@199-bio/engram 0.11.1 → 0.13.0

package/README.md

@@ -1,12 +1,16 @@
 # Engram
 
+<p align="center">
+  <img src="logo.png" alt="Engram" width="480" />
+</p>
+
 **Give your AI a perfect memory.**
 
-Every conversation you have with your AI disappears the moment it ends. Names you've mentioned, preferences you've shared, the context of your life—all gone. You find yourself repeating the same information, re-explaining who people are, reminding it of things you've already said.
+Every conversation you have with your AI disappears the moment it ends. Names you've mentioned, preferences you've shared, the context of your life — gone. You repeat yourself. You re-explain who people are. You remind it of things you've already said.
 
-Engram changes that.
+Engram fixes that.
 
-It gives your AI the ability to remember. Not just store text, but truly remember—the way you do. Important things stick. Trivial things fade. And everything connects to everything else.
+It lets your AI remember. Not just store text — actually remember, the way you do. Important things stick. Trivial things fade. And everything connects.
 
 > *An engram is a unit of cognitive information imprinted in a physical substance—the biological basis of memory.*
 
@@ -14,7 +18,7 @@ It gives your AI the ability to remember. Not just store text, but truly remembe
 
 ## How It Works
 
-Tell your AI something once:
+Tell your AI something once. Just once:
 
 > "My colleague Sarah is allergic to shellfish and prefers window seats. She's leading the Q1 product launch."
 
@@ -22,7 +26,7 @@ Weeks later, ask:
 
 > "I'm booking a team lunch and flights for the offsite—what should I know?"
 
-Engram connects the dots. It remembers Sarah, her allergy, her seating preference, and her current workload. Your AI can now actually help—suggesting restaurants without shellfish, booking her a window seat, and noting she might be busy with the launch.
+Engram connects the dots. It remembers Sarah — the allergy, the seating preference, the workload. Your AI can now actually help. It'll suggest restaurants without shellfish and book her a window seat. It'll flag that she's probably swamped with the launch.
 
 This isn't keyword matching. It's understanding.
 
@@ -32,11 +36,11 @@ This isn't keyword matching. It's understanding.
 
 Engram models memory the way your brain does.
 
-**Things fade.** A memory from six months ago that you've never thought about becomes harder to find. But something important—a name, a birthday, a preference—stays accessible even as time passes.
+**Things fade.** A memory from six months ago that you've never revisited becomes harder to find. But something important — a name, a birthday, a preference — stays accessible even as time passes.
 
-**Recall strengthens.** Every time a memory surfaces, it becomes more permanent. The things you think about often become the things you'll never forget.
+**Recall strengthens.** Every time a memory surfaces, it becomes more permanent. The things you think about often are the things you won't forget.
 
-**Everything connects.** People link to places. Places link to events. When you ask about one thing, related things come along for the ride. Ask about Sarah, and her company, her projects, and her preferences all surface together.
+**Everything connects.** People link to places, places to events. Ask about Sarah, and her company, her projects, her preferences all surface together.
 
 ---
 
@@ -48,7 +52,7 @@ Install globally:
 npm install -g @199-bio/engram
 ```
 
-Add to **Claude Desktop** (`~/Library/Application Support/Claude/claude_desktop_config.json`):
+Add to **MCP desktop client** (`~/Library/Application Support/AI/AI_desktop_config.json`):
 
 ```json
 {
@@ -64,10 +68,10 @@ Add to **Claude Desktop** (`~/Library/Application Support/Claude/claude_desktop_
 }
 ```
 
-Or with **Claude Code**:
+Or with **AI coding assistant**:
 
 ```bash
-claude mcp add engram -- npx -y @199-bio/engram
+AI mcp add engram -- npx -y @199-bio/engram
 ```
 
 That's it. Your AI now remembers.
@@ -89,15 +93,15 @@ Just talk naturally. Your AI handles the rest.
 - "When is my anniversary?"
 
 **Build a knowledge graph** of your world:
-- People, places, organizations, and how they connect
+- People, places, organizations — how they connect
 - Observations about each entity
-- Relationships that span your entire life
+- Relationships that span your whole life
 
 ---
 
 ## Privacy
 
-Your memories stay on your machine. Everything is stored locally in `~/.engram/`. The only external call is optional—if you provide an API key, Engram can periodically compress old memories into summaries. But the core functionality works entirely offline.
+Your memories stay on your machine. Everything lives in `~/.engram/`. The only external call is optional — if you provide an API key, Engram can compress old memories into summaries. But core functionality works offline.
 
 ---
 
@@ -110,14 +114,14 @@ Your AI gets these capabilities:
 
 | Tool | Purpose |
 |------|---------|
-| `remember` | Store new information with importance, emotions, and timing |
+| `remember` | Store new information with importance and emotional weight |
 | `recall` | Find relevant memories ranked by relevance and recency |
 | `forget` | Remove a specific memory |
-| `create_entity` | Add a person, place, or concept to the knowledge graph |
+| `create_entity` | Add a person, place, or concept to the graph |
 | `observe` | Record a fact about an entity |
 | `relate` | Connect two entities (e.g., "works at", "married to") |
 | `query_entity` | Get everything known about someone or something |
-| `list_entities` | See all tracked people, places, and things |
+| `list_entities` | See all tracked people and places |
 | `stats` | View memory statistics |
 | `consolidate` | Compress old memories and detect contradictions |
 | `engram_web` | Launch a visual memory browser |
@@ -127,13 +131,13 @@ Your AI gets these capabilities:
 <details>
 <summary><strong>How Search Works</strong></summary>
 
-Engram uses three search methods simultaneously:
+Engram runs three search methods at once:
 
-1. **Keywords** — SQLite FTS5 finds exact matches for names, dates, and phrases
-2. **Meaning** — Neural embeddings find conceptually related content
+1. **Keywords** — SQLite FTS5 finds exact matches for names and phrases
+2. **Meaning** — Jina v5 embeddings find conceptually related content
 3. **Connections** — The knowledge graph expands to related entities
 
-Results are fused together, then adjusted for how recent and important each memory is. Fresh memories surface first. Important memories resist fading.
+Results are merged, then ranked by recency and importance. Fresh memories surface first. Important memories resist fading.
 
 </details>
 
@@ -147,23 +151,23 @@ Retention = e^(-time / stability)
 ```
 
 - **Time** is days since the memory was last accessed
-- **Stability** is memory strength, which increases each time you recall something
+- **Stability** is memory strength, which grows each time you recall something
 
-High-importance and emotionally significant memories decay slower. Frequently accessed memories become essentially permanent.
+High-importance and emotionally weighted memories decay slower. Frequently accessed memories become permanent.
 
