feat: upgrade embedding model to bge-small-en-v1.5

Replaces all-MiniLM-L6-v2 (256 token context, MTEB ~60th) with
bge-small-en-v1.5 (512 token context, MTEB ~30th). Same 384
dimensions — no migration or re-index schema change needed.

Key improvement: chunks up to 512 tokens are now fully embedded
instead of silently truncated at 256. This fixes a quality issue
where the second half of larger chunks was invisible to semantic
search.

Trade-off: model download grows from 23MB to 127MB (one-time).

Users must delete ~/.engraph/models/ and run engraph index --rebuild
after upgrading.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: devwhodevs

CLAUDE.md

@@ -9,7 +9,7 @@ Single binary with 20 modules behind a lib crate:
 - `config.rs` — loads `~/.engraph/config.toml` and `vault.toml`, merges CLI args, provides `data_dir()`
 - `chunker.rs` — smart chunking with break-point scoring algorithm. Finds optimal split points considering headings, code fences, blank lines, and thematic breaks. `split_oversized_chunks()` handles token-aware secondary splitting with overlap
 - `docid.rs` — deterministic 6-char hex IDs for files (SHA-256 of path, truncated). Shown in search results for quick reference
-- `embedder.rs` — downloads and runs `all-MiniLM-L6-v2` ONNX model (384-dim). SHA256-verified on download. Uses `ort` for inference, `tokenizers` for tokenization. Implements `ModelBackend` trait. **Not `Send`** — all embedding is serial
+- `embedder.rs` — downloads and runs `bge-small-en-v1.5` ONNX model (384-dim, 512 token context). SHA256-verified on download. Uses `ort` for inference, `tokenizers` for tokenization. Implements `ModelBackend` trait. **Not `Send`** — all embedding is serial
 - `model.rs` — pluggable `ModelBackend` trait, model registry, and `parse_model_spec()`. Enables future model swapping without changing consumer code
 - `fts.rs` — FTS5 full-text search support. Re-exports `FtsResult` from store. BM25-ranked keyword search
 - `fusion.rs` — Reciprocal Rank Fusion (RRF) engine. Merges semantic + FTS5 + graph results. Supports lane weighting, `--explain` output with per-lane detail

README.md

@@ -20,7 +20,7 @@ Plain vector search treats your notes as isolated documents. But knowledge isn't
 - **MCP server for AI agents** — `engraph serve` exposes 13 tools (search, read, context bundles, note creation) that Claude, Cursor, or any MCP client can call directly.
 - **Real-time sync** — file watcher keeps the index fresh as you edit in Obsidian. No manual re-indexing needed.
 - **Smart write pipeline** — AI agents can create notes with automatic tag resolution, wikilink discovery, and folder placement based on semantic similarity.
-- **Fully local** — ONNX embeddings (`all-MiniLM-L6-v2`, 23MB), SQLite storage, no network required after initial model download.
+- **Fully local** — ONNX embeddings (`bge-small-en-v1.5`, 127MB), SQLite storage, no network required after initial model download.
 
 ## What problem it solves
 
@@ -80,7 +80,7 @@ cargo install --git https://github.com/devwhodevs/engraph
 
 ```bash
 engraph index ~/path/to/vault
-# Downloads embedding model on first run (~23MB)
+# Downloads embedding model on first run (~127MB)
 # Incremental — only re-embeds changed files on subsequent runs
 ```
 

src/embedder.rs

@@ -11,12 +11,11 @@ use tokenizers::Tokenizer;
 use tracing::info;
 
 const MODEL_URL: &str =
-    "https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2/resolve/main/onnx/model.onnx";
+    "https://huggingface.co/BAAI/bge-small-en-v1.5/resolve/main/onnx/model.onnx";
 const TOKENIZER_URL: &str =
-    "https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2/resolve/main/tokenizer.json";
-/// SHA-256 of the ONNX model file. Set to empty string to skip verification
-/// until we can compute the real hash from a download.
-const MODEL_SHA256: &str = "6fd5d72fe4589f189f8ebc006442dbb529bb7ce38f8082112682524616046452";
+    "https://huggingface.co/BAAI/bge-small-en-v1.5/resolve/main/tokenizer.json";
+/// SHA-256 of the ONNX model file.
+const MODEL_SHA256: &str = "828e1496d7fabb79cfa4dcd84fa38625c0d3d21da474a00f08db0f559940cf35";
 pub const EMBEDDING_DIM: usize = 384;
 
 pub struct Embedder {
@@ -175,7 +174,7 @@ impl crate::model::ModelBackend for Embedder {
     }
 
     fn name(&self) -> &str {
-        "onnx:all-MiniLM-L6-v2"
+        "onnx:bge-small-en-v1.5"
     }
 }
 

