agent-working-memory 0.7.17 → 0.8.5

package/dist/coordination/stale.js

@@ -3,6 +3,8 @@
 /**
  * Stale agent detection and cleanup for the coordination module.
  */
+import { classifyFailure, MUTATION_HINTS } from './failure-modes.js';
+import { recordFailure as circuitRecordFailure } from './circuit-breaker.js';
 /** Detect agents that haven't checked in within the threshold. */
 export function detectStale(db, thresholdSeconds) {
     return db.prepare(`SELECT id, name, role, status, last_seen,
@@ -11,24 +13,66 @@ export function detectStale(db, thresholdSeconds) {
      WHERE status NOT IN ('dead')
        AND (julianday('now') - julianday(last_seen)) * 86400 > ?`).all(thresholdSeconds);
 }
-/** Clean up stale agents: fail assignments, release locks, mark dead. */
+/** Maximum retry attempts before an assignment is permanently failed. */
+const MAX_RETRY_ATTEMPTS = 3;
+/**
+ * Retry or permanently fail a single assignment.
+ * Called from cleanupStale (stale path) and POST /assignment/:id/fail (voluntary path).
+ *
+ * @returns 'retried' | 'failed'
+ */
+export function retryOrFailAssignment(db, assignmentId, agentId, failureResult, mode) {
+    const row = db.prepare(`SELECT attempt_count, description FROM coord_assignments WHERE id = ?`).get(assignmentId);
+    const attemptCount = row?.attempt_count ?? 0;
+    const failureMode = mode ?? classifyFailure(failureResult);
+    if (attemptCount < MAX_RETRY_ATTEMPTS) {
+        const hint = MUTATION_HINTS[failureMode];
+        const nextAttempt = attemptCount + 1;
+        const hintBlock = `\n\n--- RETRY HINT (attempt ${nextAttempt}) ---\n${hint}`;
+        const newDescription = (row?.description ?? '') + hintBlock;
+        db.prepare(`UPDATE coord_assignments
+       SET status = 'pending', agent_id = NULL, started_at = NULL, result = NULL,
+           attempt_count = ?, last_failure_mode = ?, description = ?
+       WHERE id = ?`).run(nextAttempt, failureMode, newDescription, assignmentId);
+        db.prepare(`INSERT INTO coord_events (agent_id, event_type, detail) VALUES (?, 'assignment_retry', ?)`).run(agentId, `attempt ${nextAttempt}: ${failureMode} — ${failureResult.slice(0, 200)}`);
+        return 'retried';
+    }
+    else {
+        db.prepare(`UPDATE coord_assignments
+       SET status = 'failed', result = ?, last_failure_mode = ?, completed_at = datetime('now')
+       WHERE id = ?`).run(failureResult, failureMode, assignmentId);
+        db.prepare(`INSERT INTO coord_events (agent_id, event_type, detail) VALUES (?, 'assignment_failed', ?)`).run(agentId, `permanently failed after ${attemptCount} attempt(s): ${failureResult.slice(0, 200)}`);
+        return 'failed';
+    }
+}
+/** Clean up stale agents: retry or fail assignments, release locks, mark dead. */
 export function cleanupStale(db, thresholdSeconds) {
     const stale = detectStale(db, thresholdSeconds);
     let cleaned = 0;
     for (const agent of stale) {
-        // Fail active assignments
-        const orphaned = db.prepare(`UPDATE coord_assignments SET status = 'failed', result = 'agent disconnected (stale)', completed_at = datetime('now')
-       WHERE agent_id = ? AND status IN ('assigned', 'in_progress')`).run(agent.id);
-        if (orphaned.changes > 0) {
-            db.prepare(`INSERT INTO coord_events (agent_id, event_type, detail) VALUES (?, 'assignment_failed', ?)`).run(agent.id, `auto-failed ${orphaned.changes} orphaned assignment(s) — agent stale`);
+        // Handle each orphaned assignment with retry logic
+        const orphanedRows = db.prepare(`SELECT id FROM coord_assignments WHERE agent_id = ? AND status IN ('assigned', 'in_progress')`).all(agent.id);
+        let retried = 0;
+        let failed = 0;
+        for (const row of orphanedRows) {
+            const outcome = retryOrFailAssignment(db, row.id, agent.id, 'agent disconnected (stale)');
+            if (outcome === 'retried')
+                retried++;
+            else
+                failed++;
+        }
+        if (orphanedRows.length > 0) {
+            db.prepare(`INSERT INTO coord_events (agent_id, event_type, detail) VALUES (?, 'assignment_failed', ?)`).run(agent.id, `stale cleanup: retried ${retried}, permanently failed ${failed}`);
+            // Record failure in circuit breaker for each orphaned assignment
+            circuitRecordFailure(db, agent.id);
         }
         // Release locks
         const locks = db.prepare(`DELETE FROM coord_locks WHERE agent_id = ?`).run(agent.id);
         // Mark dead
         db.prepare(`UPDATE coord_agents SET status = 'dead', current_task = NULL WHERE id = ?`).run(agent.id);
-        cleaned += orphaned.changes + locks.changes;
-        if (orphaned.changes > 0 || locks.changes > 0) {
-            db.prepare(`INSERT INTO coord_events (agent_id, event_type, detail) VALUES (?, 'stale_cleanup', ?)`).run(agent.id, `failed ${orphaned.changes} assignment(s), released ${locks.changes} lock(s)`);
+        cleaned += orphanedRows.length + locks.changes;
+        if (orphanedRows.length > 0 || locks.changes > 0) {
+            db.prepare(`INSERT INTO coord_events (agent_id, event_type, detail) VALUES (?, 'stale_cleanup', ?)`).run(agent.id, `failed ${orphanedRows.length} assignment(s), released ${locks.changes} lock(s)`);
         }
     }
     return { stale, cleaned };

