@tiens.nguyen/gonext-local-worker 1.0.6 → 1.0.8

package/README.md

@@ -1,120 +1,3 @@
-# gonext-local-worker
-
-Runs on **your Mac** next to **Ollama** or any **OpenAI-compatible** local server. It:
-
-1. Polls **`POST /api/worker/jobs/next`** on your GoNext cloud API (Lambda).
-2. Runs the job against **`payload.baseURL`** (your LAN `http://127.0.0.1:11434/v1` etc.).
-3. **`PATCH`**es **`running`** → **`completed`** / **`failed`** so DynamoDB and the web app update.
-
-Set your Firebase **worker user id** in the worker env so it claims your own queued jobs.
-
-## Install
-
-```bash
-npm install -g @tiens.nguyen/gonext-local-worker
-```
-
-Or from source:
-
-```bash
-cd tools/gonext-local-worker
-npm install
-npm link
-```
-
-Requires **Node.js 18+**.
-
-## Configure
-
-Create `~/.gonext/worker.env` (optional):
-
-```bash
-mkdir -p ~/.gonext
-cat > ~/.gonext/worker.env << 'EOF'
-GONEXT_API_BASE=https://YOUR_API.execute-api.ap-southeast-1.amazonaws.com
-GONEXT_WORKER_USER_ID=paste-your-firebase-user-id
-GONEXT_POLL_MS=1500
-EOF
-chmod 600 ~/.gonext/worker.env
-```
-
-Or export in the shell:
-
-```bash
-export GONEXT_API_BASE=https://....execute-api....amazonaws.com
-export GONEXT_WORKER_USER_ID=...
-```
-
-## Run
-
-```bash
-gonext-local-worker
-```
-
-Options:
-
-```bash
-gonext-local-worker --poll-ms 2000
-gonext-local-worker --api-base https://other-host.example
-gonext-local-worker --mode webhook --webhook-port 5001
-gonext-local-worker --mode both
-gonext-local-worker --help
-```
-
-Leave this process **running** while you use async local models from the web app.
-
-## Push-dispatch mode (new option)
-
-You now have 2 ways to update job status:
-
-1. **Polling mode** (existing): worker polls `/api/worker/jobs/next`.
-2. **Webhook mode** (new): API push-dispatches job payloads to your local worker endpoint.
-
-Run local webhook endpoint:
-
-```bash
-gonext-local-worker --mode webhook --webhook-port 5001
-```
-
-or keep both (recommended transition):
-
-```bash
-gonext-local-worker --mode both --webhook-port 5001
-```
-
-Expose webhook with ngrok:
-
-```bash
-ngrok http 5001
-```
-
-Then in web app Settings set **Worker webhook URL** to:
-
-`https://<your-ngrok-domain>/api/dispatch-job`
-
-Local worker API endpoints on port 5001:
-
-- `POST /api/chat` -> proxies to `http://localhost:11434/api/chat`
-- `POST /api/generate` -> proxies to `http://localhost:11434/api/generate`
-- `GET /api/tags` -> proxies to `http://localhost:11434/api/tags`
-- `POST /api/dispatch-job` -> cloud dispatch entrypoint
-- `GET /api/health`
-
-Legacy non-prefixed routes (`/chat`, `/generate`, `/tags`, `/dispatch-job`, `/health`) are also available for compatibility.
-
-## Run at login (macOS LaunchAgent)
-
-1. Copy `launchd/com.gonext.worker.plist.example` to `~/Library/LaunchAgents/com.gonext.worker.plist`.
-2. Edit paths: set **full path** to `gonext-local-worker` (`which gonext-local-worker` after `npm link`).
-3. `launchctl load ~/Library/LaunchAgents/com.gonext.worker.plist`
-
-Unload:
-
-```bash
-launchctl unload ~/Library/LaunchAgents/com.gonext.worker.plist
-```
-
-## Troubleshooting
-
-- **`400` / missing workerUserId** — Set `GONEXT_WORKER_USER_ID` correctly.
-- **Job never claimed** — Ensure DynamoDB + worker routes work; Settings must include **Ollama base URL** so the queued payload points at a reachable host from this Mac (`127.0.0.1` is fine).
+# @tiens.nguyen/gonext-local-worker
+Run:
+  GONEXT_API_BASE=... GONEXT_WORKER_KEY=... npx -y --package @tiens.nguyen/gonext-local-worker gonext-local-worker