 </details>
 
 <details>
 <summary><strong>How Consolidation Works</strong></summary>
 
-With an API key, Engram can compress old memories—like how sleep consolidates your experiences into long-term storage.
+With an API key, Engram compresses old memories — like sleep turning experiences into long-term storage.
 
-1. Groups related low-importance memories together
+1. Groups related low-importance memories
 2. Creates AI-generated summaries (digests)
-3. Detects contradictory information
+3. Flags contradictory information
 4. Archives the originals
 
-This keeps storage efficient while preserving everything important.
+Storage stays lean, but nothing important gets lost.
 
 </details>
 
@@ -177,7 +181,7 @@ engram/
 │   ├── storage/database.ts   # SQLite with temporal fields
 │   ├── graph/knowledge-graph.ts
 │   ├── retrieval/
-│   │   ├── colbert.ts        # Semantic search
+│   │   ├── jina.ts           # Jina v5 semantic search
 │   │   └── hybrid.ts         # Fusion + decay + salience
 │   ├── consolidation/consolidator.ts
 │   └── web/server.ts         # Visual browser
@@ -208,13 +212,13 @@ npm run build
 npm install -g .
 ```
 
-For semantic search, install Python dependencies:
+For semantic search, install the Jina embeddings package:
 
 ```bash
-pip install ragatouille torch
+pip install jina-grep
 ```
 
-If unavailable, Engram falls back to keyword-only search automatically.
+This uses Jina v5 embeddings with MLX Metal acceleration (~9ms/query). If unavailable, Engram falls back to keyword-only search.
 
 </details>
 

package/dist/consolidation/plan.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"plan.d.ts","sourceRoot":"","sources":["../../src/consolidation/plan.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAGH,OAAO,EAAE,cAAc,EAAE,MAAM,EAAE,OAAO,EAAE,uBAAuB,EAAE,MAAM,wBAAwB,CAAC;AAiClG,MAAM,WAAW,iBAAiB;IAChC,sBAAsB,EAAE,MAAM,CAAC;IAC/B,sBAAsB,EAAE,MAAM,CAAC;IAC/B,cAAc,EAAE,OAAO,CAAC;IACxB,gBAAgB,EAAE,MAAM,CAAC;IACzB,aAAa,EAAE,MAAM,CAAC;IACtB,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,CAAC;IACnB,eAAe,EAAE,MAAM,CAAC;IACxB,UAAU,EAAE,OAAO,CAAC;IACpB,kBAAkB,EAAE,MAAM,CAAC;IAC3B,MAAM,EAAE,SAAS,EAAE,CAAC;CACrB;AAED,MAAM,WAAW,SAAS;IACxB,KAAK,EAAE,UAAU,GAAG,UAAU,GAAG,OAAO,GAAG,SAAS,CAAC;IACrD,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,eAAe,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,WAAW,qBAAqB;IACpC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,uBAAuB,CAAC,OAAO,CAAC,CAAC;IACxC,gBAAgB,EAAE,MAAM,CAAC;IACzB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,iBAAiB,EAAE,MAAM,CAAC;IAC1B,cAAc,EAAE,MAAM,CAAC;IACvB,mBAAmB,EAAE,MAAM,CAAC;IAC5B,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,SAAS,EAAE,IAAI,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,YAAY,GAAG,eAAe,GAAG,oBAAoB,GAAG,iBAAiB,CAAC;IAChF,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,OAAO,CAAC;IACnB,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,qBAAa,iBAAiB;IAC5B,OAAO,CAAC,EAAE,CAAiB;IAC3B,OAAO,CAAC,KAAK,CAAS;IACtB,OAAO,CAAC,MAAM,CAAgB;IAC9B,OAAO,CAAC,YAAY,CAAa;IACjC,OAAO,CAAC,YAAY,CAAa;IACjC,OAAO,CAAC,QAAQ,CAAa;IAC7B,OAAO,CAAC,SAAS,CAAa;gBAElB,EAAE,EAAE,cAAc;IAK9B;;OAEG;IACH,aAAa,IAAI,iBAAiB;IA2ElC;;;OAGG;IACH,sBAAsB,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,EAAE;IAsC/C;;OAEG;IACH,sBAAsB,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,EAAE;IAShD;;OAEG;IACH,gBAAgB,CAAC,KAAK,EAAE,uBAAuB,CAAC,OAAO,CAAC,EAAE,YAAY,EAAE,MAAM,GAAG,uBAAuB;IAIxG;;OAEG;IACH,cAAc,CAAC,OAAO,EAAE,OAAO,CAAC;QAC9B,KAAK,EAAE,uBAAuB,CAAC,OAAO,CAAC,CAAC;QACxC,gBAAgB,EAAE,MAAM,CAAC;QACzB,YAAY,EAAE,MAAM,CAAC;QACrB,iBAAiB,EAAE,MAAM,CAAC;QAC1B,iBAAiB,EAAE,MAAM,CAAC;QAC1B,cAAc,EAAE,MAAM,CAAC;QACvB,mBAAmB,EAAE,MAAM,CAAC;QAC5B,UAAU,EAAE,MAAM,CAAC;QACnB,aAAa,EAAE,MAAM,CAAC;KACvB,CAAC,GAAG,IAAI;IAcT;;OAEG;IACH,QAAQ,IAAI,IAAI;IAIhB;;OAEG;IACH,IAAI,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI;IAIzB;;OAEG;IACH,WAAW,IAAI,qBAAqB;IAoBpC;;OAEG;IACH,cAAc,IAAI,uBAAuB,GAAG,IAAI;IAIhD;;OAEG;IACH,UAAU,CAAC,UAAU,EAAE,uBAAuB,GAAG,IAAI;IAIrD;;OAEG;IACH,aAAa,CAAC,OAAO,EAAE,OAAO,EAAE,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI;IAO1D;;OAEG;IACH,YAAY,CAAC,OAAO,EAAE,OAAO,GAAG,IAAI;IAOpC;;OAEG;IACH,WAAW,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI;IAIhC;;OAEG;IACH,qBAAqB,IAAI,eAAe,EAAE;IA2C1C;;OAEG;IACG,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC;IAK5B;;OAEG;IACH,OAAO,CAAC,wBAAwB;IAKhC;;OAEG;IACH,OAAO,CAAC,uBAAuB;IAO/B;;OAEG;IACH,aAAa,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,EAAE,WAAW,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,EAAE,cAAc,CAAC,EAAE,MAAM,GAAG,MAAM;IAWlH;;OAEG;IACH,QAAQ,IAAI,MAAM;CAGnB"}