package/dist/coordination/stale.js.map

@@ -1 +1 @@
-{"version":3,"file":"stale.js","sourceRoot":"","sources":["../../src/coordination/stale.ts"],"names":[],"mappings":"AAAA,gDAAgD;AAChD,sCAAsC;AACtC;;GAEG;AAaH,kEAAkE;AAClE,MAAM,UAAU,WAAW,CAAC,EAAqB,EAAE,gBAAwB;IACzE,OAAO,EAAE,CAAC,OAAO,CACf;;;;iEAI6D,CAC9D,CAAC,GAAG,CAAC,gBAAgB,CAAiB,CAAC;AAC1C,CAAC;AAED,yEAAyE;AACzE,MAAM,UAAU,YAAY,CAAC,EAAqB,EAAE,gBAAwB;IAC1E,MAAM,KAAK,GAAG,WAAW,CAAC,EAAE,EAAE,gBAAgB,CAAC,CAAC;IAChD,IAAI,OAAO,GAAG,CAAC,CAAC;IAEhB,KAAK,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;QAC1B,0BAA0B;QAC1B,MAAM,QAAQ,GAAG,EAAE,CAAC,OAAO,CACzB;oEAC8D,CAC/D,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAEhB,IAAI,QAAQ,CAAC,OAAO,GAAG,CAAC,EAAE,CAAC;YACzB,EAAE,CAAC,OAAO,CACR,4FAA4F,CAC7F,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,EAAE,eAAe,QAAQ,CAAC,OAAO,uCAAuC,CAAC,CAAC;QAC1F,CAAC;QAED,gBAAgB;QAChB,MAAM,KAAK,GAAG,EAAE,CAAC,OAAO,CAAC,4CAA4C,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAErF,YAAY;QACZ,EAAE,CAAC,OAAO,CAAC,2EAA2E,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAEtG,OAAO,IAAI,QAAQ,CAAC,OAAO,GAAG,KAAK,CAAC,OAAO,CAAC;QAE5C,IAAI,QAAQ,CAAC,OAAO,GAAG,CAAC,IAAI,KAAK,CAAC,OAAO,GAAG,CAAC,EAAE,CAAC;YAC9C,EAAE,CAAC,OAAO,CACR,wFAAwF,CACzF,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,EAAE,UAAU,QAAQ,CAAC,OAAO,4BAA4B,KAAK,CAAC,OAAO,UAAU,CAAC,CAAC;QACjG,CAAC;IACH,CAAC;IAED,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC;AAC5B,CAAC;AAED,8GAA8G;AAC9G,MAAM,UAAU,kBAAkB,CAAC,EAAqB;IACtD,MAAM,MAAM,GAAG,EAAE,CAAC,OAAO,CACvB,qGAAqG,CACtG,CAAC,GAAG,EAAE,CAAC;IACR,OAAO,MAAM,CAAC,OAAO,CAAC;AACxB,CAAC;AAED,oEAAoE;AACpE,MAAM,UAAU,eAAe,CAAC,EAAqB,EAAE,WAAW,GAAG,EAAE;IACrE,MAAM,MAAM,GAAG,EAAE,CAAC,OAAO,CACvB,sGAAsG,CACvG,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC;IACnB,OAAO,MAAM,CAAC,OAAO,CAAC;AACxB,CAAC;AAED,wFAAwF;AACxF,MAAM,UAAU,UAAU,CAAC,EAAqB;IAC9C,wDAAwD;IACxD,EAAE,CAAC,OAAO,CAAC,iFAAiF,CAAC,CAAC,GAAG,EAAE,CAAC;IAEpG,MAAM,KAAK,GAAG,EAAE,CAAC,OAAO,CACtB,0DAA0D,CAC3D,CAAC,GAAG,EAAyC,CAAC;IAE/C,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO;IAE/B,KAAK,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;QAC1B,EAAE,CAAC,OAAO,CAAC,2EAA2E,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACtG,EAAE,CAAC,OAAO,CAAC,4CAA4C,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IACzE,CAAC;IAED,OAAO,CAAC,GAAG,CAAC,sCAAsC,KAAK,CAAC,MAAM,yCAAyC,CAAC,CAAC;AAC3G,CAAC"}
+{"version":3,"file":"stale.js","sourceRoot":"","sources":["../../src/coordination/stale.ts"],"names":[],"mappings":"AAAA,gDAAgD;AAChD,sCAAsC;AACtC;;GAEG;AAGH,OAAO,EAAE,eAAe,EAAe,cAAc,EAAE,MAAM,oBAAoB,CAAC;AAClF,OAAO,EAAE,aAAa,IAAI,oBAAoB,EAAE,MAAM,sBAAsB,CAAC;AAW7E,kEAAkE;AAClE,MAAM,UAAU,WAAW,CAAC,EAAqB,EAAE,gBAAwB;IACzE,OAAO,EAAE,CAAC,OAAO,CACf;;;;iEAI6D,CAC9D,CAAC,GAAG,CAAC,gBAAgB,CAAiB,CAAC;AAC1C,CAAC;AAED,yEAAyE;AACzE,MAAM,kBAAkB,GAAG,CAAC,CAAC;AAE7B;;;;;GAKG;AACH,MAAM,UAAU,qBAAqB,CACnC,EAAqB,EACrB,YAAoB,EACpB,OAAe,EACf,aAAqB,EACrB,IAAkB;IAElB,MAAM,GAAG,GAAG,EAAE,CAAC,OAAO,CACpB,uEAAuE,CACxE,CAAC,GAAG,CAAC,YAAY,CAAsE,CAAC;IAEzF,MAAM,YAAY,GAAG,GAAG,EAAE,aAAa,IAAI,CAAC,CAAC;IAC7C,MAAM,WAAW,GAAG,IAAI,IAAI,eAAe,CAAC,aAAa,CAAC,CAAC;IAE3D,IAAI,YAAY,GAAG,kBAAkB,EAAE,CAAC;QACtC,MAAM,IAAI,GAAG,cAAc,CAAC,WAAW,CAAC,CAAC;QACzC,MAAM,WAAW,GAAG,YAAY,GAAG,CAAC,CAAC;QACrC,MAAM,SAAS,GAAG,+BAA+B,WAAW,UAAU,IAAI,EAAE,CAAC;QAC7E,MAAM,cAAc,GAAG,CAAC,GAAG,EAAE,WAAW,IAAI,EAAE,CAAC,GAAG,SAAS,CAAC;QAE5D,EAAE,CAAC,OAAO,CACR;;;oBAGc,CACf,CAAC,GAAG,CAAC,WAAW,EAAE,WAAW,EAAE,cAAc,EAAE,YAAY,CAAC,CAAC;QAE9D,EAAE,CAAC,OAAO,CACR,2FAA2F,CAC5F,CAAC,GAAG,CAAC,OAAO,EAAE,WAAW,WAAW,KAAK,WAAW,MAAM,aAAa,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;QAE1F,OAAO,SAAS,CAAC;IACnB,CAAC;SAAM,CAAC;QACN,EAAE,CAAC,OAAO,CACR;;oBAEc,CACf,CAAC,GAAG,CAAC,aAAa,EAAE,WAAW,EAAE,YAAY,CAAC,CAAC;QAEhD,EAAE,CAAC,OAAO,CACR,4FAA4F,CAC7F,CAAC,GAAG,CAAC,OAAO,EAAE,4BAA4B,YAAY,gBAAgB,aAAa,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC;QAEtG,OAAO,QAAQ,CAAC;IAClB,CAAC;AACH,CAAC;AAED,kFAAkF;AAClF,MAAM,UAAU,YAAY,CAAC,EAAqB,EAAE,gBAAwB;IAC1E,MAAM,KAAK,GAAG,WAAW,CAAC,EAAE,EAAE,gBAAgB,CAAC,CAAC;IAChD,IAAI,OAAO,GAAG,CAAC,CAAC;IAEhB,KAAK,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;QAC1B,mDAAmD;QACnD,MAAM,YAAY,GAAG,EAAE,CAAC,OAAO,CAC7B,+FAA+F,CAChG,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAA0B,CAAC;QAEzC,IAAI,OAAO,GAAG,CAAC,CAAC;QAChB,IAAI,MAAM,GAAG,CAAC,CAAC;QAEf,KAAK,MAAM,GAAG,IAAI,YAAY,EAAE,CAAC;YAC/B,MAAM,OAAO,GAAG,qBAAqB,CAAC,EAAE,EAAE,GAAG,CAAC,EAAE,EAAE,KAAK,CAAC,EAAE,EAAE,4BAA4B,CAAC,CAAC;YAC1F,IAAI,OAAO,KAAK,SAAS;gBAAE,OAAO,EAAE,CAAC;;gBAChC,MAAM,EAAE,CAAC;QAChB,CAAC;QAED,IAAI,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC5B,EAAE,CAAC,OAAO,CACR,4FAA4F,CAC7F,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,EAAE,0BAA0B,OAAO,wBAAwB,MAAM,EAAE,CAAC,CAAC;YAEnF,iEAAiE;YACjE,oBAAoB,CAAC,EAAE,EAAE,KAAK,CAAC,EAAE,CAAC,CAAC;QACrC,CAAC;QAED,gBAAgB;QAChB,MAAM,KAAK,GAAG,EAAE,CAAC,OAAO,CAAC,4CAA4C,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAErF,YAAY;QACZ,EAAE,CAAC,OAAO,CAAC,2EAA2E,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAEtG,OAAO,IAAI,YAAY,CAAC,MAAM,GAAG,KAAK,CAAC,OAAO,CAAC;QAE/C,IAAI,YAAY,CAAC,MAAM,GAAG,CAAC,IAAI,KAAK,CAAC,OAAO,GAAG,CAAC,EAAE,CAAC;YACjD,EAAE,CAAC,OAAO,CACR,wFAAwF,CACzF,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,EAAE,UAAU,YAAY,CAAC,MAAM,4BAA4B,KAAK,CAAC,OAAO,UAAU,CAAC,CAAC;QACpG,CAAC;IACH,CAAC;IAED,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,CAAC;AAC5B,CAAC;AAED,8GAA8G;AAC9G,MAAM,UAAU,kBAAkB,CAAC,EAAqB;IACtD,MAAM,MAAM,GAAG,EAAE,CAAC,OAAO,CACvB,qGAAqG,CACtG,CAAC,GAAG,EAAE,CAAC;IACR,OAAO,MAAM,CAAC,OAAO,CAAC;AACxB,CAAC;AAED,oEAAoE;AACpE,MAAM,UAAU,eAAe,CAAC,EAAqB,EAAE,WAAW,GAAG,EAAE;IACrE,MAAM,MAAM,GAAG,EAAE,CAAC,OAAO,CACvB,sGAAsG,CACvG,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC;IACnB,OAAO,MAAM,CAAC,OAAO,CAAC;AACxB,CAAC;AAED,wFAAwF;AACxF,MAAM,UAAU,UAAU,CAAC,EAAqB;IAC9C,wDAAwD;IACxD,EAAE,CAAC,OAAO,CAAC,iFAAiF,CAAC,CAAC,GAAG,EAAE,CAAC;IAEpG,MAAM,KAAK,GAAG,EAAE,CAAC,OAAO,CACtB,0DAA0D,CAC3D,CAAC,GAAG,EAAyC,CAAC;IAE/C,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO;IAE/B,KAAK,MAAM,KAAK,IAAI,KAAK,EAAE,CAAC;QAC1B,EAAE,CAAC,OAAO,CAAC,2EAA2E,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QACtG,EAAE,CAAC,OAAO,CAAC,4CAA4C,CAAC,CAAC,GAAG,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;IACzE,CAAC;IAED,OAAO,CAAC,GAAG,CAAC,sCAAsC,KAAK,CAAC,MAAM,yCAAyC,CAAC,CAAC;AAC3G,CAAC"}

