@tiens.nguyen/gonext-local-worker 1.0.175 → 1.0.178

package/gonext-local-worker.mjs

@@ -1799,6 +1799,9 @@ async function runAgentChatJob(job) {
           return Array.isArray(parsed?.workspaces) ? parsed.workspaces : [];
         })
         .catch(() => []),
+      // The `gonext` terminal REPL's launch folder — download_file/unzip_file output
+      // lands here (must be a registered workspace; the python side re-validates).
+      activeWorkspace: payload?.activeWorkspace ?? "",
     });
     // 30 min max for an agent run: multi-step ReAct on a 14B/31B with a cold prompt
     // cache — or a remote Ollama box on slow GPU cold-loading a big model — can run

package/gonext-repl.mjs

@@ -8,13 +8,15 @@
  *
  * It offers to register the CURRENT folder as an agent workspace (with a trust prompt
  * that maps to --allow-run), authenticates with the worker key from ~/.gonext/worker.env
- * (the same file `gonext-local-worker set` / the Wizard write), pulls your agent/coding
- * model + budgets + RAG config FROM YOUR USER SETTINGS via POST /api/worker/agent-payload,
- * then loops on a `>> ` prompt. Each question executes gonext_agent_chat.py DIRECTLY
- * (no job queue — no race with a polling worker daemon), streaming the agent's steps
- * and thinking into the terminal.
+ * (the same file `gonext-local-worker set` / the Wizard write), and loops on a `>> `
+ * prompt. Each question is ENQUEUED as a normal agent_chat job (models/budgets/RAG
+ * resolved from your user settings by POST /api/worker/agent-ask) and executed by the
+ * RUNNING gonext-local-worker daemon — so its terminal shows the full [gonext-agent]
+ * logs, exactly like a web question. The REPL polls the job and streams the persisted
+ * chunks live into the terminal.
  *
- * Commands: /exit  /revert [runId]  /help      Ctrl-C aborts the current run.
+ * Requires the worker daemon to be running (it claims the job).
+ * Commands: /exit  /revert [runId]  /help      Ctrl-C stops following the current run.
  */
 import { readFile, mkdir, writeFile } from "node:fs/promises";
 import { existsSync, statSync } from "node:fs";
@@ -43,9 +45,9 @@ if (argv.includes("--help") || argv.includes("-h")) {
       "  -v, --verbose   show the agent's diagnostic logs ([gonext-agent] …) and python stderr\n" +
       "  -h, --help      this help\n\n" +
       "Auth comes from ~/.gonext/worker.env; models/budgets/RAG come from your web Settings.\n" +
-      "Inside the REPL: /exit · /revert [runId] · /help — Ctrl-C aborts the running turn.\n" +
-      "Note: REPL turns run in THIS process (no job queue), so the gonext-local-worker\n" +
-      "daemon shows no logs for them — use -v to see the agent diagnostics here."
+      "Inside the REPL: /exit · /revert [runId] · /help — Ctrl-C stops following a turn.\n" +
+      "Questions are enqueued like web questions and executed by the RUNNING\n" +
+      "gonext-local-worker daemon — its terminal shows the full [gonext-agent] logs."
   );
   process.exit(0);
 }
@@ -176,104 +178,88 @@ async function fetchAgentPayload(messages) {
   return data;
 }
 
-// ---------- run one agent turn by executing gonext_agent_chat.py directly ----------
-let child = null;
+// ---------- run one agent turn via the JOB QUEUE (the daemon executes it) ----------
+// The question is enqueued exactly like a web question, so the RUNNING
+// gonext-local-worker daemon claims + executes it — its terminal shows the full
+// [gonext-agent] logs. We poll the job here and stream the persisted chunks live.
+let following = false;
+let followAborted = false;
 
