_HIGHLIGHT_POST_TAG, DEFAULT_HIGHLIGHT_PRE_TAG, DEFAULT_SEARCH_LIMIT,
@@ -95,6 +96,9 @@ pub struct SearchAggregator<Method: AggregateMethod> {
    show_ranking_score_details: bool,
    ranking_score_threshold: bool,

+    // personalization
+    total_personalized: usize,
+
    marker: std::marker::PhantomData<Method>,
 }

@@ -129,6 +133,7 @@ impl<Method: AggregateMethod> SearchAggregator<Method> {
            hybrid,
            ranking_score_threshold,
            locales,
+            personalize,
        } = query;

        let mut ret = Self::default();
@@ -204,6 +209,12 @@ impl<Method: AggregateMethod> SearchAggregator<Method> {
            ret.locales = locales.iter().copied().collect();
        }

+        // personalization
+        if personalize.is_some() {
+            ret.total_personalized = 1;
+            MEILISEARCH_PERSONALIZED_SEARCH_REQUESTS.inc();
+        }
+
        ret.highlight_pre_tag = *highlight_pre_tag != DEFAULT_HIGHLIGHT_PRE_TAG();
        ret.highlight_post_tag = *highlight_post_tag != DEFAULT_HIGHLIGHT_POST_TAG();
        ret.crop_marker = *crop_marker != DEFAULT_CROP_MARKER();
@@ -296,6 +307,7 @@ impl<Method: AggregateMethod> Aggregate for SearchAggregator<Method> {
            total_used_negative_operator,
            ranking_score_threshold,
            mut locales,
+            total_personalized,
            marker: _,
        } = *new;

@@ -381,6 +393,9 @@ impl<Method: AggregateMethod> Aggregate for SearchAggregator<Method> {
        // locales
        self.locales.append(&mut locales);

+        // personalization
+        self.total_personalized = self.total_personalized.saturating_add(total_personalized);
+
        self
    }

@@ -426,6 +441,7 @@ impl<Method: AggregateMethod> Aggregate for SearchAggregator<Method> {
            total_used_negative_operator,
            ranking_score_threshold,
            locales,
+            total_personalized,
            marker: _,
        } = *self;

@@ -499,6 +515,9 @@ impl<Method: AggregateMethod> Aggregate for SearchAggregator<Method> {
                "show_ranking_score_details": show_ranking_score_details,
                "ranking_score_threshold": ranking_score_threshold,
            },
+            "personalization": {
+                "total_personalized": total_personalized,
+            },
        })
    }
 }
--- a/crates/meilisearch/src/routes/mod.rs
+++ b/crates/meilisearch/src/routes/mod.rs
@@ -41,7 +41,9 @@ use crate::routes::indexes::IndexView;
 use crate::routes::multi_search::SearchResults;
 use crate::routes::network::{Network, Remote};
 use crate::routes::swap_indexes::SwapIndexesPayload;
-use crate::routes::webhooks::{WebhookResults, WebhookSettings, WebhookWithMetadata};
+use crate::routes::webhooks::{
+    WebhookResults, WebhookSettings, WebhookWithMetadataRedactedAuthorization,
+};
 use crate::search::{
    FederatedSearch, FederatedSearchResult, Federation, FederationOptions, MergeFacets,
    SearchQueryWithIndex, SearchResultWithIndex, SimilarQuery, SimilarResult,
@@ -103,7 +105,7 @@ mod webhooks;
        url = "/",
        description = "Local server",
    )),
-    components(schemas(PaginationView<KeyView>, PaginationView<IndexView>, IndexView, DocumentDeletionByFilter, AllBatches, BatchStats, ProgressStepView, ProgressView, BatchView, RuntimeTogglableFeatures, SwapIndexesPayload, DocumentEditionByFunction, MergeFacets, FederationOptions, SearchQueryWithIndex, Federation, FederatedSearch, FederatedSearchResult, SearchResults, SearchResultWithIndex, SimilarQuery, SimilarResult, PaginationView<serde_json::Value>, BrowseQuery, UpdateIndexRequest, IndexUid, IndexCreateRequest, KeyView, Action, CreateApiKey, UpdateStderrLogs, LogMode, GetLogs, IndexStats, Stats, HealthStatus, HealthResponse, VersionResponse, Code, ErrorType, AllTasks, TaskView, Status, DetailsView, ResponseError, Settings<Unchecked>, Settings<Checked>, TypoSettings, MinWordSizeTyposSetting, FacetingSettings, PaginationSettings, SummarizedTaskView, Kind, Network, Remote, FilterableAttributesRule, FilterableAttributesPatterns, AttributePatterns, FilterableAttributesFeatures, FilterFeatures, Export, WebhookSettings, WebhookResults, WebhookWithMetadata, meilisearch_types::milli::vector::VectorStoreBackend))
+    components(schemas(PaginationView<KeyView>, PaginationView<IndexView>, IndexView, DocumentDeletionByFilter, AllBatches, BatchStats, ProgressStepView, ProgressView, BatchView, RuntimeTogglableFeatures, SwapIndexesPayload, DocumentEditionByFunction, MergeFacets, FederationOptions, SearchQueryWithIndex, Federation, FederatedSearch, FederatedSearchResult, SearchResults, SearchResultWithIndex, SimilarQuery, SimilarResult, PaginationView<serde_json::Value>, BrowseQuery, UpdateIndexRequest, IndexUid, IndexCreateRequest, KeyView, Action, CreateApiKey, UpdateStderrLogs, LogMode, GetLogs, IndexStats, Stats, HealthStatus, HealthResponse, VersionResponse, Code, ErrorType, AllTasks, TaskView, Status, DetailsView, ResponseError, Settings<Unchecked>, Settings<Checked>, TypoSettings, MinWordSizeTyposSetting, FacetingSettings, PaginationSettings, SummarizedTaskView, Kind, Network, Remote, FilterableAttributesRule, FilterableAttributesPatterns, AttributePatterns, FilterableAttributesFeatures, FilterFeatures, Export, WebhookSettings, WebhookResults, WebhookWithMetadataRedactedAuthorization, meilisearch_types::milli::vector::VectorStoreBackend))
 )]
 pub struct MeilisearchApi;

--- a/crates/meilisearch/src/routes/multi_search.rs
+++ b/crates/meilisearch/src/routes/multi_search.rs
@@ -146,6 +146,7 @@ pub struct SearchResults {
 pub async fn multi_search_with_post(
    index_scheduler: GuardedData<ActionPolicy<{ actions::SEARCH }>, Data<IndexScheduler>>,
    search_queue: Data<SearchQueue>,
+    personalization_service: web::Data<crate::personalization::PersonalizationService>,
    params: AwebJson<FederatedSearch, DeserrJsonError>,
    req: HttpRequest,
    analytics: web::Data<Analytics>,
@@ -236,7 +237,7 @@ pub async fn multi_search_with_post(
            // changes.
            let search_results: Result<_, (ResponseError, usize)> = async {
                let mut search_results = Vec::with_capacity(queries.len());
-                for (query_index, (index_uid, query, federation_options)) in queries
+                for (query_index, (index_uid, mut query, federation_options)) in queries
                    .into_iter()
                    .map(SearchQueryWithIndex::into_index_query_federation)
                    .enumerate()
@@ -269,6 +270,13 @@ pub async fn multi_search_with_post(
                        })
                        .with_index(query_index)?;

+                    // Extract personalization and query string before moving query
+                    let personalize = query.personalize.take();
+
+                    // Save the query string for personalization if requested
+                    let personalize_query =
+                        personalize.is_some().then(|| query.q.clone()).flatten();
+
                    let index_uid_str = index_uid.to_string();

                    let search_kind = search_kind(
@@ -280,7 +288,7 @@ pub async fn multi_search_with_post(
                    .with_index(query_index)?;
                    let retrieve_vector = RetrieveVectors::new(query.retrieve_vectors);

-                    let search_result = tokio::task::spawn_blocking(move || {
+                    let (mut search_result, time_budget) = tokio::task::spawn_blocking(move || {
                        perform_search(
                            SearchParams {
                                index_uid: index_uid_str.clone(),
@@ -295,11 +303,25 @@ pub async fn multi_search_with_post(
                        )
                    })
                    .await
+                    .with_index(query_index)?
                    .with_index(query_index)?;

+                    // Apply personalization if requested
+                    if let Some(personalize) = personalize.as_ref() {
+                        search_result = personalization_service
+                            .rerank_search_results(
+                                search_result,
+                                personalize,
+                                personalize_query.as_deref(),
+                                time_budget,
+                            )
+                            .await
+                            .with_index(query_index)?;
+                    }
+
                    search_results.push(SearchResultWithIndex {
                        index_uid: index_uid.into_inner(),
-                        result: search_result.with_index(query_index)?,
+                        result: search_result,
                    });
                }
                Ok(search_results)
--- a/crates/meilisearch/src/routes/multi_search_analytics.rs
+++ b/crates/meilisearch/src/routes/multi_search_analytics.rs
@@ -67,6 +67,7 @@ impl MultiSearchAggregator {
            hybrid: _,
            ranking_score_threshold: _,
            locales: _,
+            personalize: _,
        } in &federated_search.queries
        {
            if let Some(federation_options) = federation_options {
--- a/crates/meilisearch/src/routes/webhooks.rs
+++ b/crates/meilisearch/src/routes/webhooks.rs
@@ -90,7 +90,7 @@ fn deny_immutable_fields_webhook(
 #[derive(Debug, Serialize, ToSchema)]
 #[serde(rename_all = "camelCase")]
 #[schema(rename_all = "camelCase")]
-pub(super) struct WebhookWithMetadata {
+pub(super) struct WebhookWithMetadataRedactedAuthorization {
    uuid: Uuid,
    is_editable: bool,
    #[schema(value_type = WebhookSettings)]
@@ -98,8 +98,9 @@ pub(super) struct WebhookWithMetadata {
    webhook: Webhook,
 }