package/dist/core/embeddings.d.ts

@@ -1,12 +1,15 @@
 /**
- * Embedding Engine — local vector embeddings via transformers.js
+ * Embedding Engine - vector embeddings via the ML worker pool.
  *
- * Default: bge-small-en-v1.5 (384 dimensions, ~90MB, MTEB retrieval-optimized).
- * Better short-text similarity than MiniLM for agent memory concepts.
+ * Default model: bge-small-en-v1.5 (384 dimensions, ~90MB, MTEB retrieval-optimized).
  * Configurable via AWM_EMBED_MODEL env var.
- * Model is downloaded once on first use and cached locally.
  *
- * Singleton pattern — call getEmbedder() to get the shared instance.
+ * AWM 0.8.x: inference dispatches through ml-worker.ts. The worker_threads
+ * path was planned but reverted to in-process because onnxruntime-node's
+ * native bindings store V8 handles that don't cross isolate boundaries
+ * safely — see ml-worker.ts for the full status. The dispatch abstraction
+ * is preserved for a future child_process or HTTP sidecar pool.
+ * `AWM_ML_INPROCESS=1` is honored as a no-op (in-process is now the default).
  *
  * NOTE: Changing the model invalidates existing embeddings.
  * Set AWM_EMBED_MODEL=Xenova/all-MiniLM-L6-v2 for backward compatibility.
@@ -14,21 +17,24 @@
 import { type FeatureExtractionPipeline } from '@huggingface/transformers';
 /**
  * Get or initialize the embedding pipeline (singleton).
- * First call downloads the model (~22MB), subsequent calls are instant.
+ * Kept for backwards compat — returns the in-process pipeline only.
+ * Most consumers should use embed() / embedBatch() which dispatch
+ * to the worker pool by default.
  */
 export declare function getEmbedder(): Promise<FeatureExtractionPipeline>;
 /**
  * Generate an embedding vector for a text string.
- * Returns a normalized float32 array of length DIMENSIONS.
+ * Dispatches to the worker pool (or in-process fallback).
  */
 export declare function embed(text: string): Promise<number[]>;
-/** Get the current embedding model ID (for version tracking in stored embeddings) */
-export declare function getModelId(): string;
 /**
  * Generate embeddings for multiple texts in a batch.
- * More efficient than calling embed() in a loop.
+ * More efficient than calling embed() in a loop — the worker batches the
+ * tokenization + forward pass.
  */
 export declare function embedBatch(texts: string[]): Promise<number[][]>;
+/** Get the current embedding model ID (for version tracking in stored embeddings) */
+export declare function getModelId(): string;
 /**
  * Cosine similarity between two normalized vectors.
  * Since vectors are pre-normalized, this is just the dot product.

package/dist/core/embeddings.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"embeddings.d.ts","sourceRoot":"","sources":["../../src/core/embeddings.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAY,KAAK,yBAAyB,EAAE,MAAM,2BAA2B,CAAC;AASrF;;;GAGG;AACH,wBAAsB,WAAW,IAAI,OAAO,CAAC,yBAAyB,CAAC,CAatE;AAED;;;GAGG;AACH,wBAAsB,KAAK,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC,CAK3D;AAED,qFAAqF;AACrF,wBAAgB,UAAU,IAAI,MAAM,CAEnC;AAED;;;GAGG;AACH,wBAAsB,UAAU,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAWrE;AAED;;;GAGG;AACH,wBAAgB,gBAAgB,CAAC,CAAC,EAAE,MAAM,EAAE,EAAE,CAAC,EAAE,MAAM,EAAE,GAAG,MAAM,CAQjE;AAED,uCAAuC;AACvC,eAAO,MAAM,oBAAoB,QAAa,CAAC"}
+{"version":3,"file":"embeddings.d.ts","sourceRoot":"","sources":["../../src/core/embeddings.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;GAeG;AAEH,OAAO,EAAY,KAAK,yBAAyB,EAAE,MAAM,2BAA2B,CAAC;AAyCrF;;;;;GAKG;AACH,wBAAsB,WAAW,IAAI,OAAO,CAAC,yBAAyB,CAAC,CAEtE;AAED;;;GAGG;AACH,wBAAsB,KAAK,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC,CAG3D;AAED;;;;GAIG;AACH,wBAAsB,UAAU,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAGrE;AAED,qFAAqF;AACrF,wBAAgB,UAAU,IAAI,MAAM,CAEnC;AAED;;;GAGG;AACH,wBAAgB,gBAAgB,CAAC,CAAC,EAAE,MAAM,EAAE,EAAE,CAAC,EAAE,MAAM,EAAE,GAAG,MAAM,CAQjE;AAED,uCAAuC;AACvC,eAAO,MAAM,oBAAoB,QAAa,CAAC"}