+{"version":3,"file":"plan.d.ts","sourceRoot":"","sources":["../../src/consolidation/plan.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAGH,OAAO,EAAE,cAAc,EAAE,MAAM,EAAE,OAAO,EAAE,uBAAuB,EAAE,MAAM,wBAAwB,CAAC;AAiClG,MAAM,WAAW,iBAAiB;IAChC,sBAAsB,EAAE,MAAM,CAAC;IAC/B,sBAAsB,EAAE,MAAM,CAAC;IAC/B,cAAc,EAAE,OAAO,CAAC;IACxB,gBAAgB,EAAE,MAAM,CAAC;IACzB,aAAa,EAAE,MAAM,CAAC;IACtB,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,CAAC;IACnB,eAAe,EAAE,MAAM,CAAC;IACxB,UAAU,EAAE,OAAO,CAAC;IACpB,kBAAkB,EAAE,MAAM,CAAC;IAC3B,MAAM,EAAE,SAAS,EAAE,CAAC;CACrB;AAED,MAAM,WAAW,SAAS;IACxB,KAAK,EAAE,UAAU,GAAG,UAAU,GAAG,OAAO,GAAG,SAAS,CAAC;IACrD,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,eAAe,EAAE,MAAM,CAAC;CACzB;AAED,MAAM,WAAW,qBAAqB;IACpC,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,uBAAuB,CAAC,OAAO,CAAC,CAAC;IACxC,gBAAgB,EAAE,MAAM,CAAC;IACzB,YAAY,EAAE,MAAM,CAAC;IACrB,iBAAiB,EAAE,MAAM,CAAC;IAC1B,iBAAiB,EAAE,MAAM,CAAC;IAC1B,cAAc,EAAE,MAAM,CAAC;IACvB,mBAAmB,EAAE,MAAM,CAAC;IAC5B,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;IACtB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,SAAS,EAAE,IAAI,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,YAAY,GAAG,eAAe,GAAG,oBAAoB,GAAG,iBAAiB,CAAC;IAChF,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,OAAO,CAAC;IACnB,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,qBAAa,iBAAiB;IAC5B,OAAO,CAAC,EAAE,CAAiB;IAC3B,OAAO,CAAC,KAAK,CAAS;IACtB,OAAO,CAAC,MAAM,CAAgB;IAC9B,OAAO,CAAC,YAAY,CAAa;IACjC,OAAO,CAAC,YAAY,CAAa;IACjC,OAAO,CAAC,QAAQ,CAAa;IAC7B,OAAO,CAAC,SAAS,CAAa;gBAElB,EAAE,EAAE,cAAc;IAK9B;;OAEG;IACH,aAAa,IAAI,iBAAiB;IA2ElC;;;OAGG;IACH,sBAAsB,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,EAAE;IAsC/C;;OAEG;IACH,sBAAsB,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,EAAE;IAShD;;OAEG;IACH,gBAAgB,CAAC,KAAK,EAAE,uBAAuB,CAAC,OAAO,CAAC,EAAE,YAAY,EAAE,MAAM,GAAG,uBAAuB;IAIxG;;OAEG;IACH,cAAc,CAAC,OAAO,EAAE,OAAO,CAAC;QAC9B,KAAK,EAAE,uBAAuB,CAAC,OAAO,CAAC,CAAC;QACxC,gBAAgB,EAAE,MAAM,CAAC;QACzB,YAAY,EAAE,MAAM,CAAC;QACrB,iBAAiB,EAAE,MAAM,CAAC;QAC1B,iBAAiB,EAAE,MAAM,CAAC;QAC1B,cAAc,EAAE,MAAM,CAAC;QACvB,mBAAmB,EAAE,MAAM,CAAC;QAC5B,UAAU,EAAE,MAAM,CAAC;QACnB,aAAa,EAAE,MAAM,CAAC;KACvB,CAAC,GAAG,IAAI;IAcT;;OAEG;IACH,QAAQ,IAAI,IAAI;IAIhB;;OAEG;IACH,IAAI,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI;IAIzB;;OAEG;IACH,WAAW,IAAI,qBAAqB;IAoBpC;;OAEG;IACH,cAAc,IAAI,uBAAuB,GAAG,IAAI;IAIhD;;OAEG;IACH,UAAU,CAAC,UAAU,EAAE,uBAAuB,GAAG,IAAI;IAIrD;;OAEG;IACH,aAAa,CAAC,OAAO,EAAE,OAAO,EAAE,UAAU,CAAC,EAAE,MAAM,GAAG,IAAI;IAO1D;;OAEG;IACH,YAAY,CAAC,OAAO,EAAE,OAAO,GAAG,IAAI;IAOpC;;OAEG;IACH,WAAW,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI;IAIhC;;OAEG;IACH,qBAAqB,IAAI,eAAe,EAAE;IA2C1C;;OAEG;IACG,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC;IAK5B;;OAEG;IACH,OAAO,CAAC,wBAAwB;IAKhC;;OAEG;IACH,OAAO,CAAC,uBAAuB;IAO/B;;OAEG;IACH,aAAa,CAAC,KAAK,EAAE,MAAM,GAAG,QAAQ,EAAE,WAAW,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,EAAE,cAAc,CAAC,EAAE,MAAM,GAAG,MAAM;IAWnH;;OAEG;IACH,QAAQ,IAAI,MAAM;CAGnB"}

package/dist/index.js

@@ -2,7 +2,7 @@
 /**
  * Engram - High-quality personal memory for AI assistants
  *
- * Local-first MCP server with ColBERT + BM25 hybrid search
+ * Local-first MCP server with Jina v5 + BM25 hybrid search
  * and a lightweight knowledge graph.
  */
 import { Server } from "@modelcontextprotocol/sdk/server/index.js";
@@ -15,7 +15,7 @@ import { getTransportMode, getHttpPort } from "./transport/index.js";
 import { startHttpServer } from "./transport/http.js";
 import { EngramDatabase } from "./storage/database.js";
 import { KnowledgeGraph } from "./graph/knowledge-graph.js";
-import { createRetriever } from "./retrieval/colbert.js";
+import { createRetriever } from "./retrieval/jina.js";
 import { HybridSearch } from "./retrieval/hybrid.js";
 import { EngramWebServer, getRunningServerUrl } from "./web/server.js";
 import { Consolidator } from "./consolidation/consolidator.js";
@@ -81,6 +81,9 @@ function cleanup() {
         if (webServer) {
             webServer.stop();
         }