-impl WebhookWithMetadata {
-    pub fn from(uuid: Uuid, webhook: Webhook) -> Self {
+impl WebhookWithMetadataRedactedAuthorization {
+    pub fn from(uuid: Uuid, mut webhook: Webhook) -> Self {
+        webhook.redact_authorization_header();
        Self { uuid, is_editable: uuid != Uuid::nil(), webhook }
    }
 }
@@ -107,7 +108,7 @@ impl WebhookWithMetadata {
 #[derive(Debug, Serialize, ToSchema)]
 #[serde(rename_all = "camelCase")]
 pub(super) struct WebhookResults {
-    results: Vec<WebhookWithMetadata>,
+    results: Vec<WebhookWithMetadataRedactedAuthorization>,
 }

 #[utoipa::path(
@@ -150,7 +151,7 @@ async fn get_webhooks(
    let results = webhooks
        .webhooks
        .into_iter()
-        .map(|(uuid, webhook)| WebhookWithMetadata::from(uuid, webhook))
+        .map(|(uuid, webhook)| WebhookWithMetadataRedactedAuthorization::from(uuid, webhook))
        .collect::<Vec<_>>();
    let results = WebhookResults { results };

@@ -301,7 +302,7 @@ fn check_changed(uuid: Uuid, webhook: &Webhook) -> Result<(), WebhooksError> {
    tag = "Webhooks",
    security(("Bearer" = ["webhooks.get", "webhooks.*", "*.get", "*"])),
    responses(
-        (status = 200, description = "Webhook found", body = WebhookWithMetadata, content_type = "application/json", example = json!({
+        (status = 200, description = "Webhook found", body = WebhookWithMetadataRedactedAuthorization, content_type = "application/json", example = json!({
            "uuid": "550e8400-e29b-41d4-a716-446655440000",
            "url": "https://your.site/on-tasks-completed",
            "headers": {
@@ -324,7 +325,7 @@ async fn get_webhook(
    let mut webhooks = index_scheduler.webhooks_view();

    let webhook = webhooks.webhooks.remove(&uuid).ok_or(WebhookNotFound(uuid))?;
-    let webhook = WebhookWithMetadata::from(uuid, webhook);
+    let webhook = WebhookWithMetadataRedactedAuthorization::from(uuid, webhook);

    debug!(returns = ?webhook, "Get webhook");
    Ok(HttpResponse::Ok().json(webhook))
@@ -337,7 +338,7 @@ async fn get_webhook(
    request_body = WebhookSettings,
    security(("Bearer" = ["webhooks.create", "webhooks.*", "*"])),
    responses(
-        (status = 201, description = "Webhook created successfully", body = WebhookWithMetadata, content_type = "application/json", example = json!({
+        (status = 201, description = "Webhook created successfully", body = WebhookWithMetadataRedactedAuthorization, content_type = "application/json", example = json!({
            "uuid": "550e8400-e29b-41d4-a716-446655440000",
            "url": "https://your.site/on-tasks-completed",
            "headers": {
@@ -383,7 +384,7 @@ async fn post_webhook(

    analytics.publish(PostWebhooksAnalytics, &req);

-    let response = WebhookWithMetadata::from(uuid, webhook);
+    let response = WebhookWithMetadataRedactedAuthorization::from(uuid, webhook);
    debug!(returns = ?response, "Post webhook");
    Ok(HttpResponse::Created().json(response))
 }
@@ -395,7 +396,7 @@ async fn post_webhook(
    request_body = WebhookSettings,
    security(("Bearer" = ["webhooks.update", "webhooks.*", "*"])),
    responses(
-        (status = 200, description = "Webhook updated successfully", body = WebhookWithMetadata, content_type = "application/json", example = json!({
+        (status = 200, description = "Webhook updated successfully", body = WebhookWithMetadataRedactedAuthorization, content_type = "application/json", example = json!({
            "uuid": "550e8400-e29b-41d4-a716-446655440000",
            "url": "https://your.site/on-tasks-completed",
            "headers": {
@@ -435,7 +436,7 @@ async fn patch_webhook(

    analytics.publish(PatchWebhooksAnalytics, &req);

-    let response = WebhookWithMetadata::from(uuid, webhook);
+    let response = WebhookWithMetadataRedactedAuthorization::from(uuid, webhook);
    debug!(returns = ?response, "Patch webhook");
    Ok(HttpResponse::Ok().json(response))
 }
--- a/crates/meilisearch/src/search/federated/perform.rs
+++ b/crates/meilisearch/src/search/federated/perform.rs
@@ -601,6 +601,10 @@ impl PartitionedQueries {
            .into());
        }

+        if federated_query.has_personalize() {
+            return Err(MeilisearchHttpError::PersonalizationInFederatedQuery(query_index).into());
+        }
+
        let (index_uid, query, federation_options) = federated_query.into_index_query_federation();

        let federation_options = federation_options.unwrap_or_default();
--- a/crates/meilisearch/src/search/mod.rs
+++ b/crates/meilisearch/src/search/mod.rs
@@ -59,6 +59,13 @@ pub const DEFAULT_HIGHLIGHT_POST_TAG: fn() -> String = || "</em>".to_string();
 pub const DEFAULT_SEMANTIC_RATIO: fn() -> SemanticRatio = || SemanticRatio(0.5);
 pub const INCLUDE_METADATA_HEADER: &str = "Meili-Include-Metadata";

+#[derive(Clone, Default, PartialEq, Deserr, ToSchema, Debug)]
+#[deserr(error = DeserrJsonError<InvalidSearchPersonalize>, rename_all = camelCase, deny_unknown_fields)]
+pub struct Personalize {
+    #[deserr(error = DeserrJsonError<InvalidSearchPersonalizeUserContext>)]
+    pub user_context: String,
+}
+
 #[derive(Clone, Default, PartialEq, Deserr, ToSchema)]
 #[deserr(error = DeserrJsonError, rename_all = camelCase, deny_unknown_fields)]
 pub struct SearchQuery {
@@ -122,6 +129,8 @@ pub struct SearchQuery {
    pub ranking_score_threshold: Option<RankingScoreThreshold>,
    #[deserr(default, error = DeserrJsonError<InvalidSearchLocales>)]
    pub locales: Option<Vec<Locale>>,
+    #[deserr(default, error = DeserrJsonError<InvalidSearchPersonalize>, default)]
+    pub personalize: Option<Personalize>,
 }

 impl From<SearchParameters> for SearchQuery {
@@ -169,6 +178,7 @@ impl From<SearchParameters> for SearchQuery {
            highlight_post_tag: DEFAULT_HIGHLIGHT_POST_TAG(),
            crop_marker: DEFAULT_CROP_MARKER(),
            locales: None,
+            personalize: None,
        }
    }
 }
@@ -250,6 +260,7 @@ impl fmt::Debug for SearchQuery {
            attributes_to_search_on,
            ranking_score_threshold,
            locales,
+            personalize,
        } = self;

        let mut debug = f.debug_struct("SearchQuery");
@@ -338,6 +349,10 @@ impl fmt::Debug for SearchQuery {
            debug.field("locales", &locales);
        }

+        if let Some(personalize) = personalize {
+            debug.field("personalize", &personalize);
+        }
+
        debug.finish()
    }
 }
@@ -543,6 +558,9 @@ pub struct SearchQueryWithIndex {
    pub ranking_score_threshold: Option<RankingScoreThreshold>,
    #[deserr(default, error = DeserrJsonError<InvalidSearchLocales>, default)]
    pub locales: Option<Vec<Locale>>,
+    #[deserr(default, error = DeserrJsonError<InvalidSearchPersonalize>, default)]
+    #[serde(skip)]
+    pub personalize: Option<Personalize>,

    #[deserr(default)]
    pub federation_options: Option<FederationOptions>,
@@ -567,6 +585,10 @@ impl SearchQueryWithIndex {
        self.facets.as_deref().filter(|v| !v.is_empty())
    }

+    pub fn has_personalize(&self) -> bool {
+        self.personalize.is_some()
+    }
+
    pub fn from_index_query_federation(
        index_uid: IndexUid,
        query: SearchQuery,
@@ -600,6 +622,7 @@ impl SearchQueryWithIndex {
            attributes_to_search_on,
            ranking_score_threshold,
            locales,
+            personalize,
        } = query;

        SearchQueryWithIndex {
@@ -631,6 +654,7 @@ impl SearchQueryWithIndex {
            attributes_to_search_on,
            ranking_score_threshold,
            locales,
+            personalize,
            federation_options,
        }
    }
@@ -666,6 +690,7 @@ impl SearchQueryWithIndex {
            hybrid,
            ranking_score_threshold,
            locales,
+            personalize,
        } = self;
        (
            index_uid,
@@ -697,6 +722,7 @@ impl SearchQueryWithIndex {
                hybrid,
                ranking_score_threshold,
                locales,
+                personalize,
                // do not use ..Default::default() here,
                // rather add any missing field from `SearchQuery` to `SearchQueryWithIndex`
            },
@@ -1149,7 +1175,10 @@ pub struct SearchParams {
    pub include_metadata: bool,
 }