package/gonext-local-worker.mjs

@@ -1,115 +1,29 @@
 #!/usr/bin/env node
 /**
- * GoNext local LLM worker — runs on the Mac where Ollama / MLX HTTP server lives.
- * Polls your cloud API for jobs, calls the model locally, PATCHes status back.
+ * Polls GoNext Cloud API for pending local-LLM chat jobs, runs them against the
+ * URLs embedded in the job payload (your LAN Ollama / OpenAI-compatible server),
+ * then PATCHes completion back to the API.
  *
- * Env:
- *   GONEXT_API_BASE   HTTPS origin only (no trailing slash), e.g. https://xxx.execute-api....amazonaws.com
- *   GONEXT_WORKER_USER_ID Firebase user id (owner of queued jobs)
- *   GONEXT_POLL_MS    Poll interval when idle (default 1500)
+ * Usage (from api/):
+ *   export GONEXT_API_BASE=https://xxxx.execute-api....amazonaws.com
+ *   export GONEXT_WORKER_KEY=<plaintext secret from Settings → Worker API key>
+ *   node scripts/local-llm-worker.mjs
  *
- * Optional env files (loaded in order; shell exports win if set before launch):
- *   ~/.gonext/worker.env
- *   ./.env (cwd)
+ * Requires Node 18+ (global fetch). Uses the OpenAI SDK from this package.
  */
-import { homedir } from "node:os";
-import os from "node:os";
-import { join } from "node:path";
-import dotenv from "dotenv";
-import express from "express";
 import OpenAI from "openai";
 
-dotenv.config({ path: join(homedir(), ".gonext", "worker.env") });
-dotenv.config();
+const apiBase = (process.env.GONEXT_API_BASE ?? "").replace(/\/+$/, "");
+const workerKey = process.env.GONEXT_WORKER_KEY ?? "";
+const pollMs = Number(process.env.GONEXT_POLL_MS ?? "1500") || 1500;
 
-function parseArgs(argv) {
-  const out = {
-    help: false,
-    pollMs: undefined,
-    apiBase: undefined,
-    mode: undefined,
-    webhookPort: undefined,
-  };
-  for (let i = 2; i < argv.length; i++) {
-    const a = argv[i];
-    if (a === "--help" || a === "-h") out.help = true;
-    else if (a === "--poll-ms" && argv[i + 1])
-      out.pollMs = Number(argv[++i]);
-    else if (a === "--api-base" && argv[i + 1]) out.apiBase = argv[++i];
-    else if (a === "--mode" && argv[i + 1]) out.mode = argv[++i];
-    else if (a === "--webhook-port" && argv[i + 1])
-      out.webhookPort = Number(argv[++i]);
-  }
-  return out;
-}
-
-function printHelp() {
-  console.log(`
-gonext-local-worker — bridge cloud GoNext jobs ↔ local LLM (Ollama / OpenAI-compatible)
-
-Usage:
-  export GONEXT_API_BASE=https://....amazonaws.com
-  export GONEXT_WORKER_USER_ID=your-firebase-user-id
-  gonext-local-worker
-
-Options:
-  --poll-ms <ms>    Idle poll interval (default 1500 or GONEXT_POLL_MS)
-  --api-base <url>  Override GONEXT_API_BASE
-  --mode <poll|webhook|both>  Worker mode (default: poll)
-  --webhook-port <n>          Local webhook port (default: 5001)
-
-Config files (optional):
-  ~/.gonext/worker.env
-  .env in current directory
-
-Install:
-  npm install -g @tiens.nguyen/gonext-local-worker
-
-Then keep this running while you use the web app with local models.
-`);
-}
-
-const args = parseArgs(process.argv);
-if (args.help) {
-  printHelp();
-  process.exit(0);
-}
-
-const apiBase = (
-  args.apiBase ??
-  process.env.GONEXT_API_BASE ??
-  ""
-).replace(/\/+$/, "");
-const workerUserId = (process.env.GONEXT_WORKER_USER_ID ?? "").trim();
-const pollMs =
-  (Number.isFinite(args.pollMs) && args.pollMs > 0
-    ? args.pollMs
-    : Number(process.env.GONEXT_POLL_MS ?? "1500")) || 1500;
-const modeRaw = String(args.mode ?? process.env.GONEXT_WORKER_MODE ?? "poll");
-const mode = ["poll", "webhook", "both"].includes(modeRaw)
-  ? modeRaw
-  : "poll";
-const webhookPort =
-  (Number.isFinite(args.webhookPort) && args.webhookPort > 0
-    ? args.webhookPort
-    : Number(process.env.GONEXT_WEBHOOK_PORT ?? "5001")) || 5001;
-const ollamaBase = (
-  process.env.GONEXT_LOCAL_OLLAMA_BASE ?? "http://127.0.0.1:11434"
-).replace(/\/+$/, "");
-
-if (!apiBase || !workerUserId) {
+if (!apiBase || !workerKey) {
   console.error(
-    "Missing GONEXT_API_BASE or GONEXT_WORKER_USER_ID.\nSet them or put them in ~/.gonext/worker.env — run with --help."
+    "Set GONEXT_API_BASE (HTTP API origin, no /api suffix) and GONEXT_WORKER_KEY."
   );
   process.exit(1);
 }
 
