npm - @rubytech/taskmaster - Versions diffs - 1.0.67 → 1.0.68 - Mend

@rubytech/taskmaster 1.0.67 → 1.0.68

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (12) hide show

package/dist/build-info.json +2 -2
package/dist/control-ui/assets/{index-Uo_tQYx1.css → index-BCh3mx9Z.css} +1 -1
package/dist/control-ui/assets/{index-8pJBjxcK.js → index-Tpr1NFEw.js} +162 -162
package/dist/control-ui/assets/index-Tpr1NFEw.js.map +1 -0
package/dist/control-ui/index.html +2 -2
package/dist/gateway/public-chat/session.js +16 -24
package/dist/gateway/server-methods/memory.js +2 -0
package/dist/memory/embeddings-gemini.js +55 -23
package/dist/memory/embeddings.js +11 -12
package/dist/memory/manager.js +14 -1
package/package.json +1 -1
package/dist/control-ui/assets/index-8pJBjxcK.js.map +0 -1

package/dist/control-ui/index.html CHANGED Viewed

@@ -6,8 +6,8 @@
     <title>Taskmaster Control</title>
     <meta name="color-scheme" content="dark light" />
     <link rel="icon" type="image/png" href="./favicon.png" />
-    <script type="module" crossorigin src="./assets/index-8pJBjxcK.js"></script>
-    <link rel="stylesheet" crossorigin href="./assets/index-Uo_tQYx1.css">
+    <script type="module" crossorigin src="./assets/index-Tpr1NFEw.js"></script>
+    <link rel="stylesheet" crossorigin href="./assets/index-BCh3mx9Z.css">
   </head>
   <body>
     <taskmaster-app></taskmaster-app>

package/dist/gateway/public-chat/session.js CHANGED Viewed

@@ -5,35 +5,27 @@
  * Anonymous sessions use a cookie-based identifier; verified sessions use the
  * phone number so they share the same DM session as WhatsApp.
  */
-import { resolveDefaultAgentId } from "../../agents/agent-scope.js";
+import { listBoundAccountIds } from "../../routing/bindings.js";
+import { resolveAgentRoute } from "../../routing/resolve-route.js";
 import { normalizeAgentId } from "../../routing/session-key.js";
 /**
  * Find the agent that handles public-facing WhatsApp DMs.
- * Priority: binding to whatsapp DM > agent named "public" > default agent.
+ *
+ * Uses the same routing logic as WhatsApp itself: calls resolveAgentRoute
+ * with a synthetic unknown-peer DM on the first WhatsApp account. This
+ * guarantees the public chat routes to the exact same agent that handles
+ * unknown WhatsApp DMs.
  */
 export function resolvePublicAgentId(cfg) {
-    const bindings = cfg.bindings ?? [];
-    // Find agent bound to whatsapp DMs (the public-facing agent)
-    for (const binding of bindings) {
-        if (binding.match.channel === "whatsapp" &&
-            binding.match.peer?.kind === "dm" &&
-            !binding.match.peer.id) {
-            return normalizeAgentId(binding.agentId);
-        }
-    }
-    // Any whatsapp binding
-    for (const binding of bindings) {
-        if (binding.match.channel === "whatsapp") {
-            return normalizeAgentId(binding.agentId);
-        }
-    }
-    // Agent explicitly named "public"
-    const agents = cfg.agents?.list ?? [];
-    const publicAgent = agents.find((a) => a.id === "public");
-    if (publicAgent)
-        return normalizeAgentId(publicAgent.id);
-    // Fall back to default agent
-    return resolveDefaultAgentId(cfg);
+    const accountIds = listBoundAccountIds(cfg, "whatsapp");
+    const accountId = accountIds[0] ?? "default";
+    const route = resolveAgentRoute({
+        cfg,
+        channel: "whatsapp",
+        accountId,
+        peer: { kind: "dm", id: "__public_chat__" },
+    });
+    return normalizeAgentId(route.agentId);
 }
 /**
  * Build the session key for a public-chat visitor.