-pub fn perform_search(params: SearchParams, index: &Index) -> Result<SearchResult, ResponseError> {
+pub fn perform_search(
+    params: SearchParams,
+    index: &Index,
+) -> Result<(SearchResult, TimeBudget), ResponseError> {
    let SearchParams {
        index_uid,
        query,
@@ -1168,7 +1197,7 @@ pub fn perform_search(params: SearchParams, index: &Index) -> Result<SearchResul
    };

    let (search, is_finite_pagination, max_total_hits, offset) =
-        prepare_search(index, &rtxn, &query, &search_kind, time_budget, features)?;
+        prepare_search(index, &rtxn, &query, &search_kind, time_budget.clone(), features)?;

    let (
        milli::SearchResult {
@@ -1226,6 +1255,7 @@ pub fn perform_search(params: SearchParams, index: &Index) -> Result<SearchResul
        attributes_to_search_on: _,
        filter: _,
        distinct: _,
+        personalize: _,
    } = query;

    let format = AttributesFormat {
@@ -1291,7 +1321,7 @@ pub fn perform_search(params: SearchParams, index: &Index) -> Result<SearchResul
        request_uid: Some(request_uid),
        metadata,
    };
-    Ok(result)
+    Ok((result, time_budget))
 }

 #[derive(Debug, Clone, Default, Serialize, Deserialize, ToSchema)]
--- a/crates/meilisearch/tests/common/server.rs
+++ b/crates/meilisearch/tests/common/server.rs
@@ -49,8 +49,8 @@ impl Server<Owned> {
        }

        let options = default_settings(dir.path());
-
-        let (index_scheduler, auth) = setup_meilisearch(&options).unwrap();
+        let handle = tokio::runtime::Handle::current();
+        let (index_scheduler, auth) = setup_meilisearch(&options, handle).unwrap();
        let service = Service { index_scheduler, auth, options, api_key: None };

        Server { service, _dir: Some(dir), _marker: PhantomData }
@@ -65,7 +65,9 @@ impl Server<Owned> {

        options.master_key = Some("MASTER_KEY".to_string());

-        let (index_scheduler, auth) = setup_meilisearch(&options).unwrap();
+        let handle = tokio::runtime::Handle::current();
+
+        let (index_scheduler, auth) = setup_meilisearch(&options, handle).unwrap();
        let service = Service { index_scheduler, auth, options, api_key: None };

        Server { service, _dir: Some(dir), _marker: PhantomData }
@@ -78,7 +80,9 @@ impl Server<Owned> {
    }

    pub async fn new_with_options(options: Opt) -> Result<Self, anyhow::Error> {
-        let (index_scheduler, auth) = setup_meilisearch(&options)?;
+        let handle = tokio::runtime::Handle::current();
+
+        let (index_scheduler, auth) = setup_meilisearch(&options, handle)?;
        let service = Service { index_scheduler, auth, options, api_key: None };

        Ok(Server { service, _dir: None, _marker: PhantomData })
@@ -217,8 +221,9 @@ impl Server<Shared> {
        }

        let options = default_settings(dir.path());
+        let handle = tokio::runtime::Handle::current();

-        let (index_scheduler, auth) = setup_meilisearch(&options).unwrap();
+        let (index_scheduler, auth) = setup_meilisearch(&options, handle).unwrap();
        let service = Service { index_scheduler, auth, api_key: None, options };

        Server { service, _dir: Some(dir), _marker: PhantomData }
--- a/crates/meilisearch/tests/common/service.rs
+++ b/crates/meilisearch/tests/common/service.rs
@@ -10,8 +10,9 @@ use actix_web::test::TestRequest;
 use actix_web::web::Data;
 use index_scheduler::IndexScheduler;
 use meilisearch::analytics::Analytics;
+use meilisearch::personalization::PersonalizationService;
 use meilisearch::search_queue::SearchQueue;
-use meilisearch::{create_app, Opt, SubscriberForSecondLayer};
+use meilisearch::{create_app, Opt, ServicesData, SubscriberForSecondLayer};
 use meilisearch_auth::AuthController;
 use tracing::level_filters::LevelFilter;
 use tracing_subscriber::Layer;
@@ -135,14 +136,24 @@ impl Service {
            self.options.experimental_search_queue_size,
            NonZeroUsize::new(1).unwrap(),
        );
+        let personalization_service = self
+            .options
+            .experimental_personalization_api_key
+            .clone()
+            .map(PersonalizationService::cohere)
+            .unwrap_or_else(PersonalizationService::disabled);

        actix_web::test::init_service(create_app(
-            self.index_scheduler.clone().into(),
-            self.auth.clone().into(),
-            Data::new(search_queue),
+            ServicesData {
+                index_scheduler: self.index_scheduler.clone().into(),
+                auth: self.auth.clone().into(),
+                search_queue: Data::new(search_queue),
+                personalization_service: Data::new(personalization_service),
+                logs_route_handle: Data::new(route_layer_handle),
+                logs_stderr_handle: Data::new(stderr_layer_handle),
+                analytics: Data::new(Analytics::no_analytics()),
+            },
            self.options.clone(),
-            (route_layer_handle, stderr_layer_handle),
-            Data::new(Analytics::no_analytics()),
            true,
        ))
        .await
--- a/crates/meilisearch/tests/features/mod.rs
+++ b/crates/meilisearch/tests/features/mod.rs
@@ -207,3 +207,118 @@ async fn errors() {
    }
    "###);
 }
+
+#[actix_rt::test]
+async fn search_with_personalization_without_enabling_the_feature() {
+    let server = Server::new().await;
+    let index = server.unique_index();
+
+    // Create the index and add some documents
+    let (task, _code) = index.create(None).await;
+    server.wait_task(task.uid()).await.succeeded();
+
+    let (task, _code) = index
+        .add_documents(
+            json!([
+                {"id": 1, "title": "The Dark Knight", "genre": "Action"},
+                {"id": 2, "title": "Inception", "genre": "Sci-Fi"},
+                {"id": 3, "title": "The Matrix", "genre": "Sci-Fi"}
+            ]),
+            None,
+        )
+        .await;
+    server.wait_task(task.uid()).await.succeeded();
+
+    // Try to search with personalization - should return feature_not_enabled error
+    let (response, code) = index
+        .search_post(json!({
+            "q": "movie",
+            "personalize": {
+                "userContext": "I love science fiction movies"
+            }
+        }))
+        .await;
+
+    meili_snap::snapshot!(code, @"400 Bad Request");
+    meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
+    {
+      "message": "reranking search results requires enabling the `personalization` experimental feature. See https://github.com/orgs/meilisearch/discussions/866",
+      "code": "feature_not_enabled",
+      "type": "invalid_request",
+      "link": "https://docs.meilisearch.com/errors#feature_not_enabled"
+    }
+    "###);
+}
+
+#[actix_rt::test]
+async fn multi_search_with_personalization_without_enabling_the_feature() {
+    let server = Server::new().await;
+    let index = server.unique_index();
+
+    // Create the index and add some documents
+    let (task, _code) = index.create(None).await;
+    server.wait_task(task.uid()).await.succeeded();
+
+    let (task, _code) = index
+        .add_documents(
+            json!([
+                {"id": 1, "title": "The Dark Knight", "genre": "Action"},
+                {"id": 2, "title": "Inception", "genre": "Sci-Fi"},
+                {"id": 3, "title": "The Matrix", "genre": "Sci-Fi"}
+            ]),
+            None,
+        )
+        .await;
+    server.wait_task(task.uid()).await.succeeded();
+
+    // Try to multi-search with personalization - should return feature_not_enabled error
+    let (response, code) = server
+        .multi_search(json!({
+            "queries": [
+                {
+                    "indexUid": index.uid,
+                    "q": "movie",
+                    "personalize": {
+                        "userContext": "I love science fiction movies"
+                    }
+                }
+            ]
+        }))
+        .await;
+
+    meili_snap::snapshot!(code, @"400 Bad Request");
+    meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
+    {
+      "message": "Inside `.queries[0]`: reranking search results requires enabling the `personalization` experimental feature. See https://github.com/orgs/meilisearch/discussions/866",
+      "code": "feature_not_enabled",
+      "type": "invalid_request",
+      "link": "https://docs.meilisearch.com/errors#feature_not_enabled"
+    }
+    "###);
+
+    // Try to federated search with personalization - should return feature_not_enabled error
+    let (response, code) = server
+        .multi_search(json!({
+          "federation": {},
+            "queries": [
+                {
+                    "indexUid": index.uid,
+                    "q": "movie",
+                    "personalize": {
+                        "userContext": "I love science fiction movies"
+                    }
+                }
+            ]
+        }))
+        .await;
+
+    meili_snap::snapshot!(code, @"400 Bad Request");
+    meili_snap::snapshot!(meili_snap::json_string!(response), @r###"
+    {
+      "message": "Inside `.queries[0]`: Using `.personalize` is not allowed in federated queries.\n - Hint: remove `personalize` from query #0 or remove `federation` from the request",
+      "code": "invalid_multi_search_query_personalization",
+      "type": "invalid_request",
+      "link": "https://docs.meilisearch.com/errors#invalid_multi_search_query_personalization"
+    }
+    "###);
+}
--- a/crates/meilisearch/tests/logs/mod.rs
+++ b/crates/meilisearch/tests/logs/mod.rs
@@ -8,8 +8,9 @@ use actix_web::http::header::ContentType;
 use actix_web::web::Data;
 use meili_snap::snapshot;
 use meilisearch::analytics::Analytics;
+use meilisearch::personalization::PersonalizationService;
 use meilisearch::search_queue::SearchQueue;
-use meilisearch::{create_app, Opt, SubscriberForSecondLayer};
+use meilisearch::{create_app, Opt, ServicesData, SubscriberForSecondLayer};
 use tracing::level_filters::LevelFilter;
 use tracing_subscriber::layer::SubscriberExt;
 use tracing_subscriber::Layer;
@@ -50,12 +51,16 @@ async fn basic_test_log_stream_route() {
    );