-const WORKER_VERSION = "1.0.0";
-const WORKER_HOST = os.hostname();
-function ts() {
-  return new Date().toISOString();
-}
-
 function toOpenAIMessages(messages) {
   return messages.map((m) => {
     if (m.role === "user" && m.attachments?.length) {
@@ -132,54 +46,19 @@ async function workerFetch(path, init = {}) {
   const url = `${apiBase}${path.startsWith("/") ? path : `/${path}`}`;
   const headers = {
     "Content-Type": "application/json",
-    "X-Worker-User-Id": workerUserId,
+    "X-Worker-Key": workerKey,
     ...(init.headers ?? {}),
   };
   return fetch(url, { ...init, headers });
 }
 
-let shuttingDown = false;
-let activeJobId = "";
-let lastError = "";
-let webhookServer = null;
-
-async function postHeartbeat(payload) {
-  try {
-    await workerFetch("/api/worker/heartbeat", {
-      method: "POST",
-      body: JSON.stringify({
-        workerVersion: WORKER_VERSION,
-        host: WORKER_HOST,
-        ...payload,
-      }),
-    });
-  } catch {
-    /* keep worker loop alive */
-  }
-}
-
 async function runChatJob(job) {
   const { jobId, payload } = job;
   const start = Date.now();
-  activeJobId = jobId;
-  lastError = "";
-  await postHeartbeat({ state: "running", currentJobId: jobId });
-  const patchRunning = await workerFetch(`/api/worker/jobs/${jobId}`, {
+  await workerFetch(`/api/worker/jobs/${jobId}`, {
     method: "PATCH",
     body: JSON.stringify({ jobStatus: "running" }),
   });
-  if (!patchRunning.ok) {
-    const t = await patchRunning.text().catch(() => "");
-    console.error(`[${ts()}] PATCH running failed ${patchRunning.status}`, t);
-    await postHeartbeat({
-      state: "error",
-      currentJobId: jobId,
-      lastError: `PATCH running failed ${patchRunning.status}`,
-    });
-    lastError = `PATCH running failed ${patchRunning.status}`;
-    activeJobId = "";
-    return;
-  }
 