package/dist/gateway/server-methods/memory.js CHANGED Viewed

@@ -20,6 +20,8 @@ export const memoryHandlers = {
                 ok: true,
                 agentId,
                 dirty: status.dirty,
+                syncing: status.syncing,
+                syncProgress: status.syncProgress,
                 files: status.files,
                 chunks: status.chunks,
                 sources: status.sources,

package/dist/memory/embeddings-gemini.js CHANGED Viewed

@@ -41,6 +41,14 @@ function normalizeGeminiBaseUrl(raw) {
 function buildGeminiModelPath(model) {
     return model.startsWith("models/") ? model : `models/${model}`;
 }
+/** Extract retry delay from a Gemini 429 response body, defaulting to 60s. */
+function parseRetryDelay(body) {
+    // Gemini includes "retryDelay": "52s" in the response.
+    const match = body.match(/"retryDelay"\s*:\s*"(\d+)s?"/);
+    if (match)
+        return Math.max(1, Number(match[1]));
+    return 60;
+}
 export async function createGeminiEmbeddingProvider(options) {
     const client = await resolveGeminiEmbeddingClient(options);
     const baseUrl = client.baseUrl.replace(/\/$/, "");
@@ -49,20 +57,31 @@ export async function createGeminiEmbeddingProvider(options) {
     const embedQuery = async (text) => {
         if (!text.trim())
             return [];
-        const res = await fetch(embedUrl, {
-            method: "POST",
-            headers: client.headers,
-            body: JSON.stringify({
-                content: { parts: [{ text }] },
-                taskType: "RETRIEVAL_QUERY",
-            }),
+        const body = JSON.stringify({
+            content: { parts: [{ text }] },
+            taskType: "RETRIEVAL_QUERY",
         });
-        if (!res.ok) {
-            const payload = await res.text();
-            throw new Error(`gemini embeddings failed: ${res.status} ${payload}`);
+        const maxRetries = 3;
+        for (let attempt = 0; attempt <= maxRetries; attempt++) {
+            const res = await fetch(embedUrl, {
+                method: "POST",
+                headers: client.headers,
+                body,
+            });
+            if (res.status === 429 && attempt < maxRetries) {
+                const retryAfter = parseRetryDelay(await res.text());
+                log.info(`gemini rate limit hit; retrying in ${retryAfter}s`);
+                await new Promise((resolve) => setTimeout(resolve, retryAfter * 1000));
+                continue;
+            }
+            if (!res.ok) {
+                const payload = await res.text();
+                throw new Error(`gemini embeddings failed: ${res.status} ${payload}`);
+            }
+            const payload = (await res.json());
+            return payload.embedding?.values ?? [];
         }
-        const payload = (await res.json());
-        return payload.embedding?.values ?? [];
+        throw new Error("gemini embeddings: exhausted retries after rate limiting");
     };
     const embedBatch = async (texts) => {
         if (texts.length === 0)
@@ -72,18 +91,31 @@ export async function createGeminiEmbeddingProvider(options) {
             content: { parts: [{ text }] },
             taskType: "RETRIEVAL_DOCUMENT",
         }));
-        const res = await fetch(batchUrl, {
-            method: "POST",
-            headers: client.headers,
-            body: JSON.stringify({ requests }),
-        });
-        if (!res.ok) {
-            const payload = await res.text();
-            throw new Error(`gemini embeddings failed: ${res.status} ${payload}`);
+        const body = JSON.stringify({ requests });
+        // Retry on 429 (rate limit) — Gemini free tier caps at 100 requests/minute.
+        // The initial bulk index can exceed this; retrying after the cooldown lets it complete.
+        const maxRetries = 3;
+        for (let attempt = 0; attempt <= maxRetries; attempt++) {
+            const res = await fetch(batchUrl, {
+                method: "POST",
+                headers: client.headers,
+                body,
+            });
+            if (res.status === 429 && attempt < maxRetries) {
+                const retryAfter = parseRetryDelay(await res.text());
+                log.info(`gemini rate limit hit; retrying in ${retryAfter}s`);
+                await new Promise((resolve) => setTimeout(resolve, retryAfter * 1000));
+                continue;
+            }
+            if (!res.ok) {
+                const payload = await res.text();
+                throw new Error(`gemini embeddings failed: ${res.status} ${payload}`);
+            }
+            const payload = (await res.json());
+            const embeddings = Array.isArray(payload.embeddings) ? payload.embeddings : [];
+            return texts.map((_, index) => embeddings[index]?.values ?? []);
         }
-        const payload = (await res.json());
-        const embeddings = Array.isArray(payload.embeddings) ? payload.embeddings : [];
-        return texts.map((_, index) => embeddings[index]?.values ?? []);
+        throw new Error("gemini embeddings: exhausted retries after rate limiting");
     };
     return {
         provider: {