    let app = actix_web::test::init_service(create_app(
-        server.service.index_scheduler.clone().into(),
-        server.service.auth.clone().into(),
-        Data::new(search_queue),
+        ServicesData {
+            index_scheduler: server.service.index_scheduler.clone().into(),
+            auth: server.service.auth.clone().into(),
+            search_queue: Data::new(search_queue),
+            personalization_service: Data::new(PersonalizationService::disabled()),
+            logs_route_handle: Data::new(route_layer_handle),
+            logs_stderr_handle: Data::new(stderr_layer_handle),
+            analytics: Data::new(Analytics::no_analytics()),
+        },
        server.service.options.clone(),
-        (route_layer_handle, stderr_layer_handle),
-        Data::new(Analytics::no_analytics()),
        true,
    ))
    .await;
--- a/crates/meilisearch/tests/search/errors.rs
+++ b/crates/meilisearch/tests/search/errors.rs
@@ -1,7 +1,11 @@
 use meili_snap::*;
+use meilisearch::Opt;
+use tempfile::TempDir;

 use super::test_settings_documents_indexing_swapping_and_search;
-use crate::common::{shared_does_not_exists_index, Server, DOCUMENTS, NESTED_DOCUMENTS};
+use crate::common::{
+    default_settings, shared_does_not_exists_index, Server, DOCUMENTS, NESTED_DOCUMENTS,
+};
 use crate::json;

 #[actix_rt::test]
@@ -1320,3 +1324,98 @@ async fn search_with_contains_without_enabling_the_feature() {
    }
    "#);
 }
+
+#[actix_rt::test]
+#[ignore]
+async fn search_with_personalization_invalid_api_key() {
+    // Create a server with a fake personalization API key
+    let dir = TempDir::new().unwrap();
+    let options = Opt {
+        experimental_personalization_api_key: Some("fake-api-key-12345".to_string()),
+        ..default_settings(dir.path())
+    };
+    let server = Server::new_with_options(options).await.unwrap();
+    let index = server.unique_index();
+
+    // Create the index and add some documents
+    let (task, _code) = index.create(None).await;
+    server.wait_task(task.uid()).await.succeeded();
+
+    let (task, _code) = index
+        .add_documents(
+            json!([
+                {"id": 1, "title": "The Dark Knight", "genre": "Action"},
+                {"id": 2, "title": "Inception", "genre": "Sci-Fi"},
+                {"id": 3, "title": "The Matrix", "genre": "Sci-Fi"}
+            ]),
+            None,
+        )
+        .await;
+    server.wait_task(task.uid()).await.succeeded();
+
+    // Try to search with personalization - should return remote_invalid_api_key error
+    let (response, code) = index
+        .search_post(json!({
+            "q": "the",
+            "personalize": {
+                "userContext": "I love science fiction movies"
+            }
+        }))
+        .await;
+
+    snapshot!(code, @"403 Forbidden");
+    snapshot!(json_string!(response), @r#"
+    {
+      "message": "Personalization service: Unauthorized: invalid API key",
+      "code": "remote_invalid_api_key",
+      "type": "auth",
+      "link": "https://docs.meilisearch.com/errors#remote_invalid_api_key"
+    }
+    "#);
+}
+
+#[actix_rt::test]
+async fn search_with_personalization_no_user_context() {
+    // Create a server with a fake personalization API key
+    let dir = TempDir::new().unwrap();
+    let options = Opt {
+        experimental_personalization_api_key: Some("fake-api-key-12345".to_string()),
+        ..default_settings(dir.path())
+    };
+    let server = Server::new_with_options(options).await.unwrap();
+    let index = server.unique_index();
+
+    // Create the index and add some documents
+    let (task, _code) = index.create(None).await;
+    server.wait_task(task.uid()).await.succeeded();
+
+    let (task, _code) = index
+        .add_documents(
+            json!([
+                {"id": 1, "title": "The Dark Knight", "genre": "Action"},
+                {"id": 2, "title": "Inception", "genre": "Sci-Fi"},
+                {"id": 3, "title": "The Matrix", "genre": "Sci-Fi"}
+            ]),
+            None,
+        )
+        .await;
+    server.wait_task(task.uid()).await.succeeded();
+
+    // Try to search with personalization - should return remote_invalid_api_key error
+    let (response, code) = index
+        .search_post(json!({
+            "q": "the",
+            "personalize": {}
+        }))
+        .await;
+
+    snapshot!(code, @"400 Bad Request");
+    snapshot!(json_string!(response), @r###"
+    {
+      "message": "Missing field `userContext` inside `.personalize`",
+      "code": "invalid_search_personalize",
+      "type": "invalid_request",
+      "link": "https://docs.meilisearch.com/errors#invalid_search_personalize"
+    }
+    "###);
+}
--- a/crates/meilisearch/tests/tasks/webhook.rs
+++ b/crates/meilisearch/tests/tasks/webhook.rs
@@ -82,7 +82,7 @@ async fn cli_only() {

    let (webhooks, code) = server.get_webhooks().await;
    snapshot!(code, @"200 OK");
-    snapshot!(webhooks, @r#"
+    snapshot!(webhooks, @r###"
    {
      "results": [
        {
@@ -90,12 +90,12 @@ async fn cli_only() {
          "isEditable": false,
          "url": "https://example-cli.com/",
          "headers": {
-            "Authorization": "Bearer a-secret-token"
+            "Authorization": "Bearer a-XXXX..."
          }
        }
      ]
    }
-    "#);
+    "###);
 }

 #[actix_web::test]
@@ -233,7 +233,7 @@ async fn cli_with_dumps() {

    let (webhooks, code) = server.get_webhooks().await;
    snapshot!(code, @"200 OK");
-    snapshot!(webhooks, @r#"
+    snapshot!(webhooks, @r###"
    {
      "results": [
        {
@@ -241,7 +241,7 @@ async fn cli_with_dumps() {
          "isEditable": false,
          "url": "http://defined-in-test-cli.com/",
          "headers": {
-            "Authorization": "Bearer a-secret-token-defined-in-test-cli"
+            "Authorization": "Bearer a-secXXXXXX..."
          }
        },
        {
@@ -255,7 +255,7 @@ async fn cli_with_dumps() {
          "isEditable": true,
          "url": "https://example.com/hook",
          "headers": {
-            "authorization": "TOKEN"
+            "authorization": "XXX..."
          }
        },
        {
@@ -266,7 +266,7 @@ async fn cli_with_dumps() {
        }
      ]
    }
-    "#);
+    "###);
 }

 #[actix_web::test]
@@ -367,30 +367,30 @@ async fn post_get_delete() {
        }))
        .await;
    snapshot!(code, @"201 Created");
-    snapshot!(json_string!(value, { ".uuid" => "[uuid]" }), @r#"
+    snapshot!(json_string!(value, { ".uuid" => "[uuid]" }), @r###"
    {
      "uuid": "[uuid]",
      "isEditable": true,
      "url": "https://example.com/hook",
      "headers": {
-        "authorization": "TOKEN"
+        "authorization": "XXX..."
      }
    }
-    "#);
+    "###);

    let uuid = value.get("uuid").unwrap().as_str().unwrap();
    let (value, code) = server.get_webhook(uuid).await;
    snapshot!(code, @"200 OK");
-    snapshot!(json_string!(value, { ".uuid" => "[uuid]" }), @r#"
+    snapshot!(json_string!(value, { ".uuid" => "[uuid]" }), @r###"
    {
      "uuid": "[uuid]",
      "isEditable": true,
      "url": "https://example.com/hook",
      "headers": {
-        "authorization": "TOKEN"
+        "authorization": "XXX..."
      }
    }
-    "#);
+    "###);

    let (_value, code) = server.delete_webhook(uuid).await;
    snapshot!(code, @"204 No Content");
@@ -430,31 +430,31 @@ async fn create_and_patch() {
    let (value, code) =
        server.patch_webhook(&uuid, json!({ "headers": { "authorization": "TOKEN" } })).await;
    snapshot!(code, @"200 OK");
-    snapshot!(json_string!(value, { ".uuid" => "[uuid]" }), @r#"
+    snapshot!(json_string!(value, { ".uuid" => "[uuid]" }), @r###"
    {
      "uuid": "[uuid]",
      "isEditable": true,
      "url": "https://example.com/hook",
      "headers": {
-        "authorization": "TOKEN"
+        "authorization": "XXX..."
      }
    }
-    "#);
+    "###);

    let (value, code) =
        server.patch_webhook(&uuid, json!({ "headers": { "authorization2": "TOKEN" } })).await;
    snapshot!(code, @"200 OK");
-    snapshot!(json_string!(value, { ".uuid" => "[uuid]" }), @r#"
+    snapshot!(json_string!(value, { ".uuid" => "[uuid]" }), @r###"
    {
      "uuid": "[uuid]",
      "isEditable": true,
      "url": "https://example.com/hook",
      "headers": {
-        "authorization": "TOKEN",
+        "authorization": "XXX...",
        "authorization2": "TOKEN"
      }
    }
-    "#);
+    "###);

    let (value, code) =
        server.patch_webhook(&uuid, json!({ "headers": { "authorization": null } })).await;
--- a/crates/meilitool/src/upgrade/v1_12.rs
+++ b/crates/meilitool/src/upgrade/v1_12.rs
@@ -68,7 +68,7 @@ fn convert_update_files(db_path: &Path) -> anyhow::Result<()> {