+        if (retriever && "stop" in retriever) {
+            retriever.stop();
+        }
         if (db) {
             db.close();
         }
@@ -119,13 +122,14 @@ if (getTransportMode() === "stdio") {
 let db;
 let graph;
 let search;
+let retriever;
 let consolidator;
 let webServer = null;
 async function initialize() {
     console.error(`[Engram] Initializing with database at ${DB_FILE}`);
     db = new EngramDatabase(DB_FILE);
     graph = new KnowledgeGraph(db);
-    const retriever = await createRetriever(DB_PATH);
+    retriever = await createRetriever(DB_PATH);
     search = new HybridSearch(db, graph, retriever);
     consolidator = new Consolidator(db, graph, search);
     // Rebuild index with existing memories

package/dist/retrieval/hybrid.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"hybrid.d.ts","sourceRoot":"","sources":["../../src/retrieval/hybrid.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAE,cAAc,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,wBAAwB,CAAC;AACxE,OAAO,EAAE,cAAc,EAAE,MAAM,6BAA6B,CAAC;AAC7D,OAAO,EAAE,gBAAgB,EAAE,eAAe,EAA0B,MAAM,cAAc,CAAC;AAEzF,MAAM,WAAW,kBAAkB;IACjC,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE;QACP,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,QAAQ,CAAC,EAAE,MAAM,CAAC;QAClB,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,SAAS,CAAC,EAAE,MAAM,CAAC;KACpB,CAAC;CACH;AAED,MAAM,WAAW,kBAAkB;IACjC,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB;AAED,MAAM,WAAW,oBAAoB;IACnC,OAAO,EAAE,kBAAkB,EAAE,CAAC;IAC9B,OAAO,EAAE,kBAAkB,EAAE,CAAC;IAC9B,SAAS,EAAE,MAAM,CAAC;IAClB,kBAAkB,EAAE,KAAK,CAAC;QACxB,MAAM,EAAE,MAAM,CAAC;QACf,YAAY,EAAE,MAAM,CAAC;QACrB,QAAQ,EAAE,MAAM,CAAC;KAClB,CAAC,CAAC;CACJ;AAwDD,qBAAa,YAAY;IAIrB,OAAO,CAAC,EAAE;IACV,OAAO,CAAC,KAAK;IACb,OAAO,CAAC,SAAS;IALnB,OAAO,CAAC,SAAS,CAAS;gBAGhB,EAAE,EAAE,cAAc,EAClB,KAAK,EAAE,cAAc,EACrB,SAAS,EAAE,gBAAgB,GAAG,eAAe;IAMvD;;OAEG;IACH,OAAO,CAAC,gBAAgB;IAIxB;;;;;OAKG;IACG,MAAM,CACV,KAAK,EAAE,MAAM,EACb,OAAO,GAAE;QACP,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,YAAY,CAAC,EAAE,OAAO,CAAC;QACvB,kBAAkB,CAAC,EAAE,OAAO,CAAC;QAC7B,gBAAgB,CAAC,EAAE,MAAM,CAAC;QAC1B,qBAAqB,CAAC,EAAE,MAAM,CAAC;QAC/B,UAAU,CAAC,EAAE,MAAM,CAAC;QACpB,cAAc,CAAC,EAAE,MAAM,CAAC;QACxB,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,gBAAgB,CAAC,EAAE,MAAM,CAAC;QAC1B,YAAY,CAAC,EAAE,OAAO,CAAC;KACnB,GACL,OAAO,CAAC,oBAAoB,CAAC;IAiPhC;;;OAGG;IACH,OAAO,CAAC,aAAa;IA2BrB;;;OAGG;IACG,YAAY,CAChB,QAAQ,EAAE,MAAM,EAChB,OAAO,GAAE;QAAE,eAAe,CAAC,EAAE,MAAM,CAAA;KAAO,GACzC,OAAO,CAAC,oBAAoB,CAAC;IAyBhC;;OAEG;YACW,UAAU;IASxB;;OAEG;YACW,cAAc;IAS5B;;;OAGG;IACH,OAAO,CAAC,aAAa;IAQrB;;;OAGG;YACW,WAAW;IA8BzB;;OAEG;IACG,WAAW,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;IAOhD;;OAEG;IACG,YAAY,IAAI,OAAO,CAAC;QAAE,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IAYhD;;OAEG;IACG,eAAe,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;CAGvD"}
+{"version":3,"file":"hybrid.d.ts","sourceRoot":"","sources":["../../src/retrieval/hybrid.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,OAAO,EAAE,cAAc,EAAE,MAAM,EAAE,MAAM,EAAE,MAAM,wBAAwB,CAAC;AACxE,OAAO,EAAE,cAAc,EAAE,MAAM,6BAA6B,CAAC;AAC7D,OAAO,EAAE,aAAa,EAAE,eAAe,EAA0B,MAAM,WAAW,CAAC;AAEnF,MAAM,WAAW,kBAAkB;IACjC,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE;QACP,IAAI,CAAC,EAAE,MAAM,CAAC;QACd,QAAQ,CAAC,EAAE,MAAM,CAAC;QAClB,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,SAAS,CAAC,EAAE,MAAM,CAAC;KACpB,CAAC;CACH;AAED,MAAM,WAAW,kBAAkB;IACjC,MAAM,EAAE,MAAM,CAAC;IACf,KAAK,EAAE,MAAM,CAAC;IACd,YAAY,EAAE,MAAM,EAAE,CAAC;CACxB;AAED,MAAM,WAAW,oBAAoB;IACnC,OAAO,EAAE,kBAAkB,EAAE,CAAC;IAC9B,OAAO,EAAE,kBAAkB,EAAE,CAAC;IAC9B,SAAS,EAAE,MAAM,CAAC;IAClB,kBAAkB,EAAE,KAAK,CAAC;QACxB,MAAM,EAAE,MAAM,CAAC;QACf,YAAY,EAAE,MAAM,CAAC;QACrB,QAAQ,EAAE,MAAM,CAAC;KAClB,CAAC,CAAC;CACJ;AAED;;;;;;GAMG;AACH,wBAAgB,kBAAkB,CAAC,MAAM,EAAE,MAAM,EAAE,GAAG,EAAE,IAAI,GAAG,MAAM,CAepE;AAED;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,MAAM,GAAG,MAAM,CAOxD;AAED;;GAEG;AACH,wBAAgB,WAAW,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,EAAE,GAAG,EAAE,IAAI,GAAG;IAAE,QAAQ,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,CAYjH;AAED,qBAAa,YAAY;IAIrB,OAAO,CAAC,EAAE;IACV,OAAO,CAAC,KAAK;IACb,OAAO,CAAC,SAAS;IALnB,OAAO,CAAC,SAAS,CAAS;gBAGhB,EAAE,EAAE,cAAc,EAClB,KAAK,EAAE,cAAc,EACrB,SAAS,EAAE,aAAa,GAAG,eAAe;IAMpD;;OAEG;IACH,OAAO,CAAC,gBAAgB;IAIxB;;;;;OAKG;IACG,MAAM,CACV,KAAK,EAAE,MAAM,EACb,OAAO,GAAE;QACP,KAAK,CAAC,EAAE,MAAM,CAAC;QACf,YAAY,CAAC,EAAE,OAAO,CAAC;QACvB,kBAAkB,CAAC,EAAE,OAAO,CAAC;QAC7B,gBAAgB,CAAC,EAAE,MAAM,CAAC;QAC1B,qBAAqB,CAAC,EAAE,MAAM,CAAC;QAC/B,UAAU,CAAC,EAAE,MAAM,CAAC;QACpB,cAAc,CAAC,EAAE,MAAM,CAAC;QACxB,WAAW,CAAC,EAAE,MAAM,CAAC;QACrB,gBAAgB,CAAC,EAAE,MAAM,CAAC;QAC1B,YAAY,CAAC,EAAE,OAAO,CAAC;KACnB,GACL,OAAO,CAAC,oBAAoB,CAAC;IAiPhC;;;OAGG;IACH,OAAO,CAAC,aAAa;IA2BrB;;;OAGG;IACG,YAAY,CAChB,QAAQ,EAAE,MAAM,EAChB,OAAO,GAAE;QAAE,eAAe,CAAC,EAAE,MAAM,CAAA;KAAO,GACzC,OAAO,CAAC,oBAAoB,CAAC;IAyBhC;;OAEG;YACW,UAAU;IASxB;;OAEG;YACW,cAAc;IAS5B;;;OAGG;IACH,OAAO,CAAC,aAAa;IAQrB;;;OAGG;YACW,WAAW;IA8BzB;;OAEG;IACG,WAAW,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;IAOhD;;OAEG;IACG,YAAY,IAAI,OAAO,CAAC;QAAE,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IAYhD;;OAEG;IACG,eAAe,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;CAGvD"}

package/dist/retrieval/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/retrieval/index.ts"],"names":[],"mappings":"AAAA,cAAc,cAAc,CAAC;AAC7B,cAAc,aAAa,CAAC"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/retrieval/index.ts"],"names":[],"mappings":"AAAA,cAAc,WAAW,CAAC;AAC1B,cAAc,aAAa,CAAC"}

package/dist/retrieval/jina.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"jina.d.ts","sourceRoot":"","sources":["../../src/retrieval/jina.ts"],"names":[],"mappings":"AAAA;;GAEG;AAaH,MAAM,WAAW,QAAQ;IACvB,EAAE,EAAE,MAAM,CAAC;IACX,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,YAAY;IAC3B,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,MAAM,CAAC;CACjB;AAUD,qBAAa,aAAa;IAaZ,OAAO,CAAC,SAAS;IAZ7B,OAAO,CAAC,OAAO,CAA6B;IAC5C,OAAO,CAAC,QAAQ,CAA0B;IAC1C,OAAO,CAAC,eAAe,CAGR;IACf,OAAO,CAAC,SAAS,CAAK;IACtB,OAAO,CAAC,KAAK,CAAS;IACtB,OAAO,CAAC,YAAY,CAAgB;IACpC,OAAO,CAAC,YAAY,CAA6B;IACjD,OAAO,CAAC,MAAM,CAAM;gBAEA,SAAS,EAAE,MAAM;IAMrC;;OAEG;IACG,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC;IA0C5B,OAAO,CAAC,UAAU;YAuBJ,IAAI;IAuBlB;;OAEG;IACG,KAAK,CAAC,SAAS,EAAE,QAAQ,EAAE,GAAG,OAAO,CAAC;QAAE,OAAO,EAAE,OAAO,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IAYhF;;OAEG;IACG,GAAG,CAAC,SAAS,EAAE,QAAQ,EAAE,GAAG,OAAO,CAAC;QAAE,OAAO,EAAE,OAAO,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IAY9E;;OAEG;IACG,MAAM,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC,GAAE,MAAW,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC;IAUpE;;OAEG;IACG,MAAM,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,QAAQ,EAAE,EAAE,CAAC,GAAE,MAAW,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC;IAW3F;;OAEG;IACG,MAAM,CAAC,GAAG,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC;QAAE,OAAO,EAAE,OAAO,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IAYzE;;OAEG;IACG,IAAI,IAAI,OAAO,CAAC,OAAO,CAAC;IAS9B;;OAEG;IACG,IAAI,IAAI,OAAO,CAAC,IAAI,CAAC;CAa5B;AAED;;;GAGG;AACH,qBAAa,eAAe;IAC1B,OAAO,CAAC,SAAS,CAAoC;IAE/C,KAAK,CAAC,SAAS,EAAE,QAAQ,EAAE,GAAG,OAAO,CAAC;QAAE,OAAO,EAAE,OAAO,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IAO1E,GAAG,CAAC,SAAS,EAAE,QAAQ,EAAE,GAAG,OAAO,CAAC;QAAE,OAAO,EAAE,OAAO,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IAIxE,MAAM,CAAC,KAAK,EAAE,MAAM,EAAE,CAAC,GAAE,MAAW,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC;IAwB9D,MAAM,CAAC,KAAK,EAAE,MAAM,EAAE,SAAS,EAAE,QAAQ,EAAE,EAAE,CAAC,GAAE,MAAW,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC;IAcrF,MAAM,CAAC,GAAG,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC;QAAE,OAAO,EAAE,OAAO,CAAC;QAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;CAS1E;AAED;;GAEG;AACH,wBAAsB,eAAe,CAAC,SAAS,EAAE,MAAM,GAAG,OAAO,CAAC,aAAa,GAAG,eAAe,CAAC,CAejG"}

package/dist/web/server.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/web/server.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAOH,OAAO,EAAE,cAAc,EAAE,MAAM,wBAAwB,CAAC;AACxD,OAAO,EAAE,cAAc,EAAE,MAAM,6BAA6B,CAAC;AAC7D,OAAO,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AAyBtD;;;GAGG;AACH,wBAAgB,mBAAmB,IAAI,MAAM,GAAG,IAAI,CAmBnD;AAED,UAAU,gBAAgB;IACxB,EAAE,EAAE,cAAc,CAAC;IACnB,KAAK,EAAE,cAAc,CAAC;IACtB,MAAM,EAAE,YAAY,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,CAAC;CACf;AAED,qBAAa,eAAe;IAC1B,OAAO,CAAC,MAAM,CAA4B;IAC1C,OAAO,CAAC,EAAE,CAAiB;IAC3B,OAAO,CAAC,KAAK,CAAiB;IAC9B,OAAO,CAAC,MAAM,CAAe;IAC7B,OAAO,CAAC,IAAI,CAAc;IAC1B,OAAO,CAAC,YAAY,CAAe;IACnC,OAAO,CAAC,IAAI,CAAS;gBAET,OAAO,EAAE,gBAAgB;IAa/B,KAAK,IAAI,OAAO,CAAC,MAAM,CAAC;IAwD9B,OAAO,CAAC,aAAa;IAkBrB,OAAO,CAAC,cAAc;IActB,IAAI,IAAI,IAAI;YAQE,aAAa;YA+Bb,SAAS;YAiZT,WAAW;IAgCzB,OAAO,CAAC,SAAS;CAclB"}