   const client = new OpenAI({
     baseURL: payload.baseURL,
@@ -199,19 +78,11 @@ async function runChatJob(job) {
       body: JSON.stringify({
         jobStatus: "completed",
         resultText: text,
-        tokenCount: Math.max(1, completion.usage?.total_tokens ?? 1),
+        tokenCount: 1,
         totalTimeSeconds,
       }),
     });
-    console.log(
-      `[${ts()}] completed job ${jobId} (${totalTimeSeconds.toFixed(1)}s)`
-    );
-    await postHeartbeat({
-      state: "idle",
-      currentJobId: "",
-      lastJobCompletedAt: new Date().toISOString(),
-    });
-    activeJobId = "";
+    console.log(`[gonext-worker] completed ${jobId} (${totalTimeSeconds.toFixed(1)}s)`);
   } catch (e) {
     const message = e instanceof Error ? e.message : String(e);
     await workerFetch(`/api/worker/jobs/${jobId}`, {
@@ -222,14 +93,7 @@ async function runChatJob(job) {
         totalTimeSeconds: (Date.now() - start) / 1000,
       }),
     });
-    console.error(`[${ts()}] failed job ${jobId}:`, message);
-    await postHeartbeat({
-      state: "error",
-      currentJobId: jobId,
-      lastError: message,
-    });
-    lastError = message;
-    activeJobId = "";
+    console.error(`[gonext-worker] failed ${jobId}:`, message);
   }
 }
 
@@ -238,268 +102,24 @@ async function pollOnce() {
   if (res.status === 204) return;
   if (!res.ok) {
     const t = await res.text().catch(() => "");
-    throw new Error(`POST /api/worker/jobs/next → ${res.status}: ${t}`);
+    throw new Error(`next failed ${res.status}: ${t}`);
   }
   const job = await res.json();
-  if (job?.jobId && job.payload) {
+  if (job?.jobId) {
     await runChatJob(job);
   }
 }
 