    for uuid in file_store.all_uuids().context("while retrieving uuids from file store")? {
        let uuid = uuid.context("while retrieving uuid from file store")?;
-        let update_file_path = file_store.get_update_path(uuid);
+        let update_file_path = file_store.update_path(uuid);
        let update_file = file_store
            .get_update(uuid)
            .with_context(|| format!("while getting update file for uuid {uuid:?}"))?;
--- a/crates/milli/Cargo.toml
+++ b/crates/milli/Cargo.toml
@@ -18,7 +18,7 @@ bincode = "1.3.3"
 bstr = "1.12.0"
 bytemuck = { version = "1.23.1", features = ["extern_crate_alloc"] }
 byteorder = "1.5.0"
-charabia = { version = "0.9.7", default-features = false }
+charabia = { version = "0.9.8", default-features = false }
 cellulite = "0.3.1-nested-rtxns-2"
 concat-arrays = "0.1.2"
 convert_case = "0.8.0"
@@ -34,7 +34,7 @@ grenad = { version = "0.5.0", default-features = false, features = [
    "rayon",
    "tempfile",
 ] }
-heed = { version = "0.22.1-nested-rtxns", default-features = false, features = [
+heed = { version = "0.22.1-nested-rtxns-6", default-features = false, features = [
    "serde-json",
    "serde-bincode",
 ] }
@@ -120,14 +120,16 @@ twox-hash = { version = "2.1.1", default-features = false, features = [
 ] }
 geo-types = "0.7.16"
 zerometry = "0.3.0"
+bitpacking = "0.9.2"

 [dev-dependencies]
-mimalloc = { version = "0.1.47", default-features = false }
 # fixed version due to format breakages in v1.40
 insta = "=1.39.0"
+mimalloc = { version = "0.1.47", default-features = false }
 maplit = "1.0.2"
 md5 = "0.7.0"
 meili-snap = { path = "../meili-snap" }
+quickcheck = "1.0.3"
 rand = { version = "0.8.5", features = ["small_rng"] }

 [features]
--- a/crates/milli/src/heed_codec/de_bitmap.rs
+++ b/crates/milli/src/heed_codec/de_bitmap.rs
@@ -0,0 +1,299 @@
+use std::io::{self, ErrorKind};
+use std::mem::{size_of, size_of_val};
+
+use bitpacking::{BitPacker, BitPacker1x, BitPacker4x, BitPacker8x};
+use roaring::RoaringBitmap;
+
+/// The magic header for our custom encoding format
+const MAGIC_HEADER: u16 = 36869;
+
+pub struct DeBitmapCodec;
+
+// TODO reintroduce:
+//  - serialized_size?
+//  - serialize_into_vec
+//  - intersection_with_serialized
+//  - merge_into
+//  - merge_deladd_into
+impl DeBitmapCodec {
+    /// Returns the delta-encoded compressed version of the given roaring bitmap.
+    pub fn serialize_into<W: io::Write>(bitmap: &RoaringBitmap, writer: W) -> io::Result<()> {
+        let mut tmp_buffer = Vec::new();
+        Self::serialize_into_with_tmp_buffer(bitmap, writer, &mut tmp_buffer)
+    }
+
+    /// Same as [Self::serialize_into] but accepts a buffer to avoid allocating one.
+    pub fn serialize_into_with_tmp_buffer<W: io::Write>(
+        bitmap: &RoaringBitmap,
+        mut writer: W,
+        tmp_buffer: &mut Vec<u32>,
+    ) -> io::Result<()> {
+        // Insert the magic header
+        writer.write_all(&MAGIC_HEADER.to_ne_bytes())?;
+
+        let bitpacker8x = BitPacker8x::new();
+        let bitpacker4x = BitPacker4x::new();
+        let bitpacker1x = BitPacker1x::new();
+
+        // This temporary buffer is used to store each chunk of decompressed and
+        // compressed and delta-encoded u32s. We need room for the decompressed
+        // u32s coming from the roaring bitmap, the compressed output that can
+        // be as large as the decompressed u32s, and the chunk header.
+        tmp_buffer.resize((BitPacker8x::BLOCK_LEN * 2) + 1, 0u32);
+        let (decompressed, compressed) = tmp_buffer.split_at_mut(BitPacker8x::BLOCK_LEN);
+        let compressed = bytemuck::cast_slice_mut(compressed);
+
+        let mut buffer_index = 0;
+        let mut initial = None;
+        // We initially collect all the integers into a flat buffer of the size
+        // of the largest bitpacker. We encode them with it until we don't have
+        // enough of them...
+        for n in bitmap {
+            decompressed[buffer_index] = n;
+            buffer_index += 1;
+            if buffer_index == BitPacker8x::BLOCK_LEN {
+                let output = encode_with_packer(&bitpacker8x, decompressed, initial, compressed);
+                writer.write_all(output)?;
+                initial = Some(n);
+                buffer_index = 0;
+            }
+        }
+
+        // ...We then switch to a smaller bitpacker to encode the remaining chunks...
+        let decompressed = &decompressed[..buffer_index];
+        let mut chunks = decompressed.chunks_exact(BitPacker4x::BLOCK_LEN);
+        for decompressed in chunks.by_ref() {
+            let output = encode_with_packer(&bitpacker4x, decompressed, initial, compressed);
+            writer.write_all(output)?;
+            initial = decompressed.iter().last().copied();
+        }
+
+        // ...And so on...
+        let decompressed = chunks.remainder();
+        let mut chunks = decompressed.chunks_exact(BitPacker1x::BLOCK_LEN);
+        for decompressed in chunks.by_ref() {
+            let output = encode_with_packer(&bitpacker1x, decompressed, initial, compressed);
+            writer.write_all(output)?;
+            initial = decompressed.iter().last().copied();
+        }
+
+        // ...Until we don't have any small enough bitpacker. We put them raw
+        // at the end of out buffer with a header indicating the matter.
+        let decompressed = chunks.remainder();
+        if !decompressed.is_empty() {
+            let header = encode_chunk_header(BitPackerLevel::None, u32::BITS as u8);
+            // Note: Not convinced about the performance of writing a single
+            //       byte followed by a larger write. However, we will use this
+            //       codec with a BufWriter or directly with a Vec of bytes.
+            writer.write_all(&[header])?;
+            writer.write_all(bytemuck::cast_slice(decompressed))?;
+        }
+
+        Ok(())
+    }
+
+    /// Returns the delta-decoded roaring bitmap from the compressed bytes.
+    pub fn deserialize_from(compressed: &[u8]) -> io::Result<RoaringBitmap> {
+        let mut tmp_buffer = Vec::new();
+        Self::deserialize_from_with_tmp_buffer(compressed, &mut tmp_buffer)
+    }
+
+    /// Same as [Self::deserialize_from] but accepts a buffer to avoid allocating one.
+    pub fn deserialize_from_with_tmp_buffer(
+        input: &[u8],
+        tmp_buffer: &mut Vec<u32>,
+    ) -> io::Result<RoaringBitmap> {
+        let Some((header, mut compressed)) = input.split_at_checked(size_of_val(&MAGIC_HEADER))
+        else {
+            return Err(io::Error::new(ErrorKind::UnexpectedEof, "expecting a two-bytes header"));
+        };
+
+        // Safety: This unwrap cannot happen as the header buffer is the right size
+        let header = u16::from_ne_bytes(header.try_into().unwrap());
+
+        if header != MAGIC_HEADER {
+            return Err(io::Error::other("invalid header value"));
+        }
+
+        let bitpacker8x = BitPacker8x::new();
+        let bitpacker4x = BitPacker4x::new();
+        let bitpacker1x = BitPacker1x::new();
+
+        let mut bitmap = RoaringBitmap::new();
+        tmp_buffer.resize(BitPacker8x::BLOCK_LEN, 0u32);
+        let decompressed = &mut tmp_buffer[..];
+        let mut initial = None;
+
+        while let Some((&chunk_header, encoded)) = compressed.split_first() {
+            let (level, num_bits) = decode_chunk_header(chunk_header);
+            let (bytes_read, decompressed) = match level {
+                BitPackerLevel::None => {
+                    if num_bits != u32::BITS as u8 {
+                        return Err(io::Error::new(
+                            ErrorKind::InvalidData,
+                            "invalid number of bits to encode non-compressed u32s",
+                        ));
+                    }
+
+                    let chunks = encoded.chunks_exact(size_of::<u32>());
+                    if !chunks.remainder().is_empty() {
+                        return Err(io::Error::new(
+                            io::ErrorKind::InvalidData,
+                            "expecting last chunk to be a multiple of the size of an u32",
+                        ));
+                    }
+
+                    let integers = chunks
+                        // safety: This unwrap cannot happen as
+                        //         the size of u32 is set correctly.
+                        .map(|b| b.try_into().unwrap())
+                        .map(u32::from_ne_bytes);
+
+                    bitmap
+                        .append(integers)
+                        .map_err(|e| io::Error::new(ErrorKind::InvalidData, e))?;
+
+                    // This is basically always the last chunk that exists in
+                    // this delta-encoded format as the raw u32s are appended
+                    // when there is not enough of them to fit in a bitpacker.
+                    break;
+                }
+                BitPackerLevel::BitPacker1x => {
+                    decode_with_packer(&bitpacker1x, decompressed, initial, encoded, num_bits)
+                }
+                BitPackerLevel::BitPacker4x => {
+                    decode_with_packer(&bitpacker4x, decompressed, initial, encoded, num_bits)
+                }
+                BitPackerLevel::BitPacker8x => {
+                    decode_with_packer(&bitpacker8x, decompressed, initial, encoded, num_bits)
+                }
+            };
+
+            initial = decompressed.iter().last().copied();
+            // TODO investigate perf
+            // Safety: Bitpackers cannot output unsorter integers when
+            //         used with the compress_strictly_sorted function.
+            bitmap.append(decompressed.iter().copied()).unwrap();
+            // What the delta-decoding read plus the chunk header size
+            compressed = &compressed[bytes_read + 1..];
+        }
+
+        Ok(bitmap)
+    }
+}
+
+/// Takes a strickly sorted list of u32s and outputs delta-encoded
+/// bytes with a chunk header. We expect the output buffer to be
+/// at least BLOCK_LEN + 1.
+fn encode_with_packer<'c, B: BitPackerExt>(
+    bitpacker: &B,
+    decompressed: &[u32],
+    initial: Option<u32>,
+    output: &'c mut [u8],
+) -> &'c [u8] {
+    let num_bits = bitpacker.num_bits_strictly_sorted(initial, decompressed);
+    let compressed_len = B::compressed_block_size(num_bits);
+    let chunk_header = encode_chunk_header(B::level(), num_bits);
+    let buffer = &mut output[..compressed_len + 1];
+    // Safety: The buffer is at least one byte
+    let (header_in_buffer, encoded) = buffer.split_first_mut().unwrap();
+    *header_in_buffer = chunk_header;
+    bitpacker.compress_strictly_sorted(initial, decompressed, encoded, num_bits);
+    buffer
+}
+
+/// Returns the number of bytes read and the decoded unsigned integers.
+fn decode_with_packer<'d, B: BitPacker>(
+    bitpacker: &B,
+    decompressed: &'d mut [u32],
+    initial: Option<u32>,
+    compressed: &[u8],
+    num_bits: u8,
+) -> (usize, &'d [u32]) {
+    let decompressed = &mut decompressed[..B::BLOCK_LEN];
+    let read = bitpacker.decompress_strictly_sorted(initial, compressed, decompressed, num_bits);
+    (read, decompressed)
+}
+
+/// An identifier for the bitpacker to be able
+/// to correctly decode the compressed integers.
+#[derive(Debug, PartialEq, Eq)]
+#[repr(u8)]
+enum BitPackerLevel {
+    /// The remaining bytes are raw little endian encoded u32s.
+    None,
+    /// The remaining bits are encoded using a `BitPacker1x`.
+    BitPacker1x,
+    /// The remaining bits are encoded using a `BitPacker4x`.
+    BitPacker4x,
+    /// The remaining bits are encoded using a `BitPacker8x`.
+    BitPacker8x,
+}
+
+/// Returns the chunk header based on the bitpacker level
+/// and the number of bits to encode the list of integers.
+fn encode_chunk_header(level: BitPackerLevel, num_bits: u8) -> u8 {
+    debug_assert!(num_bits as u32 <= 2_u32.pow(6));
+    let level = level as u8;
+    debug_assert!(level <= 3);
+    num_bits | (level << 6)
+}
+
+/// Decodes the chunk header and output the bitpacker level
+/// and the number of bits to decode the following bytes.
+fn decode_chunk_header(data: u8) -> (BitPackerLevel, u8) {
+    let num_bits = data & 0b00111111;
+    let level = match data >> 6 {
+        0 => BitPackerLevel::None,
+        1 => BitPackerLevel::BitPacker1x,
+        2 => BitPackerLevel::BitPacker4x,
+        3 => BitPackerLevel::BitPacker8x,
+        invalid => panic!("Invalid bitpacker level: {invalid}"),
+    };
+    debug_assert!(num_bits as u32 <= 2_u32.pow(6));
+    (level, num_bits)
+}
+
+/// A simple helper trait to get the BitPackerLevel
+/// and correctly generate the chunk header.
+trait BitPackerExt: BitPacker {
+    /// Returns the level of the bitpacker: an identifier to be
+    /// able to decode the numbers with the right bitpacker.
+    fn level() -> BitPackerLevel;
+}
+
+impl BitPackerExt for BitPacker8x {
+    fn level() -> BitPackerLevel {
+        BitPackerLevel::BitPacker8x
+    }
+}
+
+impl BitPackerExt for BitPacker4x {
+    fn level() -> BitPackerLevel {
+        BitPackerLevel::BitPacker4x
+    }
+}
+
+impl BitPackerExt for BitPacker1x {
+    fn level() -> BitPackerLevel {
+        BitPackerLevel::BitPacker1x
+    }
+}
+
+#[cfg(test)]
+mod tests {
+    use quickcheck::quickcheck;
+    use roaring::RoaringBitmap;
+
+    use crate::heed_codec::de_bitmap::DeBitmapCodec;
+
+    quickcheck! {
+        fn qc_random(xs: Vec<u32>) -> bool {
+            let bitmap = RoaringBitmap::from_iter(xs);
+            let mut compressed = Vec::new();
+            DeBitmapCodec::serialize_into(&bitmap, &mut compressed).unwrap();
+            let decompressed = DeBitmapCodec::deserialize_from(&compressed[..]).unwrap();
+            decompressed == bitmap
+        }
+    }
+}
--- a/crates/milli/src/heed_codec/mod.rs
+++ b/crates/milli/src/heed_codec/mod.rs
@@ -1,6 +1,7 @@
 mod beu16_str_codec;
 mod beu32_str_codec;
 mod byte_slice_ref;
+mod de_bitmap;
 pub mod facet;
 mod field_id_word_count_codec;
 mod fst_set_codec;
@@ -19,6 +20,7 @@ use thiserror::Error;

