Semantic Index (#10329)

This introduces semantic indexing in Zed based on chunking text from
files in the developer's workspace and creating vector embeddings using
an embedding model. As part of this, we've created an embeddings
provider trait that allows us to work with OpenAI, a local Ollama model,
or a Zed hosted embedding.

The semantic index is built by breaking down text for known
(programming) languages into manageable chunks that are smaller than the
max token size. Each chunk is then fed to a language model to create a
high dimensional vector which is then normalized to a unit vector to
allow fast comparison with other vectors with a simple dot product.
Alongside the vector, we store the path of the file and the range within
the document where the vector was sourced from.

Zed will soon grok contextual similarity across different text snippets,
allowing for natural language search beyond keyword matching. This is
being put together both for human-based search as well as providing
results to Large Language Models to allow them to refine how they help
developers.

Remaining todo:

* [x] Change `provider` to `model` within the zed hosted embeddings
database (as its currently a combo of the provider and the model in one
name)


Release Notes:

- N/A

---------

Co-authored-by: Nathan Sobo <nathan@zed.dev>
Co-authored-by: Antonio Scandurra <me@as-cii.com>
Co-authored-by: Conrad Irwin <conrad@zed.dev>
Co-authored-by: Marshall Bowers <elliott.codes@gmail.com>
Co-authored-by: Antonio <antonio@zed.dev>

Cargo.toml

@@ -73,6 +73,7 @@ members = [
     "crates/task",
     "crates/tasks_ui",
     "crates/search",
+    "crates/semantic_index",
     "crates/semantic_version",
     "crates/settings",
     "crates/snippet",
@@ -253,9 +254,11 @@ derive_more = "0.99.17"
 emojis = "0.6.1"
 env_logger = "0.9"
 futures = "0.3"
+futures-batch = "0.6.1"
 futures-lite = "1.13"
 git2 = { version = "0.15", default-features = false }
 globset = "0.4"
+heed = { git = "https://github.com/meilisearch/heed", rev = "036ac23f73a021894974b9adc815bc95b3e0482a", features = ["read-txn-no-tls"] }
 hex = "0.4.3"
 ignore = "0.4.22"
 indoc = "1"

crates/channel/src/channel_store_tests.rs

@@ -264,7 +264,7 @@ async fn test_channel_messages(cx: &mut TestAppContext) {
     );
 
     assert_eq!(
-        channel.next_event(cx),
+        channel.next_event(cx).await,
         ChannelChatEvent::MessagesUpdated {
             old_range: 2..2,
             new_count: 1,
@@ -317,7 +317,7 @@ async fn test_channel_messages(cx: &mut TestAppContext) {
     );
 
     assert_eq!(
-        channel.next_event(cx),
+        channel.next_event(cx).await,
         ChannelChatEvent::MessagesUpdated {
             old_range: 0..0,
             new_count: 2,

crates/collab/migrations/20240409082755_create_embeddings.sql

@@ -0,0 +1,9 @@
+CREATE TABLE IF NOT EXISTS "embeddings" (
+    "model" TEXT,
+    "digest" BYTEA,
+    "dimensions" FLOAT4[1536],
+    "retrieved_at" TIMESTAMP NOT NULL DEFAULT now(),
+    PRIMARY KEY ("model", "digest")
+);
+
+CREATE INDEX IF NOT EXISTS "idx_retrieved_at_on_embeddings" ON "embeddings" ("retrieved_at");

crates/collab/src/db/queries.rs

@@ -6,6 +6,7 @@ pub mod channels;
 pub mod contacts;
 pub mod contributors;
 pub mod dev_servers;
+pub mod embeddings;
 pub mod extensions;
 pub mod hosted_projects;
 pub mod messages;