package/dist/core/embeddings.js

@@ -1,71 +1,87 @@
 // Copyright 2026 Robert Winter / Complete Ideas
 // SPDX-License-Identifier: Apache-2.0
 /**
- * Embedding Engine — local vector embeddings via transformers.js
+ * Embedding Engine - vector embeddings via the ML worker pool.
  *
- * Default: bge-small-en-v1.5 (384 dimensions, ~90MB, MTEB retrieval-optimized).
- * Better short-text similarity than MiniLM for agent memory concepts.
+ * Default model: bge-small-en-v1.5 (384 dimensions, ~90MB, MTEB retrieval-optimized).
  * Configurable via AWM_EMBED_MODEL env var.
- * Model is downloaded once on first use and cached locally.
  *
- * Singleton pattern — call getEmbedder() to get the shared instance.
+ * AWM 0.8.x: inference dispatches through ml-worker.ts. The worker_threads
+ * path was planned but reverted to in-process because onnxruntime-node's
+ * native bindings store V8 handles that don't cross isolate boundaries
+ * safely — see ml-worker.ts for the full status. The dispatch abstraction
+ * is preserved for a future child_process or HTTP sidecar pool.
+ * `AWM_ML_INPROCESS=1` is honored as a no-op (in-process is now the default).
  *
  * NOTE: Changing the model invalidates existing embeddings.
  * Set AWM_EMBED_MODEL=Xenova/all-MiniLM-L6-v2 for backward compatibility.
  */
 import { pipeline } from '@huggingface/transformers';
+import { dispatchEmbed, registerInProcessHandlers } from './ml-worker.js';
 const MODEL_ID = process.env.AWM_EMBED_MODEL ?? 'Xenova/bge-small-en-v1.5';
 const DIMENSIONS = parseInt(process.env.AWM_EMBED_DIMS ?? '384', 10);
 const POOLING = (process.env.AWM_EMBED_POOLING ?? 'mean');
-let instance = null;
-let initPromise = null;
+// --- In-process fallback (used by tests and crash recovery) ---
+let inProcessInstance = null;
+let inProcessInitPromise = null;
+async function loadInProcess() {
+    if (inProcessInstance)
+        return inProcessInstance;
+    if (inProcessInitPromise)
+        return inProcessInitPromise;
+    inProcessInitPromise = pipeline('feature-extraction', MODEL_ID, { dtype: 'fp32' }).then(pipe => {
+        inProcessInstance = pipe;
+        console.log(`Embedding model loaded in-process: ${MODEL_ID} (${DIMENSIONS}d)`);
+        return pipe;
+    });
+    return inProcessInitPromise;
+}
+async function inProcessEmbed(args) {
+    const { texts, pooling, dimensions } = args;
+    if (texts.length === 0)
+        return [];
+    const embedder = await loadInProcess();
+    const result = await embedder(texts, { pooling, normalize: true });
+    const data = result.data;
+    const vectors = [];
+    for (let i = 0; i < texts.length; i++) {
+        vectors.push(Array.from(data.slice(i * dimensions, (i + 1) * dimensions)));
+    }
+    return vectors;
+}
+// Register the in-process handler with the pool (used in test mode and as fallback)
+registerInProcessHandlers({ embed: inProcessEmbed });
+// --- Public API ---
 /**
  * Get or initialize the embedding pipeline (singleton).
- * First call downloads the model (~22MB), subsequent calls are instant.
+ * Kept for backwards compat — returns the in-process pipeline only.
+ * Most consumers should use embed() / embedBatch() which dispatch
+ * to the worker pool by default.
  */
 export async function getEmbedder() {
-    if (instance)
-        return instance;
-    if (initPromise)
-        return initPromise;
-    initPromise = pipeline('feature-extraction', MODEL_ID, {
-        dtype: 'fp32',
-    }).then(pipe => {
-        instance = pipe;
-        console.log(`Embedding model loaded: ${MODEL_ID} (${DIMENSIONS}d)`);
-        return pipe;
-    });
-    return initPromise;
+    return loadInProcess();
 }
 /**
  * Generate an embedding vector for a text string.
- * Returns a normalized float32 array of length DIMENSIONS.
+ * Dispatches to the worker pool (or in-process fallback).
  */
 export async function embed(text) {
-    const embedder = await getEmbedder();
-    const result = await embedder(text, { pooling: POOLING, normalize: true });
-    // result is a Tensor — extract the data
-    return Array.from(result.data).slice(0, DIMENSIONS);
-}
-/** Get the current embedding model ID (for version tracking in stored embeddings) */
-export function getModelId() {
-    return MODEL_ID;
+    const vectors = await dispatchEmbed({ texts: [text], pooling: POOLING, dimensions: DIMENSIONS });
+    return vectors[0] ?? new Array(DIMENSIONS).fill(0);
 }
 /**
  * Generate embeddings for multiple texts in a batch.
- * More efficient than calling embed() in a loop.
+ * More efficient than calling embed() in a loop — the worker batches the
+ * tokenization + forward pass.
  */
 export async function embedBatch(texts) {
     if (texts.length === 0)
         return [];
-    const embedder = await getEmbedder();
-    const result = await embedder(texts, { pooling: POOLING, normalize: true });
-    const data = result.data;
-    const vectors = [];
-    for (let i = 0; i < texts.length; i++) {
-        vectors.push(Array.from(data.slice(i * DIMENSIONS, (i + 1) * DIMENSIONS)));
-    }
-    return vectors;
+    return dispatchEmbed({ texts, pooling: POOLING, dimensions: DIMENSIONS });
+}
+/** Get the current embedding model ID (for version tracking in stored embeddings) */
+export function getModelId() {
+    return MODEL_ID;
 }
 /**
  * Cosine similarity between two normalized vectors.

