osborn 0.9.221 → 0.9.223

package/dist/embedder.js

@@ -15,8 +15,8 @@
  *
  * Model cache honors TRANSFORMERS_CACHE / HF_HOME so it can be baked into the image.
  */
-import { EMBED_DIM } from './session-store.js';
-const MODEL_ID = process.env.OSBORN_EMBED_MODEL || 'Xenova/all-MiniLM-L6-v2';
+import { EMBED_DIM, EMBED_MODEL } from './session-store.js';
+const MODEL_ID = EMBED_MODEL;
 let pipelinePromise = null;
 let disabled = false;
 /** Quantize a normalized float32 embedding to int8[dim] (×127, clamped). */

package/dist/env-keys.d.ts

@@ -0,0 +1,27 @@
+/**
+ * env-keys.ts — the agent's source-of-truth list of environment variables it
+ * expects to find in process.env at runtime, mirroring the frontend's
+ * platform-env.ts injection spec.
+ *
+ * Today the agent reads these directly from process.env (some explicitly, some
+ * implicitly via the LiveKit plugin SDKs which read env internally). This module
+ * gives that set ONE name so:
+ *   • a future runtime path can hydrate process.env from a cloud/user config
+ *     layer before the plugins read it, using this list as the set to fetch, and
+ *   • startup can log which expected keys are missing for diagnosis.
+ *
+ * Keep in sync with frontend/src/lib/platform-env.ts (PLATFORM_ENV_SPEC).
+ */
+export interface AgentEnvKey {
+    key: string;
+    required: boolean;
+    /** true = read directly via process.env.X in agent source; false = consumed
+     *  implicitly by a plugin SDK (Deepgram/Soniox/Gemini realtime) reading env. */
+    explicit: boolean;
+    description: string;
+}
+export declare const AGENT_ENV_KEYS: AgentEnvKey[];
+/** Keys the agent treats as required — absence should be surfaced at startup. */
+export declare const REQUIRED_AGENT_ENV_KEYS: string[];
+/** Return the required keys that are missing from the given env (default process.env). */
+export declare function missingRequiredEnv(env?: NodeJS.ProcessEnv): string[];

package/dist/env-keys.js

@@ -0,0 +1,47 @@
+/**
+ * env-keys.ts — the agent's source-of-truth list of environment variables it
+ * expects to find in process.env at runtime, mirroring the frontend's
+ * platform-env.ts injection spec.
+ *
+ * Today the agent reads these directly from process.env (some explicitly, some
+ * implicitly via the LiveKit plugin SDKs which read env internally). This module
+ * gives that set ONE name so:
+ *   • a future runtime path can hydrate process.env from a cloud/user config
+ *     layer before the plugins read it, using this list as the set to fetch, and
+ *   • startup can log which expected keys are missing for diagnosis.
+ *
+ * Keep in sync with frontend/src/lib/platform-env.ts (PLATFORM_ENV_SPEC).
+ */
+export const AGENT_ENV_KEYS = [
+    // Voice transport
+    { key: 'LIVEKIT_URL', required: true, explicit: true, description: 'LiveKit server URL.' },
+    { key: 'LIVEKIT_API_KEY', required: true, explicit: true, description: 'LiveKit API key.' },
+    { key: 'LIVEKIT_API_SECRET', required: true, explicit: true, description: 'LiveKit API secret.' },
+    // STT / inference (plugin-consumed unless noted)
+    { key: 'DEEPGRAM_API_KEY', required: true, explicit: false, description: 'Deepgram STT (LiveKit plugin).' },
+    { key: 'SONIOX_API_KEY', required: false, explicit: false, description: 'Soniox STT (LiveKit plugin).' },
+    { key: 'GROQ_API_KEY', required: false, explicit: true, description: 'Groq fast-brain inference.' },
+    { key: 'OPENAI_API_KEY', required: true, explicit: true, description: 'OpenAI realtime + pipeline.' },
+    { key: 'GOOGLE_API_KEY', required: false, explicit: false, description: 'Google Gemini realtime (plugin).' },
+    // LLM auth (Claude vs OpenRouter switch)
+    { key: 'ANTHROPIC_API_KEY', required: false, explicit: true, description: 'Anthropic/Claude auth (per-user; may be OAuth file instead).' },
+    { key: 'OPENROUTER_API_KEY', required: false, explicit: true, description: 'OpenRouter model routing.' },
+    // Integrations
+    { key: 'RECALL_API_KEY', required: false, explicit: true, description: 'Recall.ai meeting bot.' },
+    { key: 'RECALL_REGION', required: false, explicit: true, description: 'Recall.ai region.' },
+    { key: 'SMITHERY_API_KEY', required: false, explicit: true, description: 'Smithery hosted-MCP catalog.' },
+    // Platform / infra
+    { key: 'OSBORN_API_PORT', required: true, explicit: true, description: 'Agent HTTP server port.' },
+    { key: 'OSBORN_CWD', required: false, explicit: true, description: 'Agent working directory.' },
+    { key: 'OSBORN_SYNC_TOKEN', required: false, explicit: true, description: 'Session export/import auth token.' },
+    { key: 'OSBORN_FRONTEND_URL', required: false, explicit: true, description: 'Frontend base URL for artifact upload.' },
+    { key: 'DEV_DOMAIN', required: false, explicit: true, description: 'Wildcard dev-routing domain.' },
+];
+/** Keys the agent treats as required — absence should be surfaced at startup. */
+export const REQUIRED_AGENT_ENV_KEYS = AGENT_ENV_KEYS
+    .filter((k) => k.required)
+    .map((k) => k.key);
+/** Return the required keys that are missing from the given env (default process.env). */
+export function missingRequiredEnv(env = process.env) {
+    return REQUIRED_AGENT_ENV_KEYS.filter((k) => !env[k]);
+}