 pub use self::beu16_str_codec::BEU16StrCodec;
 pub use self::beu32_str_codec::BEU32StrCodec;
+pub use self::de_bitmap::DeBitmapCodec;
 pub use self::field_id_word_count_codec::FieldIdWordCountCodec;
 pub use self::fst_set_codec::FstSetCodec;
 pub use self::obkv_codec::ObkvCodec;
--- a/crates/milli/src/index.rs
+++ b/crates/milli/src/index.rs
@@ -425,6 +425,10 @@ impl Index {
        self.env.info().map_size
    }

+    pub fn try_clone_inner_file(&self) -> heed::Result<File> {
+        self.env.try_clone_inner_file()
+    }
+
    pub fn copy_to_file(&self, file: &mut File, option: CompactionOption) -> Result<()> {
        self.env.copy_to_file(file, option).map_err(Into::into)
    }
--- a/crates/milli/src/update/indexer_config.rs
+++ b/crates/milli/src/update/indexer_config.rs
@@ -1,3 +1,6 @@
+use std::num::NonZeroUsize;
+use std::time::Duration;
+
 use grenad::CompressionType;

 use super::GrenadParameters;
@@ -20,6 +23,7 @@ pub struct IndexerConfig {
    pub experimental_no_edition_2024_for_dumps: bool,
    pub experimental_no_edition_2024_for_prefix_post_processing: bool,
    pub experimental_no_edition_2024_for_facet_post_processing: bool,
+    pub s3_snapshot_options: Option<S3SnapshotOptions>,
 }

 impl IndexerConfig {
@@ -37,6 +41,20 @@ impl IndexerConfig {
    }
 }

+#[derive(Debug, Clone)]
+pub struct S3SnapshotOptions {
+    pub s3_bucket_url: String,
+    pub s3_bucket_region: String,
+    pub s3_bucket_name: String,
+    pub s3_snapshot_prefix: String,
+    pub s3_access_key: String,
+    pub s3_secret_key: String,
+    pub s3_max_in_flight_parts: NonZeroUsize,
+    pub s3_compression_level: u32,
+    pub s3_signature_duration: Duration,
+    pub s3_multipart_part_size: u64,
+}
+
 /// By default use only 1 thread for indexing in tests
 #[cfg(test)]
 pub fn default_thread_pool_and_threads() -> (ThreadPoolNoAbort, Option<usize>) {
@@ -76,6 +94,7 @@ impl Default for IndexerConfig {
            experimental_no_edition_2024_for_dumps: false,
            experimental_no_edition_2024_for_prefix_post_processing: false,
            experimental_no_edition_2024_for_facet_post_processing: false,
+            s3_snapshot_options: None,
        }
    }
 }
--- a/crates/milli/src/update/mod.rs
+++ b/crates/milli/src/update/mod.rs
@@ -5,7 +5,7 @@ pub use self::concurrent_available_ids::ConcurrentAvailableIds;
 pub use self::facet::bulk::FacetsUpdateBulk;
 pub use self::facet::incremental::FacetsUpdateIncrementalInner;
 pub use self::index_documents::{request_threads, *};
-pub use self::indexer_config::{default_thread_pool_and_threads, IndexerConfig};
+pub use self::indexer_config::{default_thread_pool_and_threads, IndexerConfig, S3SnapshotOptions};
 pub use self::new::ChannelCongestion;
 pub use self::settings::{validate_embedding_settings, Setting, Settings};
 pub use self::update_step::UpdateIndexingStep;
--- a/crates/milli/src/update/new/words_prefix_docids.rs
+++ b/crates/milli/src/update/new/words_prefix_docids.rs
@@ -4,7 +4,7 @@ use std::io::{BufReader, BufWriter, Read, Seek, Write};
 use std::iter;

 use hashbrown::HashMap;
-use heed::types::{Bytes, DecodeIgnore};
+use heed::types::{Bytes, DecodeIgnore, Str};
 use heed::{BytesDecode, Database, Error, RoTxn, RwTxn};
 use rayon::iter::{IndexedParallelIterator as _, IntoParallelIterator, ParallelIterator as _};
 use roaring::MultiOps;
@@ -16,22 +16,29 @@ use crate::heed_codec::StrBEU16Codec;
 use crate::update::GrenadParameters;
 use crate::{CboRoaringBitmapCodec, Index, Prefix, Result};