package/dist/core/embeddings.js.map

@@ -1 +1 @@
-{"version":3,"file":"embeddings.js","sourceRoot":"","sources":["../../src/core/embeddings.ts"],"names":[],"mappings":"AAAA,gDAAgD;AAChD,sCAAsC;AACtC;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,QAAQ,EAAkC,MAAM,2BAA2B,CAAC;AAErF,MAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,CAAC,eAAe,IAAI,0BAA0B,CAAC;AAC3E,MAAM,UAAU,GAAG,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,KAAK,EAAE,EAAE,CAAC,CAAC;AACrE,MAAM,OAAO,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,iBAAiB,IAAI,MAAM,CAAmB,CAAC;AAE5E,IAAI,QAAQ,GAAqC,IAAI,CAAC;AACtD,IAAI,WAAW,GAA8C,IAAI,CAAC;AAElE;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,WAAW;IAC/B,IAAI,QAAQ;QAAE,OAAO,QAAQ,CAAC;IAC9B,IAAI,WAAW;QAAE,OAAO,WAAW,CAAC;IAEpC,WAAW,GAAG,QAAQ,CAAC,oBAAoB,EAAE,QAAQ,EAAE;QACrD,KAAK,EAAE,MAAM;KACd,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;QACb,QAAQ,GAAG,IAAI,CAAC;QAChB,OAAO,CAAC,GAAG,CAAC,2BAA2B,QAAQ,KAAK,UAAU,IAAI,CAAC,CAAC;QACpE,OAAO,IAAI,CAAC;IACd,CAAC,CAAC,CAAC;IAEH,OAAO,WAAW,CAAC;AACrB,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,KAAK,CAAC,IAAY;IACtC,MAAM,QAAQ,GAAG,MAAM,WAAW,EAAE,CAAC;IACrC,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,IAAI,EAAE,EAAE,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC3E,wCAAwC;IACxC,OAAO,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,IAAoB,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,UAAU,CAAC,CAAC;AACtE,CAAC;AAED,qFAAqF;AACrF,MAAM,UAAU,UAAU;IACxB,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,UAAU,CAAC,KAAe;IAC9C,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAClC,MAAM,QAAQ,GAAG,MAAM,WAAW,EAAE,CAAC;IACrC,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,KAAK,EAAE,EAAE,OAAO,EAAE,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC5E,MAAM,IAAI,GAAG,MAAM,CAAC,IAAoB,CAAC;IAEzC,MAAM,OAAO,GAAe,EAAE,CAAC;IAC/B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACtC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,UAAU,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC;IAC7E,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,gBAAgB,CAAC,CAAW,EAAE,CAAW;IACvD,IAAI,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IACtD,IAAI,GAAG,GAAG,CAAC,CAAC;IACZ,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAClC,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IACrB,CAAC;IACD,kDAAkD;IAClD,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;AACxC,CAAC;AAED,uCAAuC;AACvC,MAAM,CAAC,MAAM,oBAAoB,GAAG,UAAU,CAAC"}
+{"version":3,"file":"embeddings.js","sourceRoot":"","sources":["../../src/core/embeddings.ts"],"names":[],"mappings":"AAAA,gDAAgD;AAChD,sCAAsC;AACtC;;;;;;;;;;;;;;;GAeG;AAEH,OAAO,EAAE,QAAQ,EAAkC,MAAM,2BAA2B,CAAC;AACrF,OAAO,EAAE,aAAa,EAAE,yBAAyB,EAAE,MAAM,gBAAgB,CAAC;AAE1E,MAAM,QAAQ,GAAG,OAAO,CAAC,GAAG,CAAC,eAAe,IAAI,0BAA0B,CAAC;AAC3E,MAAM,UAAU,GAAG,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,cAAc,IAAI,KAAK,EAAE,EAAE,CAAC,CAAC;AACrE,MAAM,OAAO,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,iBAAiB,IAAI,MAAM,CAAmB,CAAC;AAE5E,iEAAiE;AAEjE,IAAI,iBAAiB,GAAqC,IAAI,CAAC;AAC/D,IAAI,oBAAoB,GAA8C,IAAI,CAAC;AAE3E,KAAK,UAAU,aAAa;IAC1B,IAAI,iBAAiB;QAAE,OAAO,iBAAiB,CAAC;IAChD,IAAI,oBAAoB;QAAE,OAAO,oBAAoB,CAAC;IACtD,oBAAoB,GAAG,QAAQ,CAAC,oBAAoB,EAAE,QAAQ,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE;QAC7F,iBAAiB,GAAG,IAAI,CAAC;QACzB,OAAO,CAAC,GAAG,CAAC,sCAAsC,QAAQ,KAAK,UAAU,IAAI,CAAC,CAAC;QAC/E,OAAO,IAAI,CAAC;IACd,CAAC,CAAC,CAAC;IACH,OAAO,oBAAoB,CAAC;AAC9B,CAAC;AAED,KAAK,UAAU,cAAc,CAAC,IAAsE;IAClG,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,UAAU,EAAE,GAAG,IAAI,CAAC;IAC5C,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAClC,MAAM,QAAQ,GAAG,MAAM,aAAa,EAAE,CAAC;IACvC,MAAM,MAAM,GAAG,MAAM,QAAQ,CAAC,KAAK,EAAE,EAAE,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IACnE,MAAM,IAAI,GAAG,MAAM,CAAC,IAAoB,CAAC;IACzC,MAAM,OAAO,GAAe,EAAE,CAAC;IAC/B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACtC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,UAAU,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC;IAC7E,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,oFAAoF;AACpF,yBAAyB,CAAC,EAAE,KAAK,EAAE,cAAc,EAAE,CAAC,CAAC;AAErD,qBAAqB;AAErB;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,WAAW;IAC/B,OAAO,aAAa,EAAE,CAAC;AACzB,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,KAAK,CAAC,IAAY;IACtC,MAAM,OAAO,GAAG,MAAM,aAAa,CAAC,EAAE,KAAK,EAAE,CAAC,IAAI,CAAC,EAAE,OAAO,EAAE,OAAO,EAAE,UAAU,EAAE,UAAU,EAAE,CAAC,CAAC;IACjG,OAAO,OAAO,CAAC,CAAC,CAAC,IAAI,IAAI,KAAK,CAAC,UAAU,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AACrD,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,UAAU,CAAC,KAAe;IAC9C,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAClC,OAAO,aAAa,CAAC,EAAE,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,UAAU,EAAE,UAAU,EAAE,CAAC,CAAC;AAC5E,CAAC;AAED,qFAAqF;AACrF,MAAM,UAAU,UAAU;IACxB,OAAO,QAAQ,CAAC;AAClB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,gBAAgB,CAAC,CAAW,EAAE,CAAW;IACvD,IAAI,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IACtD,IAAI,GAAG,GAAG,CAAC,CAAC;IACZ,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAClC,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IACrB,CAAC;IACD,kDAAkD;IAClD,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC;AACxC,CAAC;AAED,uCAAuC;AACvC,MAAM,CAAC,MAAM,oBAAoB,GAAG,UAAU,CAAC"}

package/dist/core/ml-worker-entry.d.ts

@@ -0,0 +1,17 @@
+/**
+ * ML worker entry — runs INSIDE a worker_thread.
+ *
+ * Loaded once per worker, sets up the model for the assigned role
+ * (embed | rerank | expand), then handles request messages from the
+ * main thread via the parentPort.
+ *
+ * Protocol:
+ *   Main thread → worker:  { id, op, args }    (op matches the worker's role)
+ *   Worker → main thread:  { id, ok: true, result } or { id, ok: false, error }
+ *   Worker → main thread:  { ready: true }     (one-time signal after model load)
+ *   Main thread → worker:  { shutdown: true }  (drain queue, then terminate)
+ *
+ * The worker stays loaded — the model lives in memory for the worker's lifetime.
+ */
+export {};
+//# sourceMappingURL=ml-worker-entry.d.ts.map

package/dist/core/ml-worker-entry.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"ml-worker-entry.d.ts","sourceRoot":"","sources":["../../src/core/ml-worker-entry.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;GAcG"}