package/dist/index.js

@@ -40,6 +40,7 @@ import { DIRECT_MODE_PROMPT } from './prompts.js';
 import { MCP_CATALOG } from './config.js';
 import { getRecallClient } from './recall-client.js';
 import { MeetingTranscriptPoller } from './meeting-transcript-poller.js';
+import { missingRequiredEnv } from './env-keys.js';
 import { llm } from '@livekit/agents';
 // ============================================================
 // PIPELINE MODE: STT (Deepgram) → Claude Agent SDK → TTS
@@ -1816,6 +1817,16 @@ async function main() {
         console.error('\nSet these in your .env file or environment.');
         process.exit(1);
     }
+    // Non-fatal diagnostic: surface any OTHER expected-required keys that are
+    // missing (per the shared env-keys.ts manifest — mirror of the frontend's
+    // platform-env.ts injection spec). The LiveKit trio above is the only
+    // hard-fail; the rest degrade gracefully (a plugin falls back, a feature
+    // no-ops), so we log rather than exit — but a missing OPENAI/DEEPGRAM key is
+    // the usual root cause of "voice works but STT/realtime is dead", worth naming.
+    const missingEnv = missingRequiredEnv().filter((k) => !['LIVEKIT_URL', 'LIVEKIT_API_KEY', 'LIVEKIT_API_SECRET'].includes(k));
+    if (missingEnv.length > 0) {
+        console.warn(`⚠️   Missing expected env keys (non-fatal): ${missingEnv.join(', ')}`);
+    }
     // Parse CLI args
     const cliArgs = parseArgs();
     // Load configuration

package/dist/session-store.d.ts

@@ -30,6 +30,9 @@
 import Database from 'better-sqlite3';
 export declare const STORE_VERSION = 1;
 export declare const EMBED_DIM = 384;
+/** Canonical embed model id — the single source of truth (embedder.ts imports this).
+ * Recorded in meta.embed_model so a model swap is detected as a stale store. */
+export declare const EMBED_MODEL: string;
 /** An embedder turns text into int8[EMBED_DIM] vectors (quantized, normalized). */
 export type Embedder = (texts: string[]) => Promise<Int8Array[] | null>;
 export interface RecallHit {