-struct WordPrefixDocids {
+struct WordPrefixDocids<'i> {
+    index: &'i Index,
    database: Database<Bytes, CboRoaringBitmapCodec>,
    prefix_database: Database<Bytes, CboRoaringBitmapCodec>,
    max_memory_by_thread: Option<usize>,
+    /// Do not use an experimental LMDB feature to read uncommitted data in parallel.
+    no_experimental_post_processing: bool,
 }

-impl WordPrefixDocids {
+impl<'i> WordPrefixDocids<'i> {
    fn new(
+        index: &'i Index,
        database: Database<Bytes, CboRoaringBitmapCodec>,
        prefix_database: Database<Bytes, CboRoaringBitmapCodec>,
        grenad_parameters: &GrenadParameters,
-    ) -> WordPrefixDocids {
+    ) -> WordPrefixDocids<'i> {
        WordPrefixDocids {
+            index,
            database,
            prefix_database,
            max_memory_by_thread: grenad_parameters.max_memory_by_thread(),
+            no_experimental_post_processing: grenad_parameters
+                .experimental_no_edition_2024_for_prefix_post_processing,
        }
    }

@@ -42,7 +49,77 @@ impl WordPrefixDocids {
        prefix_to_delete: &BTreeSet<Prefix>,
    ) -> Result<()> {
        delete_prefixes(wtxn, &self.prefix_database, prefix_to_delete)?;
-        self.recompute_modified_prefixes(wtxn, prefix_to_compute)
+        if self.no_experimental_post_processing {
+            self.recompute_modified_prefixes(wtxn, prefix_to_compute)
+        } else {
+            self.recompute_modified_prefixes_no_frozen(wtxn, prefix_to_compute)
+        }
+    }
+
+    #[tracing::instrument(level = "trace", skip_all, target = "indexing::prefix")]
+    fn recompute_modified_prefixes_no_frozen(
+        &self,
+        wtxn: &mut RwTxn,
+        prefix_to_compute: &BTreeSet<Prefix>,
+    ) -> Result<()> {
+        let thread_count = rayon::current_num_threads();
+        let rtxns = iter::repeat_with(|| self.index.env.nested_read_txn(wtxn))
+            .take(thread_count)
+            .collect::<heed::Result<Vec<_>>>()?;
+
+        let outputs = rtxns
+            .into_par_iter()
+            .enumerate()
+            .map(|(thread_id, rtxn)| {
+                // `indexes` represent offsets at which prefixes computations were stored in the `file`.
+                let mut indexes = Vec::new();
+                let mut file = BufWriter::new(spooled_tempfile(
+                    self.max_memory_by_thread.unwrap_or(usize::MAX),
+                ));
+
+                let mut buffer = Vec::new();
+                for (prefix_index, prefix) in prefix_to_compute.iter().enumerate() {
+                    // Is prefix for another thread?
+                    if prefix_index % thread_count != thread_id {
+                        continue;
+                    }
+
+                    let output = self
+                        .database
+                        .prefix_iter(&rtxn, prefix.as_bytes())?
+                        .remap_types::<Str, CboRoaringBitmapCodec>()
+                        .map(|result| result.map(|(_word, bitmap)| bitmap))
+                        .union()?;
+
+                    buffer.clear();
+                    CboRoaringBitmapCodec::serialize_into_vec(&output, &mut buffer);
+                    indexes.push(PrefixEntry { prefix, serialized_length: buffer.len() });
+                    file.write_all(&buffer)?;
+                }
+
+                Ok((indexes, file))
+            })
+            .collect::<Result<Vec<_>>>()?;
+
+        // We iterate over all the collected and serialized bitmaps through
+        // the files and entries to eventually put them in the final database.
+        let mut buffer = Vec::new();
+        for (index, file) in outputs {
+            let mut file = file.into_inner().map_err(|e| e.into_error())?;
+            file.rewind()?;
+            let mut file = BufReader::new(file);
+            for PrefixEntry { prefix, serialized_length } in index {
+                buffer.resize(serialized_length, 0);
+                file.read_exact(&mut buffer)?;
+                self.prefix_database.remap_data_type::<Bytes>().put(
+                    wtxn,
+                    prefix.as_bytes(),
+                    &buffer,
+                )?;
+            }
+        }
+
+        Ok(())
    }