+{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/web/server.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAOH,OAAO,EAAE,cAAc,EAAE,MAAM,wBAAwB,CAAC;AACxD,OAAO,EAAE,cAAc,EAAE,MAAM,6BAA6B,CAAC;AAC7D,OAAO,EAAE,YAAY,EAAE,MAAM,wBAAwB,CAAC;AAyBtD;;;GAGG;AACH,wBAAgB,mBAAmB,IAAI,MAAM,GAAG,IAAI,CAmBnD;AAED,UAAU,gBAAgB;IACxB,EAAE,EAAE,cAAc,CAAC;IACnB,KAAK,EAAE,cAAc,CAAC;IACtB,MAAM,EAAE,YAAY,CAAC;IACrB,IAAI,CAAC,EAAE,MAAM,CAAC;CACf;AAED,qBAAa,eAAe;IAC1B,OAAO,CAAC,MAAM,CAA4B;IAC1C,OAAO,CAAC,EAAE,CAAiB;IAC3B,OAAO,CAAC,KAAK,CAAiB;IAC9B,OAAO,CAAC,MAAM,CAAe;IAC7B,OAAO,CAAC,IAAI,CAAc;IAC1B,OAAO,CAAC,YAAY,CAAe;IACnC,OAAO,CAAC,IAAI,CAAS;gBAET,OAAO,EAAE,gBAAgB;IAa/B,KAAK,IAAI,OAAO,CAAC,MAAM,CAAC;IAwD9B,OAAO,CAAC,aAAa;IAkBrB,OAAO,CAAC,cAAc;IActB,IAAI,IAAI,IAAI;YAQE,aAAa;YA+Bb,SAAS;YAofT,WAAW;IAgCzB,OAAO,CAAC,SAAS;CA0BlB"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@199-bio/engram",
-  "version": "0.11.1",
+  "version": "0.13.0",
   "description": "Give Claude a perfect memory. Local-first MCP server with hybrid search.",
   "type": "module",
   "main": "dist/index.js",
@@ -15,7 +15,6 @@
     "start": "node dist/index.js",
     "test": "vitest",
     "test:run": "vitest run",
-    "download-models": "node scripts/download-models.js",
     "lint": "eslint src/",
     "format": "prettier --write src/"
   },
@@ -28,7 +27,7 @@
     "memory",
     "ai",
     "claude",
-    "colbert",
+    "jina",
     "rag",
     "knowledge-graph",
     "semantic-search",

package/src/consolidation/consolidator.ts

@@ -1,7 +1,7 @@
 /**
  * Memory Consolidator
  *
- * Uses Opus 4.5 with extended thinking to consolidate memories into digests
+ * Uses Opus 4.6 with extended thinking to consolidate memories into digests
  * and detect contradictions. Inspired by how the brain consolidates
  * short-term memories into long-term storage during sleep.
  *
@@ -241,7 +241,7 @@ export class Consolidator {
   }
 
   /**
-   * Consolidate a batch of memories using Opus 4.5 with extended thinking
+   * Consolidate a batch of memories using Opus 4.6 with extended thinking
    */
   private async consolidateBatch(
     memories: Memory[]
@@ -273,7 +273,7 @@ Create a detailed digest that preserves all important information. Respond with
 
     try {
       const response = await client.messages.create({
-        model: "claude-opus-4-5-20251101",
+        model: "claude-opus-4-6-20250514",
         max_tokens: 16000,
         temperature: 1, // Required for extended thinking
         thinking: {
@@ -363,7 +363,7 @@ Create a rich, detailed profile. Do not summarize away important nuances. Respon
 
     try {
       const response = await client.messages.create({
-        model: "claude-opus-4-5-20251101",
+        model: "claude-opus-4-6-20250514",
         max_tokens: 16000,
         temperature: 1, // Required for extended thinking
         thinking: {
@@ -560,7 +560,7 @@ Create a rich, detailed profile. Do not summarize away important nuances. Respon
   }
 
   /**
-   * Extract memories from conversation episodes using Haiku (fast, cheap)
+   * Extract memories from conversation episodes using Sonnet 4.6 (fast, affordable)
    */
   private async extractMemoriesFromEpisodes(
     episodes: Episode[]
@@ -583,9 +583,9 @@ Remember: Only extract information worth remembering long-term. Skip transient t
 Respond with JSON only.`;
 
     try {
-      // Use Haiku for speed/cost (no extended thinking needed)
+      // Use Sonnet 4.6 for speed/cost (no extended thinking needed)
       const response = await client.messages.create({
-        model: "claude-haiku-4-5-20251201",
+        model: "claude-sonnet-4-6-20250514",
         max_tokens: 4000,
         messages: [
           {
@@ -781,10 +781,10 @@ Respond with JSON only.`;
           episodesProcessed += sessionEpisodes.length;
           batchIndex++;
 
-          // Estimate tokens (Haiku)
+          // Estimate tokens (Sonnet 4.6)
           const batchTokens = 3000; // Conservative estimate
           totalTokens += batchTokens;
-          totalCost += plan.calculateCost("haiku", 2000, 1000);
+          totalCost += plan.calculateCost("sonnet", 2000, 1000);
 
           plan.updateProgress({
             batchesCompleted: batchIndex,