package/dist/session-store.js

@@ -39,6 +39,9 @@ import { getSessionPaths, getSessionSubAgents, projectPathToSlug } from './sessi
 // ============================================================
 export const STORE_VERSION = 1;
 export const EMBED_DIM = 384;
+/** Canonical embed model id — the single source of truth (embedder.ts imports this).
+ * Recorded in meta.embed_model so a model swap is detected as a stale store. */
+export const EMBED_MODEL = process.env.OSBORN_EMBED_MODEL || 'Xenova/all-MiniLM-L6-v2';
 // ============================================================
 // PATHS
 // ============================================================
@@ -64,6 +67,41 @@ export function openStore(dbPath) {
     sqliteVec.load(db);
     db.pragma('journal_mode = WAL');
     db.pragma('synchronous = NORMAL');
+    // ── Migration gate ──────────────────────────────────────────────
+    // The derived tables (content/fts/vec/sources) are a re-derivable cache of the
+    // JSONL, which is the source of truth. If this DB was built by a different
+    // STORE_VERSION, EMBED_DIM, or embed model, the cache is stale — most fatally the
+    // vec table is fixed at its old dimension, so queries embedded at the new dim throw
+    // or return garbage. Detect that and drop the derived tables + reset offsets; the
+    // next updateSessionStore() rebuilds them fresh from the JSONL at offset 0.
+    //
+    // A store with NO recorded embed_model (i.e. predating this stamp) is treated as
+    // NOT stale — otherwise the first rollout of this code would wipe every existing DB.
+    // We only wipe on a real mismatch, never on absence.
+    const hasMeta = db
+        .prepare(`SELECT 1 FROM sqlite_master WHERE type='table' AND name='meta'`)
+        .get();
+    if (hasMeta) {
+        const readMeta = db.prepare('SELECT value FROM meta WHERE key = ?');
+        const prevVersion = readMeta.get('version')?.value ?? null;
+        const prevDim = readMeta.get('embed_dim')?.value ?? null;
+        const prevModel = readMeta.get('embed_model')?.value ?? null;
+        const stale = (prevVersion != null && prevVersion !== String(STORE_VERSION)) ||
+            (prevDim != null && prevDim !== String(EMBED_DIM)) ||
+            (prevModel != null && prevModel !== EMBED_MODEL);
+        if (stale) {
+            console.warn(`⚠️  recall store ${dbPath} is stale ` +
+                `(version ${prevVersion}→${STORE_VERSION}, dim ${prevDim}→${EMBED_DIM}, ` +
+                `model ${prevModel}→${EMBED_MODEL}) — rebuilding derived tables from JSONL`);
+            db.exec(`
+        DROP TABLE IF EXISTS content;
+        DROP TABLE IF EXISTS fts;
+        DROP TABLE IF EXISTS vec;
+        DROP TABLE IF EXISTS sources;
+        DELETE FROM meta WHERE key IN ('version', 'embed_dim', 'embed_model');
+      `);
+        }
+    }
     db.exec(`
     CREATE TABLE IF NOT EXISTS content (
       id          INTEGER PRIMARY KEY,
@@ -94,9 +132,13 @@ export function openStore(dbPath) {
 
     CREATE TABLE IF NOT EXISTS meta (key TEXT PRIMARY KEY, value TEXT);
   `);
-    const setMeta = db.prepare('INSERT OR IGNORE INTO meta(key, value) VALUES (?, ?)');
+    // INSERT OR REPLACE (not IGNORE) so a rebuilt store restamps its identity; the
+    // migration gate above cleared these rows on a stale open, so this writes the
+    // current version/dim/model that the freshly-created tables were built for.
+    const setMeta = db.prepare('INSERT OR REPLACE INTO meta(key, value) VALUES (?, ?)');
     setMeta.run('version', String(STORE_VERSION));
     setMeta.run('embed_dim', String(EMBED_DIM));
+    setMeta.run('embed_model', EMBED_MODEL);
     return db;
 }
 // ============================================================

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "osborn",
-  "version": "0.9.221",
+  "version": "0.9.223",
   "description": "Voice AI coding assistant - local agent that connects to Osborn frontend",
   "type": "module",
   "bin": {