src/model.rs

@@ -42,12 +42,12 @@ impl Default for ModelRegistry {
     fn default() -> Self {
         Self {
             entries: vec![ModelRegistryEntry {
-                name: "onnx:all-MiniLM-L6-v2".to_string(),
+                name: "onnx:bge-small-en-v1.5".to_string(),
                 format: ModelFormat::Onnx,
-                url: "https://huggingface.co/sentence-transformers/all-MiniLM-L6-v2/resolve/main/onnx/model.onnx".to_string(),
-                sha256: "6fd5d72fe4589f189f8ebc006442dbb529bb7ce38f8082112682524616046452".to_string(),
+                url: "https://huggingface.co/BAAI/bge-small-en-v1.5/resolve/main/onnx/model.onnx".to_string(),
+                sha256: "828e1496d7fabb79cfa4dcd84fa38625c0d3d21da474a00f08db0f559940cf35".to_string(),
                 dim: 384,
-                description: "Lightweight general-purpose sentence embeddings".to_string(),
+                description: "High-quality English embeddings, 512 token context".to_string(),
             }],
         }
     }
@@ -96,16 +96,16 @@ mod tests {
         let registry = ModelRegistry::default();
         assert_eq!(registry.entries.len(), 1);
         let entry = &registry.entries[0];
-        assert_eq!(entry.name, "onnx:all-MiniLM-L6-v2");
+        assert_eq!(entry.name, "onnx:bge-small-en-v1.5");
         assert_eq!(entry.dim, 384);
         assert_eq!(entry.format, ModelFormat::Onnx);
     }
 
     #[test]
     fn test_parse_model_spec_onnx() {
-        let spec = parse_model_spec("onnx:all-MiniLM-L6-v2");
+        let spec = parse_model_spec("onnx:bge-small-en-v1.5");
         assert_eq!(spec.format, ModelFormat::Onnx);
-        assert_eq!(spec.name, "all-MiniLM-L6-v2");
+        assert_eq!(spec.name, "bge-small-en-v1.5");
         assert!(spec.path.is_empty());
     }
 
@@ -128,7 +128,7 @@ mod tests {
     #[test]
     fn test_registry_get_existing() {
         let registry = ModelRegistry::default();
-        let entry = registry.get("onnx:all-MiniLM-L6-v2");
+        let entry = registry.get("onnx:bge-small-en-v1.5");
         assert!(entry.is_some());
         assert_eq!(entry.unwrap().dim, 384);
     }

src/search.rs

@@ -233,7 +233,7 @@ pub fn run_status(json: bool, data_dir: &Path) -> Result<()> {
     // Compute index size on disk (sqlite db file).
     let index_size = std::fs::metadata(&db_path).map(|m| m.len()).unwrap_or(0);
 
-    let model_name = "all-MiniLM-L6-v2";
+    let model_name = "bge-small-en-v1.5";
 
     let output = format_status(&stats, index_size, model_name, json);
     print!("{output}");
@@ -451,15 +451,15 @@ mod tests {
             wikilink_count: None,
             mention_count: None,
         };
-        let output = format_status(&stats, 2_516_582, "all-MiniLM-L6-v2", false);
+        let output = format_status(&stats, 2_516_582, "bge-small-en-v1.5", false);
 
         assert!(output.contains("/path/to/vault"), "missing vault path");
         assert!(output.contains("42"), "missing file count");
         assert!(output.contains("187"), "missing chunk count");
         assert!(output.contains("3"), "missing tombstone count");
         assert!(output.contains("2026-03-19 14:30:00"), "missing last index");
         assert!(output.contains("2.4 MB"), "missing index size");
-        assert!(output.contains("all-MiniLM-L6-v2"), "missing model");
+        assert!(output.contains("bge-small-en-v1.5"), "missing model");
     }
 
     #[test]
@@ -474,7 +474,7 @@ mod tests {
             wikilink_count: None,
             mention_count: None,
         };
-        let output = format_status(&stats, 2_516_582, "all-MiniLM-L6-v2", true);
+        let output = format_status(&stats, 2_516_582, "bge-small-en-v1.5", true);
         let parsed: serde_json::Value = serde_json::from_str(&output).unwrap();
 
         assert_eq!(parsed["vault"], "/path/to/vault");
@@ -483,7 +483,7 @@ mod tests {
         assert_eq!(parsed["tombstones"], 3);
         assert_eq!(parsed["last_indexed"], "2026-03-19 14:30:00");
         assert_eq!(parsed["index_size"], 2_516_582);
-        assert_eq!(parsed["model"], "all-MiniLM-L6-v2");
+        assert_eq!(parsed["model"], "bge-small-en-v1.5");
     }
 
     #[test]

tests/integration.rs

@@ -1,6 +1,6 @@
 //! Integration tests for engraph.
 //!
-//! All tests are `#[ignore]` because they require the ONNX model download (~23MB).
+//! All tests are `#[ignore]` because they require the ONNX model download (~127MB).
 //! Run with: `cargo test --test integration -- --ignored`
 
 use std::path::{Path, PathBuf};