package/src/consolidation/plan.ts

@@ -14,16 +14,16 @@
 import { randomUUID } from "crypto";
 import { EngramDatabase, Memory, Episode, ConsolidationCheckpoint } from "../storage/database.js";
 
-// Token pricing (Opus 4.5 with extended thinking)
+// Token pricing (Opus 4.6 with extended thinking)
 const PRICING = {
   opus: {
     input: 15 / 1_000_000,      // $15 per 1M input tokens
     output: 75 / 1_000_000,     // $75 per 1M output tokens
     thinking: 15 / 1_000_000,   // $15 per 1M thinking tokens (same as input)
   },
-  haiku: {
-    input: 0.80 / 1_000_000,    // $0.80 per 1M input tokens
-    output: 4.00 / 1_000_000,   // $4.00 per 1M output tokens
+  sonnet: {
+    input: 3.00 / 1_000_000,    // $3.00 per 1M input tokens
+    output: 15.00 / 1_000_000,  // $15.00 per 1M output tokens
   },
 };
 
@@ -130,7 +130,7 @@ export class ConsolidationPlan {
     const phases: PhasePlan[] = [];
     const delayMs = this.db.getConfigNumber("delay_between_calls_ms", 2000);
 
-    // Episode phase (Haiku - cheap)
+    // Episode phase (Sonnet 4.6)
     if (unconsolidatedEp.length >= 4) {
       const batchCount = Math.min(episodeBatches, maxBatchesPerRun);
       const cost = batchCount * this.estimateEpisodeBatchCost();
@@ -139,7 +139,7 @@ export class ConsolidationPlan {
         itemCount: Math.min(unconsolidatedEp.length, batchCount * 20),
         batchCount,
         estimatedCost: cost,
-        estimatedTimeMs: batchCount * (2000 + delayMs), // ~2s per Haiku call + delay
+        estimatedTimeMs: batchCount * (2000 + delayMs), // ~2s per Sonnet call + delay
       });
     }
 
@@ -404,11 +404,11 @@ export class ConsolidationPlan {
   }
 
   /**
-   * Estimate cost for an episode batch (Haiku)
+   * Estimate cost for an episode batch (Sonnet 4.6)
    */
   private estimateEpisodeBatchCost(): number {
     const { input, output } = TOKEN_ESTIMATES.episodeBatch;
-    return (input * PRICING.haiku.input) + (output * PRICING.haiku.output);
+    return (input * PRICING.sonnet.input) + (output * PRICING.sonnet.output);
   }
 
   /**
@@ -424,7 +424,7 @@ export class ConsolidationPlan {
   /**
    * Calculate actual cost from token usage
    */
-  calculateCost(model: "opus" | "haiku", inputTokens: number, outputTokens: number, thinkingTokens?: number): number {
+  calculateCost(model: "opus" | "sonnet", inputTokens: number, outputTokens: number, thinkingTokens?: number): number {
     const pricing = PRICING[model];
     let cost = (inputTokens * pricing.input) + (outputTokens * pricing.output);
 

package/src/index.ts

@@ -2,7 +2,7 @@
 /**
  * Engram - High-quality personal memory for AI assistants
  *
- * Local-first MCP server with ColBERT + BM25 hybrid search
+ * Local-first MCP server with Jina v5 + BM25 hybrid search
  * and a lightweight knowledge graph.
  */
 
@@ -21,7 +21,7 @@ import { startHttpServer } from "./transport/http.js";
 
 import { EngramDatabase } from "./storage/database.js";
 import { KnowledgeGraph } from "./graph/knowledge-graph.js";
-import { createRetriever } from "./retrieval/colbert.js";
+import { createRetriever, JinaRetriever, SimpleRetriever } from "./retrieval/jina.js";
 import { HybridSearch } from "./retrieval/hybrid.js";
 import { EngramWebServer, getRunningServerUrl } from "./web/server.js";
 import { Consolidator } from "./consolidation/consolidator.js";
@@ -91,6 +91,9 @@ function cleanup(): void {
     if (webServer) {
       webServer.stop();
     }
+    if (retriever && "stop" in retriever) {
+      (retriever as JinaRetriever).stop();
+    }
     if (db) {
       db.close();
     }
@@ -135,6 +138,7 @@ if (getTransportMode() === "stdio") {
 let db: EngramDatabase;
 let graph: KnowledgeGraph;
 let search: HybridSearch;
+let retriever: JinaRetriever | SimpleRetriever;
 let consolidator: Consolidator;
 let webServer: EngramWebServer | null = null;
 
@@ -144,7 +148,7 @@ async function initialize(): Promise<void> {
   db = new EngramDatabase(DB_FILE);
   graph = new KnowledgeGraph(db);
 
-  const retriever = await createRetriever(DB_PATH);
+  retriever = await createRetriever(DB_PATH);
   search = new HybridSearch(db, graph, retriever);
   consolidator = new Consolidator(db, graph, search);
 

package/src/retrieval/hybrid.ts

@@ -1,12 +1,12 @@
 /**
  * Hybrid Search with Reciprocal Rank Fusion (RRF)
- * Combines BM25 (keyword) and ColBERT (semantic) search