    #[tracing::instrument(level = "trace", skip_all, target = "indexing::prefix")]
@@ -463,6 +540,7 @@ pub fn compute_word_prefix_docids(
    grenad_parameters: &GrenadParameters,
 ) -> Result<()> {
    WordPrefixDocids::new(
+        index,
        index.word_docids.remap_key_type(),
        index.word_prefix_docids.remap_key_type(),
        grenad_parameters,
@@ -479,6 +557,7 @@ pub fn compute_exact_word_prefix_docids(
    grenad_parameters: &GrenadParameters,
 ) -> Result<()> {
    WordPrefixDocids::new(
+        index,
        index.exact_word_docids.remap_key_type(),
        index.exact_word_prefix_docids.remap_key_type(),
        grenad_parameters,
Author	SHA1	Message	Date
Kerollmops	faf430af06	Remove the unwraps, asserts and return actual io errors	2025-11-10 18:37:29 +01:00
Kerollmops	cb7e805967	Move to a two bytes header	2025-11-10 17:56:37 +01:00
Kerollmops	a1250001de	Introduce a first working version of the DeBitmapCodec	2025-11-10 16:56:19 +01:00
Clément Renault	8cbcaeff56	Merge pull request #5981 from meilisearch/charabia-v0.9.8 Update Charabia v0.9.8	2025-11-10 09:45:30 +00:00
ManyTheFish	ce87d5a89e	Update Charabia v0.9.8	2025-11-10 09:33:31 +01:00
Louis Dureuil	a9d6e86077	Merge pull request #5775 from meilisearch/experimental-search-personnalization Experimental search personalization	2025-11-06 18:45:18 +00:00
Clément Renault	346f9efe3a	Merge pull request #5977 from meilisearch/fix-rust-analyzer-false-positive Fix error that rust-analyzer reports because it is compiling all code with the `test` cfg	2025-11-06 18:42:43 +00:00
Louis Dureuil	a987d698c1	Fix error that rust-analyzer reports because it is compiling all code with the `test` cfg	2025-11-06 18:10:59 +01:00
ManyTheFish	dbb45dec1a	ignore flaky test	2025-11-06 17:55:27 +01:00
ManyTheFish	5f69a43846	Early return if time budget is already exceeded	2025-11-06 17:55:27 +01:00
ManyTheFish	fe1e4814fa	Return an error when personalize is used in federated queries	2025-11-06 17:55:27 +01:00
ManyTheFish	c29749741b	Use the time budget instead of defining a deadline outside the scope	2025-11-06 17:55:27 +01:00
ManyTheFish	3e47201365	Fix too many argument clippy warning	2025-11-06 17:55:27 +01:00
ManyTheFish	ec9719f3b1	Fix simple PR comments	2025-11-06 17:55:27 +01:00
ManyTheFish	b2cc9e4db8	remove useless clones	2025-11-06 17:55:27 +01:00
ManyTheFish	56198bae48	Initialize personalization API once	2025-11-06 17:55:27 +01:00
ManyTheFish	888059b2d0	Fix PR comments	2025-11-06 17:55:27 +01:00
ManyTheFish	410f2fc8c3	add some failure tests for personalization	2025-11-06 17:55:01 +01:00
ManyTheFish	54e244d2f3	Return an error when the feature is disabled	2025-11-06 17:55:01 +01:00
ManyTheFish	e0c36972fb	remove deadcode	2025-11-06 17:55:01 +01:00
ManyTheFish	daadcddb5e	Reduce personalization footprint on the codebase	2025-11-06 17:55:01 +01:00
ManyTheFish	7f92dafa02	User context is no more optional	2025-11-06 17:55:01 +01:00
ManyTheFish	cc5d12a368	FIx padding	2025-11-06 17:55:01 +01:00
ManyTheFish	0f98b996b5	Fix PR comments	2025-11-06 17:55:01 +01:00
ManyTheFish	d005ca5bf7	remove irrelevant tests	2025-11-06 17:55:01 +01:00
ManyTheFish	7e65fb1d3e	feat(metrics): add personalization count to metrics endpoint - Add MEILISEARCH_PERSONALIZED_SEARCH_REQUESTS metric to track personalized searches - Increment metric directly in search analytics when personalization is used - Metric automatically exposed in /metrics endpoint for monitoring	2025-11-06 17:55:01 +01:00
ManyTheFish	cdefb3f665	feat(analytics): add personalization tracking to segment analytics - Add total_personalized field to SearchAggregator to track personalization usage - Track when search requests include personalization parameters - Include personalization data in analytics JSON output - Maintain clean personalization service interface	2025-11-06 17:55:01 +01:00
ManyTheFish	a91887221a	refactor(personalization): improve Cohere reranking logic and error handling - Replace and_then() with early return for missing personalization - Simplify reranking by building new hits vector instead of swapping - Add debug logging for reranked indices - Fix potential index out-of-bounds issues in reranking	2025-11-06 17:55:01 +01:00
ManyTheFish	9c66b20a97	refactor: split PersonalizationService into enum with CohereService - Refactor PersonalizationService as enum with Cohere and Uninitialized variants - Create dedicated CohereService struct with rerank_search_results method - Split constructor into cohere() and uninitialized() methods - Move all Cohere logic into CohereService for better separation of concerns - Update tests and lib.rs to use new API - Improve code organization and maintainability	2025-11-06 17:55:01 +01:00
ManyTheFish	a48283527e	feat: refine personalization query by merging with user context - Merge initial query with user context to create a comprehensive prompt - Only skip reranking if both query and user_context are None - Support reranking with query-only, user_context-only, or both - Use 'let else' pattern for cleaner error handling - Add comprehensive tests for different parameter combinations - Improve prompt format for better reranking effectiveness	2025-11-06 17:55:01 +01:00
ManyTheFish	73f78c19b0	refactor: rename personalization API fields and move checks inside service - Rename 'personalization' field to 'personalize' in API - Rename 'userProfile' to 'userContext' in personalization object - Remove 'personalized' boolean field (activation now based on non-null 'personalize') - Move personalization checks inside rerank_search_results function - Use 'let else' pattern for better error handling - Update error types and messages to reflect new field names - Update all search routes and analytics to use new field names	2025-11-06 17:55:01 +01:00
ManyTheFish	34639e346e	feat: add personalization service with EnglishV3-only reranking - Add new personalization module with Cohere integration - Implement rerank_search_results method using EnglishV3 model - Remove fallback logic to EnglishV2 for simplified behavior - Add comprehensive error handling and logging - Include unit tests for service behavior - Update search route to support personalization feature	2025-11-06 17:55:01 +01:00
ManyTheFish	7af2a254d6	feat: add personalization parameters to /search route - Add Personalization struct with personalized boolean and user_profile string - Add personalizationPersonalized and personalizationUserProfile query parameters to SearchQueryGet - Follow same pattern as hybrid parameters (hybridEmbedder, hybridSemanticRatio) - Add validation: personalizationUserProfile requires personalizationPersonalized - Add error codes for personalization parameters - Update analytics and facet search to handle new personalization field - Remove serde dependencies from Personalization struct, use Deserr only	2025-11-06 17:55:01 +01:00
ManyTheFish	0f9d262a1c	feat: add experimental_personalization_api_key feature to RoFeatures - Add MEILI_EXPERIMENTAL_PERSONALIZATION_API_KEY environment variable - Add experimental_personalization_api_key field to Opt struct with CLI and env support - Add experimental_personalization_api_key field to InstanceTogglableFeatures - Store personalization API key in FeatureData for access through IndexScheduler - Add experimental_personalization_api_key() method to IndexScheduler - Update analytics destructuring to include new field - Maintain RoFeatures Copy trait while properly handling Option<String>	2025-11-06 17:55:01 +01:00
Clément Renault	34765b556b	Merge pull request #5948 from meilisearch/new-s3-snapshots Upload snapshot tarballs to S3	2025-11-06 15:10:37 +00:00
Kerollmops	dfb4860578	Make some parameters experimental and link the dedicated issue	2025-11-06 12:43:12 +01:00
Kerollmops	ce62713f02	Always create the update_files directory	2025-11-06 12:43:12 +01:00
Kerollmops	8b5d04d60f	Move the code to append a file to the tarball in a dedicated function	2025-11-06 12:07:08 +01:00
Kerollmops	1b74709b91	Extract more logic into dedicated functions	2025-11-06 12:00:25 +01:00
Kerollmops	a5c0a282c5	Better document safety problems and unwraps	2025-11-06 11:26:45 +01:00
Kerollmops	4fc048ff20	Better handle errors when trying to clone the handles	2025-11-06 11:15:34 +01:00
Kerollmops	375b5600cd	Move the tarball streaming into a dedicated function	2025-11-06 11:11:26 +01:00
Kerollmops	32b997d817	Extracts the streaming closure into an async function	2025-11-06 11:04:16 +01:00
Kerollmops	ff3090e3cc	Remove the slash-path dependency	2025-11-06 10:53:07 +01:00
Kerollmops	6c6645f945	Better seeking-to-the-start explanation	2025-11-06 10:45:47 +01:00
Clément Renault	af6473d999	Improve the comments Co-authored-by: Louis Dureuil <louis@meilisearch.com>	2025-11-06 10:43:14 +01:00
Kerollmops	11851f9701	Fix warnings on Windows	2025-11-05 16:40:33 +01:00
Kerollmops	cc4654eabd	Remove useless clippy flag	2025-11-05 16:38:05 +01:00
Kerollmops	0bb91f4a77	Immediately panic when using S3 snapshot options on Windows	2025-11-05 15:16:43 +01:00
Kerollmops	f9d57f54df	Keep the smae naming behavior as before	2025-11-05 15:16:43 +01:00
Kerollmops	3ef1afc0f1	Introduce some backoff retries	2025-11-05 15:16:43 +01:00
Kerollmops	dbb5abebb6	Support cancelation	2025-11-05 15:16:42 +01:00
Kerollmops	700f33bd39	Clean up the code	2025-11-05 15:16:42 +01:00
Kerollmops	d01bbbccde	Support clean CLI options	2025-11-05 15:16:42 +01:00
Kerollmops	4fc506f267	Remove unused imports/code on Windows	2025-11-05 15:16:42 +01:00
Kerollmops	dc456276e5	Make clippy happy	2025-11-05 15:16:42 +01:00
Kerollmops	b2ea50cb10	Make the compression level configurable	2025-11-05 15:16:42 +01:00
Kerollmops	5074cf92ab	Disable compression entirely to avoid being CPU bound	2025-11-05 15:16:42 +01:00
Clément Renault	a92bc8d192	Improve the way we create the snapshot path	2025-11-05 15:16:42 +01:00
Clément Renault	ee538cf045	Remove useless dependencies	2025-11-05 15:16:42 +01:00
Kerollmops	2b05d63a0f	Make it finaly work but without async on the write side	2025-11-05 15:16:42 +01:00
Kerollmops	104e8918ce	Seeking the tasks/data.mdb file to the begining made the trick	2025-11-05 15:16:42 +01:00
Kerollmops	d6ec4d4f4a	Improve understanding of S3-related errors	2025-11-05 15:16:42 +01:00
Kerollmops	f0e7326b7a	Retrieve the bytesMut only when released	2025-11-05 15:16:42 +01:00
Kerollmops	c8106a0006	Fix minimum part size	2025-11-05 15:16:42 +01:00
Kerollmops	c9ab5bc0b6	Improve error messaging when missing env var	2025-11-05 15:16:42 +01:00
Clément Renault	5e0f15fd43	WIP	2025-11-05 15:16:42 +01:00
Kerollmops	4c30f090c7	WIP Do more tests	2025-11-05 15:16:42 +01:00
Clément Renault	63f247cdda	WIP sending multiparts of 250MiB	2025-11-05 15:16:42 +01:00
Clément Renault	e109fa9529	Rename the update_path function	2025-11-05 15:16:42 +01:00
Clément Renault	76e4ec2168	Geenrate an async tarball	2025-11-05 15:16:42 +01:00
Kerollmops	982babdb74	WIP	2025-11-05 15:16:42 +01:00
Kerollmops	7ae2ae33d9	Make max in flights parts fro upload configurable	2025-11-05 15:16:42 +01:00
Kerollmops	cb0788ae07	Use a good mem advice for uploads	2025-11-05 15:16:42 +01:00
Kerollmops	cb3e5dc234	Move the S3 snapshots to disk into a dedicated method	2025-11-05 15:16:41 +01:00
Clément Renault	59d40a2821	Upload ten parts at a time	2025-11-05 15:16:41 +01:00
Clément Renault	98a678e73d	Use the Bytes crate to send the parts	2025-11-05 15:16:41 +01:00
Clément Renault	70292aae3c	Upload indexes under their uuids	2025-11-05 15:16:41 +01:00
Clément Renault	73521f0069	Initial working S3 uploads to RustFS	2025-11-05 15:16:41 +01:00
Louis Dureuil	4533179604	Pass tokio handle to index-scheduler	2025-11-05 15:16:41 +01:00
Clément Renault	1a21cc1a17	Merge pull request #5959 from meilisearch/parallelize-word-prefix-docids Parallelize the word prefix docids	2025-11-05 10:10:07 +00:00
Clément Renault	d08042f8a7	Merge pull request #5967 from meilisearch/bump-lmdb-version Fix the LMDB fork memory leak	2025-11-05 09:33:59 +00:00
Louis Dureuil	77aadb5f22	Merge pull request #5968 from meilisearch/engprod-2116-privilege-escalation-from-webhook-api-key-to-master-key Redact Authorization header in webhooks	2025-11-05 08:51:07 +00:00
Kerollmops	4fd913f7eb	Use the latest version of heed	2025-11-05 09:42:03 +01:00
Louis Dureuil	8618a4d2ba	document `hide_secret`	2025-11-04 17:03:12 +01:00
Louis Dureuil	e9c5df7993	happy clippy	2025-11-03 17:23:27 +01:00
Louis Dureuil	8a28b3aa77	Update snap	2025-11-03 15:52:35 +01:00
Louis Dureuil	1a0b100ad9	rename webhook to highlight redaction	2025-11-03 15:52:22 +01:00
Louis Dureuil	ff93563f41	Redact webhook authorize header on display	2025-11-03 15:51:56 +01:00
Louis Dureuil	2f25258191	Extract crate::settings::hide_secret as a public function	2025-11-03 15:50:37 +01:00
Clément Renault	2859079c32	Merge pull request #5961 from meilisearch/add-flickr-demo Add Flickr example to README	2025-11-03 14:44:54 +00:00
Clément Renault	74b83d305f	Add Flickr example to README This PR adds the new Flickr demo to the README.	2025-10-29 18:27:36 +01:00
Clément Renault	70f6e4b828	Parallelize the word prefix docids	2025-10-27 17:20:12 +01:00