package/dist/core/ml-worker-entry.js

@@ -0,0 +1,193 @@
+// Copyright 2026 Robert Winter / Complete Ideas
+// SPDX-License-Identifier: Apache-2.0
+/**
+ * ML worker entry — runs INSIDE a worker_thread.
+ *
+ * Loaded once per worker, sets up the model for the assigned role
+ * (embed | rerank | expand), then handles request messages from the
+ * main thread via the parentPort.
+ *
+ * Protocol:
+ *   Main thread → worker:  { id, op, args }    (op matches the worker's role)
+ *   Worker → main thread:  { id, ok: true, result } or { id, ok: false, error }
+ *   Worker → main thread:  { ready: true }     (one-time signal after model load)
+ *   Main thread → worker:  { shutdown: true }  (drain queue, then terminate)
+ *
+ * The worker stays loaded — the model lives in memory for the worker's lifetime.
+ */
+import { parentPort, workerData } from 'node:worker_threads';
+if (!parentPort) {
+    throw new Error('ml-worker-entry: must be loaded as a worker_thread');
+}
+const role = workerData?.role;
+if (role !== 'embed' && role !== 'rerank' && role !== 'expand') {
+    throw new Error(`ml-worker-entry: invalid role '${role}'`);
+}
+// --- Lazy model loaders (each worker loads only its own model) ---
+let embedderPipeline = null;
+let rerankerTokenizer = null;
+let rerankerModel = null;
+let expanderPipeline = null;
+// Inside worker_threads we must use the WASM ONNX backend, not the native one.
+// onnxruntime-node's native bindings store V8 handles that get invalidated when
+// crossing isolate boundaries — calling from a worker crashes with
+// `v8::HandleScope::CreateHandle()` failures. The WASM backend is V8-safe.
+const WORKER_DEVICE = 'wasm';
+async function loadEmbedder() {
+    const { pipeline } = await import('@huggingface/transformers');
+    const modelId = process.env.AWM_EMBED_MODEL ?? 'Xenova/bge-small-en-v1.5';
+    embedderPipeline = await pipeline('feature-extraction', modelId, { dtype: 'fp32', device: WORKER_DEVICE });
+}
+async function loadReranker() {
+    const { AutoTokenizer, AutoModelForSequenceClassification } = await import('@huggingface/transformers');
+    const modelId = process.env.AWM_RERANKER_MODEL || 'Xenova/ms-marco-MiniLM-L-6-v2';
+    rerankerTokenizer = await AutoTokenizer.from_pretrained(modelId);
+    rerankerModel = await AutoModelForSequenceClassification.from_pretrained(modelId, { dtype: 'fp32', device: WORKER_DEVICE });
+}
+async function loadExpander() {
+    const { pipeline } = await import('@huggingface/transformers');
+    expanderPipeline = await pipeline('text2text-generation', 'Xenova/flan-t5-small', { dtype: 'fp32', device: WORKER_DEVICE });
+}
+// --- Per-role inference handlers ---
+function sigmoid(x) {
+    return 1 / (1 + Math.exp(-x));
+}
+async function handleEmbed(args) {
+    if (!embedderPipeline)
+        throw new Error('embedder not loaded');
+    const { texts, pooling, dimensions } = args;
+    if (texts.length === 0)
+        return [];
+    const result = await embedderPipeline(texts, { pooling, normalize: true });
+    const data = result.data;
+    const vectors = [];
+    for (let i = 0; i < texts.length; i++) {
+        vectors.push(Array.from(data.slice(i * dimensions, (i + 1) * dimensions)));
+    }
+    return vectors;
+}
+async function handleRerank(args) {
+    if (!rerankerTokenizer || !rerankerModel)
+        throw new Error('reranker not loaded');
+    const { query, passages } = args;
+    if (passages.length === 0)
+        return [];
+    // Batch path
+    try {
+        const queries = passages.map(() => query);
+        const inputs = rerankerTokenizer(queries, {
+            text_pair: passages,
+            padding: true,
+            truncation: true,
+            return_tensors: 'pt',
+        });
+        const output = await rerankerModel(inputs);
+        const logits = output.logits ?? output.last_hidden_state;
+        const data = logits.data;
+        const results = [];
+        for (let i = 0; i < passages.length; i++) {
+            const rawLogit = Number(data[i] ?? 0);
+            results.push({ index: i, score: sigmoid(rawLogit) });
+        }
+        results.sort((a, b) => b.score - a.score);
+        return results;
+    }
+    catch {
+        // Per-passage fallback
+        const results = [];
+        for (let i = 0; i < passages.length; i++) {
+            try {
+                const inputs = rerankerTokenizer(query, {
+                    text_pair: passages[i],
+                    padding: true,
+                    truncation: true,
+                    return_tensors: 'pt',
+                });
+                const output = await rerankerModel(inputs);
+                const logits = output.logits ?? output.last_hidden_state;
+                const rawLogit = logits.data[0];
+                results.push({ index: i, score: sigmoid(rawLogit) });
+            }
+            catch {
+                results.push({ index: i, score: 0 });
+            }
+        }
+        results.sort((a, b) => b.score - a.score);
+        return results;
+    }
+}
+async function handleExpand(args) {
+    if (!expanderPipeline)
+        throw new Error('expander not loaded');
+    const result = await expanderPipeline(args.prompt, {
+        max_new_tokens: args.maxNewTokens,
+        no_repeat_ngram_size: args.noRepeatNgramSize,
+    });
+    const text = Array.isArray(result) ? result[0]?.generated_text ?? '' : '';
+    return String(text).trim();
+}
+// --- Main loop ---
+let shuttingDown = false;
+const inflight = new Set();
+async function loadModel() {
+    switch (role) {
+        case 'embed':
+            await loadEmbedder();
+            break;
+        case 'rerank':
+            await loadReranker();
+            break;
+        case 'expand':
+            await loadExpander();
+            break;
+    }
+}
+async function handleMessage(msg) {
+    try {
+        let result;
+        switch (msg.op) {
+            case 'embed':
+                result = await handleEmbed(msg.args);
+                break;
+            case 'rerank':
+                result = await handleRerank(msg.args);
+                break;
+            case 'expand':
+                result = await handleExpand(msg.args);
+                break;
+        }
+        parentPort.postMessage({ id: msg.id, ok: true, result });
+    }
+    catch (err) {
+        parentPort.postMessage({ id: msg.id, ok: false, error: String(err?.message ?? err) });
+    }
+}
+(async () => {
+    try {
+        await loadModel();
+        parentPort.postMessage({ ready: true, role });
+    }
+    catch (err) {
+        parentPort.postMessage({ ready: false, role, error: String(err?.message ?? err) });
+        process.exit(1);
+    }
+})();
+parentPort.on('message', (msg) => {
+    if (msg?.shutdown) {
+        shuttingDown = true;
+        // Wait for in-flight work, then exit
+        void Promise.allSettled([...inflight]).then(() => {
+            parentPort.postMessage({ shutdown: 'done' });
+            process.exit(0);
+        });
+        return;
+    }
+    if (shuttingDown)
+        return;
+    if (typeof msg?.id !== 'number' || typeof msg?.op !== 'string')
+        return;
+    const promise = handleMessage(msg);
+    inflight.add(promise);
+    void promise.finally(() => inflight.delete(promise));
+});
+//# sourceMappingURL=ml-worker-entry.js.map