+ * Combines BM25 (keyword) and Jina v5 (semantic) search
  * Enhanced with temporal decay and salience scoring
  */
 
 import { EngramDatabase, Memory, Digest } from "../storage/database.js";
 import { KnowledgeGraph } from "../graph/knowledge-graph.js";
-import { ColBERTRetriever, SimpleRetriever, SearchResult, Document } from "./colbert.js";
+import { JinaRetriever, SimpleRetriever, SearchResult, Document } from "./jina.js";
 
 export interface HybridSearchResult {
   memory: Memory;
@@ -44,7 +44,7 @@ export interface HybridSearchResponse {
  * Higher stability = slower forgetting
  * Recent access = higher retention
  */
-function calculateRetention(memory: Memory, now: Date): number {
+export function calculateRetention(memory: Memory, now: Date): number {
   // Use last_accessed if available, otherwise timestamp
   const lastActive = memory.last_accessed || memory.timestamp;
   const daysSinceAccess = (now.getTime() - lastActive.getTime()) / (1000 * 60 * 60 * 24);
@@ -65,7 +65,7 @@ function calculateRetention(memory: Memory, now: Date): number {
  * Calculate salience score - how important/memorable is this?
  * Combines emotional weight, importance, and access patterns
  */
-function calculateSalience(memory: Memory): number {
+export function calculateSalience(memory: Memory): number {
   const importance = memory.importance || 0.5;
   const emotionalWeight = memory.emotional_weight || 0.5;
   const accessBonus = Math.min(1, Math.log(1 + (memory.access_count || 0)) / 5);
@@ -77,7 +77,7 @@ function calculateSalience(memory: Memory): number {
 /**
  * Apply temporal and salience adjustments to search results
  */
-function adjustScore(memory: Memory, baseScore: number, now: Date): { adjusted: number; retention: number } {
+export function adjustScore(memory: Memory, baseScore: number, now: Date): { adjusted: number; retention: number } {
   const retention = calculateRetention(memory, now);
   const salience = calculateSalience(memory);
 
@@ -97,7 +97,7 @@ export class HybridSearch {
   constructor(
     private db: EngramDatabase,
     private graph: KnowledgeGraph,
-    private retriever: ColBERTRetriever | SimpleRetriever
+    private retriever: JinaRetriever | SimpleRetriever
   ) {
     // Generate a session ID for this search instance
     this.sessionId = `session_${Date.now()}_${Math.random().toString(36).slice(2, 8)}`;
@@ -113,8 +113,8 @@ export class HybridSearch {
   /**
    * Search using all available methods and fuse results
    *
-   * Strategy: BM25 gets candidates fast, ColBERT reranks for quality
-   * This is both FASTER (fewer ColBERT computations) and BETTER (combines keyword + semantic)
+   * Strategy: BM25 gets candidates fast, Jina reranks for quality
+   * This is both FASTER (fewer Jina computations) and BETTER (combines keyword + semantic)
    */
   async search(
     query: string,
@@ -128,7 +128,7 @@ export class HybridSearch {
       semanticWeight?: number;
       graphWeight?: number;
       connectionWeight?: number;     // Weight for connected memories in RRF
-      useReranking?: boolean;        // Use ColBERT to rerank BM25 results
+      useReranking?: boolean;        // Use Jina to rerank BM25 results
     } = {}
   ): Promise<HybridSearchResponse> {
     const {
@@ -163,7 +163,7 @@ export class HybridSearch {
     // For semantic: either rerank BM25 results (faster+better) or search full index
     let semanticResults: Array<{ id: string; score: number }>;
     if (useReranking && bm25Results.length > 0) {
-      // Rerank BM25 candidates with ColBERT - faster AND better quality
+      // Rerank BM25 candidates with Jina - faster AND better quality
       const docs = bm25Results.map(r => ({ id: r.id, content: this.db.getMemory(r.id)?.content || '' }));
       const reranked = await this.retriever.rerank(query, docs, candidateLimit);
       semanticResults = reranked.map(r => ({ id: r.id, score: r.score }));
@@ -447,7 +447,7 @@ export class HybridSearch {
   }
 
   /**
-   * Semantic search via ColBERT
+   * Semantic search via Jina v5
    */
   private async searchSemantic(query: string, limit: number): Promise<Array<{ id: string; score: number }>> {
     try {

package/src/retrieval/index.ts

@@ -1,2 +1,2 @@
-export * from "./colbert.js";
+export * from "./jina.js";
 export * from "./hybrid.js";