package/dist/memory/embeddings.js CHANGED Viewed

@@ -2,21 +2,19 @@ import fsSync from "node:fs";
 import os from "node:os";
 import { createSubsystemLogger } from "../logging/subsystem.js";
 import { resolveUserPath } from "../utils.js";
-import { getCustomProviderApiKey } from "../agents/model-auth.js";
 import { createGeminiEmbeddingProvider } from "./embeddings-gemini.js";
 import { createOpenAiEmbeddingProvider } from "./embeddings-openai.js";
 import { importNodeLlamaCpp } from "./node-llama.js";
 /**
- * Default local embedding model. The 0.6B model is small enough to run on
- * any target hardware (Pi 4GB, Pi 8GB, Mac) without GPU and with minimal
- * RAM overhead (~1-2 GB runtime). Larger models (4B, 8B) can be configured
- * explicitly via `local.modelPath` for users who have tested them on their
- * specific hardware — node-llama-cpp's actual memory footprint varies
- * enormously across GPU vendors and Metal/Vulkan backends.
+ * Default local embedding model. embeddinggemma (329 MB) is proven stable on
+ * Intel x64 Macs and ARM Pis via node-llama-cpp's CPU backend. Larger models
+ * (Qwen3-Embedding 0.6B/4B/8B) can be configured explicitly via
+ * `local.modelPath` but are untested across hardware — Qwen3 models caused
+ * runaway memory consumption (40-76 GB) on Intel x64 Mac + AMD GPU.
  */
 const DEFAULT_LOCAL_MODEL = {
-    model: "hf:Qwen/Qwen3-Embedding-0.6B-GGUF/Qwen3-Embedding-0.6B-Q8_0.gguf",
-    label: "Qwen3-Embedding-0.6B",
+    model: "hf:ggml-org/embeddinggemma-300M-Q8_0-GGUF/embeddinggemma-300M-Q8_0.gguf",
+    label: "embeddinggemma-300M",
 };
 function selectDefaultLocalModel() {
     return DEFAULT_LOCAL_MODEL;
@@ -188,15 +186,16 @@ export async function createEmbeddingProvider(options) {
         }
         // 2. Try remote providers — preferred when API keys are available
         //    (faster, no RAM overhead, no model download).
-        //    Only consider providers whose key is in the config (apiKeys section).
-        //    Environment variables are ignored for auto-selection to avoid using
+        //    Only consider providers whose key is in config.apiKeys.
+        //    Environment variables are ignored for auto-selection to avoid
         //    stale dev keys that would fail at runtime.
+        const configKeys = options.config.apiKeys ?? {};
         const remoteProviders = [
             { id: "openai", configKey: "openai" },
             { id: "gemini", configKey: "google" },
         ];
         for (const { id, configKey } of remoteProviders) {
-            if (!getCustomProviderApiKey(options.config, configKey))
+            if (!configKeys[configKey]?.trim())
                 continue;
             try {
                 const result = await createProvider(id);

package/dist/memory/manager.js CHANGED Viewed

@@ -282,6 +282,7 @@ export class MemoryIndexManager {
     sessionDeltas = new Map();
     sessionWarm = new Set();
     syncing = null;
+    syncProgress = null;
     /**
      * Ensure standard memory directory structure exists.
      * Creates: memory/public, memory/shared, memory/admin, memory/users
@@ -509,8 +510,18 @@ export class MemoryIndexManager {
     async sync(params) {
         if (this.syncing)
             return this.syncing;
-        this.syncing = this.runSync(params).finally(() => {
+        this.syncProgress = { completed: 0, total: 0 };
+        const outerProgress = params?.progress;
+        const wrappedParams = {
+            ...params,
+            progress: (update) => {
+                this.syncProgress = { completed: update.completed, total: update.total };
+                outerProgress?.(update);
+            },
+        };
+        this.syncing = this.runSync(wrappedParams).finally(() => {
             this.syncing = null;
+            this.syncProgress = null;
         });
         return this.syncing;
     }
@@ -663,6 +674,8 @@ export class MemoryIndexManager {
             files: files?.c ?? 0,
             chunks: chunks?.c ?? 0,
             dirty: this.dirty,
+            syncing: this.syncing !== null,
+            syncProgress: this.syncProgress ?? undefined,
             workspaceDir: this.workspaceDir,
             dbPath: this.settings.store.path,
             provider: this.provider.id,

package/package.json CHANGED Viewed

@@ -1,6 +1,6 @@
 {
   "name": "@rubytech/taskmaster",
-  "version": "1.0.67",
+  "version": "1.0.68",
   "description": "AI-powered business assistant for small businesses",
   "publishConfig": {
     "access": "public"