-function sleep(ms) {
-  return new Promise((r) => setTimeout(r, ms));
-}
-
-function toOllamaChatMessages(messages) {
-  return (messages ?? []).map((m) => ({
-    role: m.role,
-    content: m.content,
-    ...(Array.isArray(m.attachments) && m.attachments.length > 0
-      ? { images: m.attachments.map((a) => a.data) }
-      : {}),
-  }));
-}
-
-async function runOllamaAndMaybeUpdateJob(params) {
-  const { endpoint, requestBody, jobId, extractResultText } = params;
-  const hasJob = typeof jobId === "string" && jobId.length > 0;
-  const start = Date.now();
-  if (hasJob) {
-    activeJobId = jobId;
-    lastError = "";
-    await postHeartbeat({ state: "running", currentJobId: jobId });
-    await workerFetch(`/api/worker/jobs/${jobId}`, {
-      method: "PATCH",
-      body: JSON.stringify({ jobStatus: "running" }),
-    });
-  }
-  try {
-    const res = await fetch(`${ollamaBase}${endpoint}`, {
-      method: "POST",
-      headers: { "Content-Type": "application/json" },
-      body: JSON.stringify({
-        stream: false,
-        ...requestBody,
-      }),
-    });
-    const raw = await res.text();
-    if (!res.ok) {
-      throw new Error(`Ollama ${endpoint} failed ${res.status}: ${raw}`);
-    }
-    const parsed = raw ? JSON.parse(raw) : {};
-    if (hasJob) {
-      await workerFetch(`/api/worker/jobs/${jobId}`, {
-        method: "PATCH",
-        body: JSON.stringify({
-          jobStatus: "completed",
-          resultText: extractResultText(parsed),
-          tokenCount: 1,
-          totalTimeSeconds: (Date.now() - start) / 1000,
-        }),
-      });
-      await postHeartbeat({
-        state: "idle",
-        currentJobId: "",
-        lastJobCompletedAt: new Date().toISOString(),
-      });
-      activeJobId = "";
-    }
-    return parsed;
-  } catch (e) {
-    if (hasJob) {
-      const message = e instanceof Error ? e.message : String(e);
-      await workerFetch(`/api/worker/jobs/${jobId}`, {
-        method: "PATCH",
-        body: JSON.stringify({
-          jobStatus: "failed",
-          errorMessage: message,
-          totalTimeSeconds: (Date.now() - start) / 1000,
-        }),
-      });
-      await postHeartbeat({
-        state: "error",
-        currentJobId: jobId,
-        lastError: message,
-      });
-      lastError = message;
-      activeJobId = "";
-    }
-    throw e;
-  }
-}
-
-function startWebhookServer() {
-  const app = express();
-  app.use(express.json({ limit: "10mb" }));
-
-  const healthHandler = (_req, res) => {
-    const state =
-      activeJobId.length > 0 ? "running" : lastError ? "error" : "idle";
-    res.json({
-      ok: true,
-      workerVersion: WORKER_VERSION,
-      host: WORKER_HOST,
-      mode,
-      state,
-      activeJobId: activeJobId || undefined,
-      lastError: lastError || undefined,
-      dispatchPath: "/dispatch-job",
-      chatPath: "/chat",
-      generatePath: "/generate",
-      tagsPath: "/tags",
-      ollamaBase,
-    });
-  };
-  app.get("/health", healthHandler);
-  app.get("/api/health", healthHandler);
-
-  const tagsHandler = async (_req, res) => {
-    try {
-      const r = await fetch(`${ollamaBase}/api/tags`, { method: "GET" });
-      const raw = await r.text();
-      if (!r.ok) {
-        res
-          .status(r.status)
-          .json({ error: `Ollama /api/tags failed ${r.status}`, raw });
-        return;
-      }
-      res.type("application/json").send(raw || "{}");
-    } catch (e) {
-      res.status(500).json({ error: e instanceof Error ? e.message : "error" });
-    }
-  };
-  app.get("/tags", tagsHandler);
-  app.get("/api/tags", tagsHandler);
-
-  const dispatchJobHandler = async (req, res) => {
-    const body = req.body ?? {};
-    if (!body?.jobId || !body?.payload) {
-      res.status(400).json({ error: "Expected { jobId, payload }." });
-      return;
-    }
-    if (activeJobId) {
-      res.status(409).json({ error: "Worker busy.", activeJobId });
-      return;
-    }
-    const payload = body.payload;
-    const chatBody = {
-      model: payload.modelId,
-      messages: toOllamaChatMessages(payload.messages),
-    };
-    void runOllamaAndMaybeUpdateJob({
-      endpoint: "/api/chat",
-      requestBody: chatBody,
-      jobId: String(body.jobId),
-      extractResultText: (j) => j?.message?.content ?? "",
-    }).catch((e) => {
-      console.error("[dispatch-job] error:", e instanceof Error ? e.message : e);
-    });
-    res.status(202).json({ ok: true, accepted: true, jobId: body.jobId });
-  };
-  app.post("/dispatch-job", dispatchJobHandler);
-  app.post("/api/dispatch-job", dispatchJobHandler);
-
-  const chatHandler = async (req, res) => {
-    try {
-      const body = req.body ?? {};
-      const jobId = typeof body.jobId === "string" ? body.jobId : "";
-      const requestBody =
-        body.request && typeof body.request === "object" ? body.request : body;
-      const parsed = await runOllamaAndMaybeUpdateJob({
-        endpoint: "/api/chat",
-        requestBody,
-        jobId,
-        extractResultText: (j) => j?.message?.content ?? "",
-      });
-      res.json(parsed);
-    } catch (e) {
-      res.status(500).json({ error: e instanceof Error ? e.message : "error" });
-    }
-  };
-  app.post("/chat", chatHandler);
-  app.post("/api/chat", chatHandler);
-
-  const generateHandler = async (req, res) => {
+async function main() {
+  console.log(`[gonext-worker] polling ${apiBase} every ${pollMs}ms`);
+  for (;;) {
     try {
-      const body = req.body ?? {};
-      const jobId = typeof body.jobId === "string" ? body.jobId : "";
-      const requestBody =
-        body.request && typeof body.request === "object" ? body.request : body;
-      const parsed = await runOllamaAndMaybeUpdateJob({
-        endpoint: "/api/generate",
-        requestBody,
-        jobId,
-        extractResultText: (j) => j?.response ?? "",
-      });
-      res.json(parsed);
+      await pollOnce();
     } catch (e) {
-      res.status(500).json({ error: e instanceof Error ? e.message : "error" });
+      console.error("[gonext-worker] poll error:", e);
     }
-  };
-  app.post("/generate", generateHandler);
-  app.post("/api/generate", generateHandler);
-
-  webhookServer = app.listen(webhookPort, "127.0.0.1", () => {
-    console.log(
-      `[${ts()}] local worker API on http://127.0.0.1:${webhookPort} (/chat, /generate, /tags, /dispatch-job, /health)`
-    );
-  });
-}
-
-async function main() {
-  console.log(`[${ts()}] gonext-local-worker`);
-  console.log(`  API   ${apiBase}`);
-  console.log(`  mode  ${mode}`);
-  if (mode === "poll" || mode === "both") {
-    console.log(`  poll  every ${pollMs}ms (idle)`);
-  }
-  if (mode === "webhook" || mode === "both") {
-    console.log(`  hook  http://127.0.0.1:${webhookPort}/dispatch-job`);
-  }
-  console.log(`  stop  Ctrl+C`);
-  await postHeartbeat({ state: "idle", currentJobId: "" });
-
-  const loop = async () => {
-    while (!shuttingDown) {
-      try {
-        await pollOnce();
-        await postHeartbeat({ state: "idle", currentJobId: "" });
-      } catch (e) {
-        console.error(`[${ts()}] poll error:`, e instanceof Error ? e.message : e);
-        await postHeartbeat({
-          state: "error",
-          currentJobId: "",
-          lastError: e instanceof Error ? e.message : String(e),
-        });
-      }
-      if (shuttingDown) break;
-      await sleep(pollMs);
-    }
-  };
-
-  const stop = () => {
-    if (shuttingDown) return;
-    shuttingDown = true;
-    console.log(`\n[${ts()}] shutting down…`);
-    if (webhookServer) {
-      webhookServer.close();
-    }
-    process.exit(0);
-  };
-  process.on("SIGINT", stop);
-  process.on("SIGTERM", stop);
-
-  if (mode === "webhook" || mode === "both") {
-    startWebhookServer();
-  }
-  if (mode === "poll" || mode === "both") {
-    await loop();
-  } else {
-    await new Promise(() => {});
+    await new Promise((r) => setTimeout(r, pollMs));
   }
 }
 