package/dist/core/ml-worker-entry.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"ml-worker-entry.js","sourceRoot":"","sources":["../../src/core/ml-worker-entry.ts"],"names":[],"mappings":"AAAA,gDAAgD;AAChD,sCAAsC;AACtC;;;;;;;;;;;;;;GAcG;AAEH,OAAO,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,qBAAqB,CAAC;AAE7D,IAAI,CAAC,UAAU,EAAE,CAAC;IAChB,MAAM,IAAI,KAAK,CAAC,oDAAoD,CAAC,CAAC;AACxE,CAAC;AAGD,MAAM,IAAI,GAAe,UAAU,EAAE,IAAI,CAAC;AAC1C,IAAI,IAAI,KAAK,OAAO,IAAI,IAAI,KAAK,QAAQ,IAAI,IAAI,KAAK,QAAQ,EAAE,CAAC;IAC/D,MAAM,IAAI,KAAK,CAAC,kCAAkC,IAAI,GAAG,CAAC,CAAC;AAC7D,CAAC;AAED,oEAAoE;AAEpE,IAAI,gBAAgB,GAAQ,IAAI,CAAC;AACjC,IAAI,iBAAiB,GAAQ,IAAI,CAAC;AAClC,IAAI,aAAa,GAAQ,IAAI,CAAC;AAC9B,IAAI,gBAAgB,GAAQ,IAAI,CAAC;AAEjC,+EAA+E;AAC/E,gFAAgF;AAChF,mEAAmE;AACnE,2EAA2E;AAC3E,MAAM,aAAa,GAAG,MAAe,CAAC;AAEtC,KAAK,UAAU,YAAY;IACzB,MAAM,EAAE,QAAQ,EAAE,GAAG,MAAM,MAAM,CAAC,2BAA2B,CAAC,CAAC;IAC/D,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,eAAe,IAAI,0BAA0B,CAAC;IAC1E,gBAAgB,GAAG,MAAM,QAAQ,CAAC,oBAAoB,EAAE,OAAO,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,aAAa,EAAE,CAAC,CAAC;AAC7G,CAAC;AAED,KAAK,UAAU,YAAY;IACzB,MAAM,EAAE,aAAa,EAAE,kCAAkC,EAAE,GAAG,MAAM,MAAM,CAAC,2BAA2B,CAAC,CAAC;IACxG,MAAM,OAAO,GAAG,OAAO,CAAC,GAAG,CAAC,kBAAkB,IAAI,+BAA+B,CAAC;IAClF,iBAAiB,GAAG,MAAM,aAAa,CAAC,eAAe,CAAC,OAAO,CAAC,CAAC;IACjE,aAAa,GAAG,MAAM,kCAAkC,CAAC,eAAe,CAAC,OAAO,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,aAAa,EAAE,CAAC,CAAC;AAC9H,CAAC;AAED,KAAK,UAAU,YAAY;IACzB,MAAM,EAAE,QAAQ,EAAE,GAAG,MAAM,MAAM,CAAC,2BAA2B,CAAC,CAAC;IAC/D,gBAAgB,GAAG,MAAM,QAAQ,CAAC,sBAAsB,EAAE,sBAAsB,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,MAAM,EAAE,aAAa,EAAE,CAAC,CAAC;AAC9H,CAAC;AAED,sCAAsC;AAEtC,SAAS,OAAO,CAAC,CAAS;IACxB,OAAO,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AAChC,CAAC;AAED,KAAK,UAAU,WAAW,CAAC,IAAsE;IAC/F,IAAI,CAAC,gBAAgB;QAAE,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;IAC9D,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,UAAU,EAAE,GAAG,IAAI,CAAC;IAC5C,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAClC,MAAM,MAAM,GAAG,MAAM,gBAAgB,CAAC,KAAK,EAAE,EAAE,OAAO,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAC3E,MAAM,IAAI,GAAG,MAAM,CAAC,IAAoB,CAAC;IACzC,MAAM,OAAO,GAAe,EAAE,CAAC;IAC/B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,KAAK,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACtC,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,UAAU,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,UAAU,CAAC,CAAC,CAAC,CAAC;IAC7E,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAID,KAAK,UAAU,YAAY,CAAC,IAA2C;IACrE,IAAI,CAAC,iBAAiB,IAAI,CAAC,aAAa;QAAE,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;IACjF,MAAM,EAAE,KAAK,EAAE,QAAQ,EAAE,GAAG,IAAI,CAAC;IACjC,IAAI,QAAQ,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAC;IAErC,aAAa;IACb,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,QAAQ,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,KAAK,CAAC,CAAC;QAC1C,MAAM,MAAM,GAAG,iBAAiB,CAAC,OAAO,EAAE;YACxC,SAAS,EAAE,QAAQ;YACnB,OAAO,EAAE,IAAI;YACb,UAAU,EAAE,IAAI;YAChB,cAAc,EAAE,IAAI;SACrB,CAAC,CAAC;QACH,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,MAAM,CAAC,CAAC;QAC3C,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,iBAAiB,CAAC;QACzD,MAAM,IAAI,GAAG,MAAM,CAAC,IAA+B,CAAC;QACpD,MAAM,OAAO,GAAmB,EAAE,CAAC;QACnC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACzC,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC;YACtC,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC;QACvD,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC;QAC1C,OAAO,OAAO,CAAC;IACjB,CAAC;IAAC,MAAM,CAAC;QACP,uBAAuB;QACvB,MAAM,OAAO,GAAmB,EAAE,CAAC;QACnC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACzC,IAAI,CAAC;gBACH,MAAM,MAAM,GAAG,iBAAiB,CAAC,KAAK,EAAE;oBACtC,SAAS,EAAE,QAAQ,CAAC,CAAC,CAAC;oBACtB,OAAO,EAAE,IAAI;oBACb,UAAU,EAAE,IAAI;oBAChB,cAAc,EAAE,IAAI;iBACrB,CAAC,CAAC;gBACH,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC,MAAM,CAAC,CAAC;gBAC3C,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,iBAAiB,CAAC;gBACzD,MAAM,QAAQ,GAAG,MAAM,CAAC,IAAI,CAAC,CAAC,CAAW,CAAC;gBAC1C,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,OAAO,CAAC,QAAQ,CAAC,EAAE,CAAC,CAAC;YACvD,CAAC;YAAC,MAAM,CAAC;gBACP,OAAO,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC,CAAC;YACvC,CAAC;QACH,CAAC;QACD,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC;QAC1C,OAAO,OAAO,CAAC;IACjB,CAAC;AACH,CAAC;AAED,KAAK,UAAU,YAAY,CAAC,IAAyE;IACnG,IAAI,CAAC,gBAAgB;QAAE,MAAM,IAAI,KAAK,CAAC,qBAAqB,CAAC,CAAC;IAC9D,MAAM,MAAM,GAAG,MAAM,gBAAgB,CAAC,IAAI,CAAC,MAAM,EAAE;QACjD,cAAc,EAAE,IAAI,CAAC,YAAY;QACjC,oBAAoB,EAAE,IAAI,CAAC,iBAAiB;KAC7C,CAAC,CAAC;IACH,MAAM,IAAI,GAAG,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,CAAE,MAAM,CAAC,CAAC,CAAS,EAAE,cAAc,IAAI,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IACnF,OAAO,MAAM,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,CAAC;AAC7B,CAAC;AAED,oBAAoB;AAEpB,IAAI,YAAY,GAAG,KAAK,CAAC;AACzB,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAiB,CAAC;AAE1C,KAAK,UAAU,SAAS;IACtB,QAAQ,IAAI,EAAE,CAAC;QACb,KAAK,OAAO;YAAE,MAAM,YAAY,EAAE,CAAC;YAAC,MAAM;QAC1C,KAAK,QAAQ;YAAE,MAAM,YAAY,EAAE,CAAC;YAAC,MAAM;QAC3C,KAAK,QAAQ;YAAE,MAAM,YAAY,EAAE,CAAC;YAAC,MAAM;IAC7C,CAAC;AACH,CAAC;AAED,KAAK,UAAU,aAAa,CAAC,GAA8C;IACzE,IAAI,CAAC;QACH,IAAI,MAAe,CAAC;QACpB,QAAQ,GAAG,CAAC,EAAE,EAAE,CAAC;YACf,KAAK,OAAO;gBAAG,MAAM,GAAG,MAAM,WAAW,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;gBAAC,MAAM;YAC3D,KAAK,QAAQ;gBAAE,MAAM,GAAG,MAAM,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;gBAAC,MAAM;YAC5D,KAAK,QAAQ;gBAAE,MAAM,GAAG,MAAM,YAAY,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;gBAAC,MAAM;QAC9D,CAAC;QACD,UAAW,CAAC,WAAW,CAAC,EAAE,EAAE,EAAE,GAAG,CAAC,EAAE,EAAE,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC;IAC5D,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,UAAW,CAAC,WAAW,CAAC,EAAE,EAAE,EAAE,GAAG,CAAC,EAAE,EAAE,EAAE,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,CAAE,GAAa,EAAE,OAAO,IAAI,GAAG,CAAC,EAAE,CAAC,CAAC;IACpG,CAAC;AACH,CAAC;AAED,CAAC,KAAK,IAAI,EAAE;IACV,IAAI,CAAC;QACH,MAAM,SAAS,EAAE,CAAC;QAClB,UAAW,CAAC,WAAW,CAAC,EAAE,KAAK,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC;IACjD,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,UAAW,CAAC,WAAW,CAAC,EAAE,KAAK,EAAE,KAAK,EAAE,IAAI,EAAE,KAAK,EAAE,MAAM,CAAE,GAAa,EAAE,OAAO,IAAI,GAAG,CAAC,EAAE,CAAC,CAAC;QAC/F,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC;AACH,CAAC,CAAC,EAAE,CAAC;AAEL,UAAU,CAAC,EAAE,CAAC,SAAS,EAAE,CAAC,GAAQ,EAAE,EAAE;IACpC,IAAI,GAAG,EAAE,QAAQ,EAAE,CAAC;QAClB,YAAY,GAAG,IAAI,CAAC;QACpB,qCAAqC;QACrC,KAAK,OAAO,CAAC,UAAU,CAAC,CAAC,GAAG,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,EAAE;YAC/C,UAAW,CAAC,WAAW,CAAC,EAAE,QAAQ,EAAE,MAAM,EAAE,CAAC,CAAC;YAC9C,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAClB,CAAC,CAAC,CAAC;QACH,OAAO;IACT,CAAC;IACD,IAAI,YAAY;QAAE,OAAO;IACzB,IAAI,OAAO,GAAG,EAAE,EAAE,KAAK,QAAQ,IAAI,OAAO,GAAG,EAAE,EAAE,KAAK,QAAQ;QAAE,OAAO;IAEvE,MAAM,OAAO,GAAG,aAAa,CAAC,GAAG,CAAC,CAAC;IACnC,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;IACtB,KAAK,OAAO,CAAC,OAAO,CAAC,GAAG,EAAE,CAAC,QAAQ,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC;AACvD,CAAC,CAAC,CAAC"}