crates/collab/src/db/queries/embeddings.rs

@@ -0,0 +1,94 @@
+use super::*;
+use time::Duration;
+use time::OffsetDateTime;
+
+impl Database {
+    pub async fn get_embeddings(
+        &self,
+        model: &str,
+        digests: &[Vec<u8>],
+    ) -> Result<HashMap<Vec<u8>, Vec<f32>>> {
+        self.weak_transaction(|tx| async move {
+            let embeddings = {
+                let mut db_embeddings = embedding::Entity::find()
+                    .filter(
+                        embedding::Column::Model.eq(model).and(
+                            embedding::Column::Digest
+                                .is_in(digests.iter().map(|digest| digest.as_slice())),
+                        ),
+                    )
+                    .stream(&*tx)
+                    .await?;
+
+                let mut embeddings = HashMap::default();
+                while let Some(db_embedding) = db_embeddings.next().await {
+                    let db_embedding = db_embedding?;
+                    embeddings.insert(db_embedding.digest, db_embedding.dimensions);
+                }
+                embeddings
+            };
+
+            if !embeddings.is_empty() {
+                let now = OffsetDateTime::now_utc();
+                let retrieved_at = PrimitiveDateTime::new(now.date(), now.time());
+
+                embedding::Entity::update_many()
+                    .filter(
+                        embedding::Column::Digest
+                            .is_in(embeddings.keys().map(|digest| digest.as_slice())),
+                    )
+                    .col_expr(embedding::Column::RetrievedAt, Expr::value(retrieved_at))
+                    .exec(&*tx)
+                    .await?;
+            }
+
+            Ok(embeddings)
+        })
+        .await
+    }
+
+    pub async fn save_embeddings(
+        &self,
+        model: &str,
+        embeddings: &HashMap<Vec<u8>, Vec<f32>>,
+    ) -> Result<()> {
+        self.weak_transaction(|tx| async move {
+            embedding::Entity::insert_many(embeddings.iter().map(|(digest, dimensions)| {
+                let now_offset_datetime = OffsetDateTime::now_utc();
+                let retrieved_at =
+                    PrimitiveDateTime::new(now_offset_datetime.date(), now_offset_datetime.time());
+
+                embedding::ActiveModel {
+                    model: ActiveValue::set(model.to_string()),
+                    digest: ActiveValue::set(digest.clone()),
+                    dimensions: ActiveValue::set(dimensions.clone()),
+                    retrieved_at: ActiveValue::set(retrieved_at),
+                }
+            }))
+            .on_conflict(
+                OnConflict::columns([embedding::Column::Model, embedding::Column::Digest])
+                    .do_nothing()
+                    .to_owned(),
+            )
+            .exec_without_returning(&*tx)
+            .await?;
+            Ok(())
+        })
+        .await
+    }
+
+    pub async fn purge_old_embeddings(&self) -> Result<()> {
+        self.weak_transaction(|tx| async move {
+            embedding::Entity::delete_many()
+                .filter(
+                    embedding::Column::RetrievedAt
+                        .lte(OffsetDateTime::now_utc() - Duration::days(60)),
+                )
+                .exec(&*tx)
+                .await?;
+
+            Ok(())
+        })
+        .await
+    }
+}

crates/collab/src/db/tables.rs

@@ -11,6 +11,7 @@ pub mod channel_message_mention;
 pub mod contact;
 pub mod contributor;
 pub mod dev_server;
+pub mod embedding;
 pub mod extension;
 pub mod extension_version;
 pub mod feature_flag;

crates/collab/src/db/tables/embedding.rs

@@ -0,0 +1,18 @@
+use sea_orm::entity::prelude::*;
+use time::PrimitiveDateTime;
+
+#[derive(Clone, Debug, PartialEq, DeriveEntityModel)]
+#[sea_orm(table_name = "embeddings")]
+pub struct Model {
+    #[sea_orm(primary_key)]
+    pub model: String,
+    #[sea_orm(primary_key)]
+    pub digest: Vec<u8>,
+    pub dimensions: Vec<f32>,
+    pub retrieved_at: PrimitiveDateTime,
+}
+
+#[derive(Copy, Clone, Debug, EnumIter, DeriveRelation)]
+pub enum Relation {}
+
+impl ActiveModelBehavior for ActiveModel {}

crates/collab/src/db/tests.rs

@@ -2,6 +2,7 @@ mod buffer_tests;
 mod channel_tests;
 mod contributor_tests;
 mod db_tests;
+mod embedding_tests;
 mod extension_tests;
 mod feature_flag_tests;
 mod message_tests;