-async function runAgentTurn(payload) {
-  // Map the job payload to the python cfg the same way the worker daemon does in
-  // runAgentChatJob (gonext-local-worker.mjs) — KEEP IN SYNC with that block.
-  const input = JSON.stringify({
-    messages: payload?.messages ?? [],
-    agentBaseURL: payload?.agentBaseURL ?? "",
-    agentApiKey: payload?.agentApiKey ?? "",
-    agentModelId: payload?.agentModelId ?? "",
-    codingBaseURL: payload?.codingBaseURL ?? "",
-    codingModelId: payload?.codingModelId ?? "",
-    tools: payload?.tools ?? ["http_request"],
-    maxSteps: payload?.maxSteps ?? 0,
-    researchBudget: payload?.researchBudget ?? 10,
-    agentToolMode: payload?.agentToolMode ?? "code",
-    apiBaseURL: apiBase,
-    workerKey,
-    emailEnabled: payload?.emailEnabled ?? false,
-    emailApiUrl: payload?.emailApiUrl ?? "",
-    emailApiMethod: payload?.emailApiMethod ?? "POST",
-    emailApiHeaders: payload?.emailApiHeaders ?? "",
-    emailBodyTemplate: payload?.emailBodyTemplate ?? "",
-    emailFrom: payload?.emailFrom ?? "",
-    emailAllowList: payload?.emailAllowList ?? "",
-    ragEnabled: payload?.ragEnabled ?? false,
-    ragS3Location: payload?.ragS3Location ?? "",
-    ragAwsRegion: payload?.ragAwsRegion ?? "",
-    ragAwsAccessKeyId: payload?.ragAwsAccessKeyId ?? "",
-    ragAwsSecretAccessKey: payload?.ragAwsSecretAccessKey ?? "",
-    ragEmbedModel: payload?.ragEmbedModel ?? "",
-    ragEmbedUrl: payload?.ragEmbedUrl ?? "",
-    ragTopK: payload?.ragTopK ?? 6,
-    workspaces: await readWorkspaces(),
-  });
+const sleep = (ms) => new Promise((r) => setTimeout(r, ms));
+
+/** Strip <think> markers for live display, and whole think blocks for the answer. */
+const stripThinkTags = (s) => s.replace(/<\/?think>/gi, "");
+const answerFrom = (s) =>
+  s
+    .replace(/<think>[\s\S]*?<\/think>/gi, "")
+    .replace(/<think>[\s\S]*$/i, "")
+    .trim();
 
-  const python =
-    (process.env.GONEXT_PROBE_PYTHON ?? process.env.GONEXT_MLX_LM_PYTHON ?? "").trim() ||
-    "python3";
-  // WeasyPrint (create_pdf) needs Homebrew libs on macOS — same env the worker sets.
-  let extraEnv;
-  if (process.platform === "darwin") {
-    const existing = process.env.DYLD_FALLBACK_LIBRARY_PATH ?? "";
-    extraEnv = {
-      DYLD_FALLBACK_LIBRARY_PATH: [existing, "/opt/homebrew/lib", "/usr/local/lib"]
-        .filter(Boolean)
-        .join(":"),
-    };
+async function runAgentTurn(history) {
+  const res = await fetch(`${apiBase}/api/worker/agent-ask`, {
+    method: "POST",
+    headers: { "Content-Type": "application/json", "X-Worker-Key": workerKey },
+    // cwd = this terminal's folder → agent puts download/unzip output here (when it's
+    // a registered workspace).
+    body: JSON.stringify({ messages: history, cwd: resolve(process.cwd()) }),
+  });
+  const data = await res.json().catch(() => ({}));
+  if (!res.ok) {
+    if (res.status === 404) {
+      throw new Error(
+        "the API does not have POST /api/worker/agent-ask yet — deploy the latest API."
+      );
+    }
+    throw new Error(data?.error || `agent-ask failed (HTTP ${res.status})`);
   }
+  const jobId = data?.jobId;
+  if (!jobId) throw new Error("agent-ask returned no jobId.");
 