-main().catch((e) => {
-  console.error(e);
-  process.exit(1);
-});
+main();

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiens.nguyen/gonext-local-worker",
-  "version": "1.0.6",
+  "version": "1.0.8",
   "description": "Polls GoNext cloud API for async local LLM jobs and runs them against Ollama/OpenAI-compatible servers on this Mac",
   "type": "module",
   "license": "MIT",
@@ -15,6 +15,11 @@
   "bin": {
     "gonext-local-worker": "./gonext-local-worker.mjs"
   },
+  "scripts": {
+    "deploy:local": "npm install -g .",
+    "run": "gonext-local-worker",
+    "publish:org": "npm publish --access public"
+  },
   "files": [
     "gonext-local-worker.mjs",
     "README.md",

package/launchd/com.gonext.worker.plist.example

@@ -1,34 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
-<plist version="1.0">
-<dict>
-  <key>Label</key>
-  <string>com.gonext.worker</string>
-  <key>ProgramArguments</key>
-  <array>
-    <!-- Replace with output of: which gonext-local-worker -->
-    <string>/usr/local/bin/gonext-local-worker</string>
-  </array>
-  <!-- Optional: uncomment to force Node path -->
-  <!--
-  <array>
-    <string>/usr/local/bin/node</string>
-    <string>/ABSOLUTE/PATH/TO/gonext/tools/gonext-local-worker/gonext-local-worker.mjs</string>
-  </array>
-  -->
-  <key>RunAtLoad</key>
-  <true/>
-  <key>KeepAlive</key>
-  <true/>
-  <key>StandardOutPath</key>
-  <string>/tmp/gonext-worker.log</string>
-  <key>StandardErrorPath</key>
-  <string>/tmp/gonext-worker.err</string>
-  <key>EnvironmentVariables</key>
-  <dict>
-    <!-- Or rely on ~/.gonext/worker.env (loaded by the worker) -->
-    <key>PATH</key>
-    <string>/usr/local/bin:/usr/bin:/bin:/opt/homebrew/bin</string>
-  </dict>
-</dict>
-</plist>