package/dist/core/ml-worker.d.ts

@@ -0,0 +1,59 @@
+/**
+ * ML inference dispatch pool.
+ *
+ * STATUS (AWM 0.8.x P1 REVISE, 2026-05-25): worker_threads were the original
+ * plan, but @huggingface/transformers in Node only supports the `cpu`
+ * (native ONNX) and `dml` (Windows GPU) backends. Neither is safe inside
+ * a worker_thread — onnxruntime-node's native bindings store V8 handles
+ * that get invalidated when crossing isolate boundaries, causing
+ * `v8::HandleScope::CreateHandle()` crashes on first inference call.
+ * The browser-only `wasm` backend is not loaded in Node builds of
+ * transformers.js.
+ *
+ * The dispatch abstraction is preserved so a future child_process pool
+ * or HTTP sidecar (see AWM_ML_SIDECAR_URL design in docs/awm-architecture-history.md)
+ * can plug in. For now ALL inference runs in-process. The freeze fix from
+ * P0 (sleep-only consolidation) already eliminates the multi-second
+ * in-band blocks. Individual inference calls (~50ms each) on the main
+ * thread are accepted as-is.
+ *
+ * Test / dev mode: AWM_ML_INPROCESS=1 is honored but is now the default.
+ * The env var remains as a no-op for backwards compatibility.
+ */
+declare let inProcessEmbed: ((args: any) => Promise<number[][]>) | null;
+declare let inProcessRerank: ((args: any) => Promise<Array<{
+    index: number;
+    score: number;
+}>>) | null;
+declare let inProcessExpand: ((args: any) => Promise<string>) | null;
+/** Initialize the pool. Idempotent — safe to call multiple times. */
+export declare function initMLPool(): void;
+/** Register in-process fallback handlers. Called once by the consumer modules. */
+export declare function registerInProcessHandlers(handlers: {
+    embed?: typeof inProcessEmbed;
+    rerank?: typeof inProcessRerank;
+    expand?: typeof inProcessExpand;
+}): void;
+/** True if the pool is operating in in-process mode (no workers). */
+export declare function isInProcessMode(): boolean;
+export declare function dispatchEmbed(args: {
+    texts: string[];
+    pooling: 'cls' | 'mean';
+    dimensions: number;
+}): Promise<number[][]>;
+export declare function dispatchRerank(args: {
+    query: string;
+    passages: string[];
+}): Promise<Array<{
+    index: number;
+    score: number;
+}>>;
+export declare function dispatchExpand(args: {
+    prompt: string;
+    maxNewTokens: number;
+    noRepeatNgramSize: number;
+}): Promise<string>;
+/** Graceful shutdown. Waits up to 2s for queue drain, then terminates. */
+export declare function shutdownMLPool(): Promise<void>;
+export {};
+//# sourceMappingURL=ml-worker.d.ts.map

package/dist/core/ml-worker.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"ml-worker.d.ts","sourceRoot":"","sources":["../../src/core/ml-worker.ts"],"names":[],"mappings":"AAEA;;;;;;;;;;;;;;;;;;;;;GAqBG;AAgCH,QAAA,IAAI,cAAc,EAAE,CAAC,CAAC,IAAI,EAAE,GAAG,KAAK,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,IAAW,CAAC;AACvE,QAAA,IAAI,eAAe,EAAE,CAAC,CAAC,IAAI,EAAE,GAAG,KAAK,OAAO,CAAC,KAAK,CAAC;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAAC,CAAC,CAAC,GAAG,IAAW,CAAC;AACrG,QAAA,IAAI,eAAe,EAAE,CAAC,CAAC,IAAI,EAAE,GAAG,KAAK,OAAO,CAAC,MAAM,CAAC,CAAC,GAAG,IAAW,CAAC;AAkHpE,qEAAqE;AACrE,wBAAgB,UAAU,IAAI,IAAI,CAWjC;AAED,kFAAkF;AAClF,wBAAgB,yBAAyB,CAAC,QAAQ,EAAE;IAClD,KAAK,CAAC,EAAE,OAAO,cAAc,CAAC;IAC9B,MAAM,CAAC,EAAE,OAAO,eAAe,CAAC;IAChC,MAAM,CAAC,EAAE,OAAO,eAAe,CAAC;CACjC,GAAG,IAAI,CAIP;AAED,qEAAqE;AACrE,wBAAgB,eAAe,IAAI,OAAO,CAEzC;AAmDD,wBAAsB,aAAa,CAAC,IAAI,EAAE;IAAE,KAAK,EAAE,MAAM,EAAE,CAAC;IAAC,OAAO,EAAE,KAAK,GAAG,MAAM,CAAC;IAAC,UAAU,EAAE,MAAM,CAAA;CAAE,GAAG,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAE/H;AAED,wBAAsB,cAAc,CAAC,IAAI,EAAE;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,QAAQ,EAAE,MAAM,EAAE,CAAA;CAAE,GAAG,OAAO,CAAC,KAAK,CAAC;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAA;CAAE,CAAC,CAAC,CAElI;AAED,wBAAsB,cAAc,CAAC,IAAI,EAAE;IAAE,MAAM,EAAE,MAAM,CAAC;IAAC,YAAY,EAAE,MAAM,CAAC;IAAC,iBAAiB,EAAE,MAAM,CAAA;CAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAE/H;AAED,0EAA0E;AAC1E,wBAAsB,cAAc,IAAI,OAAO,CAAC,IAAI,CAAC,CAiBpD"}