-  return new Promise((resolveTurn) => {
-    let finalText = "";
-    let buf = "";
-    child = spawn(python, [join(DIR, "gonext_agent_chat.py")], {
-      env: { ...process.env, ...(extraEnv ?? {}) },
-    });
-    child.stdin.write(input);
-    child.stdin.end();
-    child.stdout.on("data", (chunk) => {
-      buf += chunk.toString();
-      let nl;
-      while ((nl = buf.indexOf("\n")) >= 0) {
-        const line = buf.slice(0, nl);
-        buf = buf.slice(nl + 1);
-        if (!line.trim()) continue;
-        let ev;
-        try {
-          ev = JSON.parse(line);
-        } catch {
-          continue;
-        }
-        if (ev.type === "step" && typeof ev.text === "string") {
-          process.stdout.write(dim(`\n· ${ev.text}`));
-        } else if (ev.type === "stream" && typeof ev.text === "string") {
-          process.stdout.write(dim(ev.text));
-        } else if (ev.type === "final" && typeof ev.text === "string") {
-          finalText = ev.text;
-        } else if (ev.type === "log" && process.env.GONEXT_REPL_VERBOSE === "1") {
-          process.stdout.write(dim(`\n[log] ${ev.text}`));
-        }
+  following = true;
+  followAborted = false;
+  const startedAt = Date.now();
+  let shownChars = 0;
+  let warnedPending = false;
+  try {
+    for (;;) {
+      if (followAborted) {
+        process.stdout.write(
+          dim("\n(stopped following — the job keeps running on the worker)\n")
+        );
+        return "";
       }
-    });
-    child.stderr.on("data", (chunk) => {
-      if (process.env.GONEXT_REPL_VERBOSE === "1") {
-        process.stderr.write(dim(chunk.toString()));
+      const jr = await fetch(`${apiBase}/api/worker/jobs/${jobId}`, {
+        headers: { "X-Worker-Key": workerKey },
+      });
+      const job = await jr.json().catch(() => ({}));
+      if (!jr.ok) throw new Error(job?.error || `job poll failed (HTTP ${jr.status})`);
+      const text = typeof job.resultText === "string" ? job.resultText : "";
+      if (job.jobStatus === "completed") {
+        // Don't stream the final delta dim — the answer prints bright right after.
+        return answerFrom(text);
       }
-    });
-    child.on("error", (err) => {
-      console.error(red(`\ngonext: failed to start ${python}: ${err.message}`));
-      child = null;
-      resolveTurn("");
-    });
-    child.on("exit", () => {
-      child = null;
-      resolveTurn(finalText);
-    });
-  });
+      if (text.length > shownChars) {
+        process.stdout.write(dim(stripThinkTags(text.slice(shownChars))));
+        shownChars = text.length;
+      }
+      if (job.jobStatus === "failed" || job.jobStatus === "cancelled") {
+        throw new Error(job.errorMessage || `job ${job.jobStatus}`);
+      }
+      if (
+        job.jobStatus === "pending" &&
+        !warnedPending &&
+        Date.now() - startedAt > 6000
+      ) {
+        warnedPending = true;
+        process.stdout.write(
+          red("\n⚠ no worker has claimed the job — is `gonext-local-worker` running?\n")
+        );
+      }
+      await sleep(700);
+    }
+  } finally {
+    following = false;
+  }
 }
 
 // ---------- REPL ----------
@@ -299,10 +285,9 @@ async function main() {
 
   const history = [];
   rl.on("SIGINT", () => {
-    if (child) {
-      child.kill("SIGKILL");
-      child = null;
-      process.stdout.write(red("\n^C run aborted\n"));
+    if (following) {
+      followAborted = true;
+      process.stdout.write(red("\n^C "));
     } else {
       process.stdout.write(dim("\n(/exit to quit)\n") + cyan(">> "));
     }
@@ -334,14 +319,13 @@ async function main() {
     }
     history.push({ role: "user", content: line });
     try {
-      const { payload } = await fetchAgentPayload(history);
-      const answer = await runAgentTurn(payload);
+      const answer = await runAgentTurn(history);
       if (answer) {
         history.push({ role: "assistant", content: answer });
         console.log(`\n\n${answer}\n`);
       } else {
-        history.pop(); // failed turn — don't poison the next request's history
-        console.log(red("\n(no answer — run failed or was aborted)\n"));
+        history.pop(); // aborted/failed turn — don't poison the next request's history
+        console.log(red("\n(no answer — run aborted or failed)\n"));
       }
     } catch (err) {
       history.pop();

package/gonext_agent_chat.py

@@ -1158,6 +1158,17 @@ def _rag_workdir(source_key: str) -> str:
     return base
 
 
+def _rag_base_dir(source_key: str) -> str:
+    """Where a URL's downloaded/unzipped files should live. Prefer the active terminal
+    workspace (the folder the `gonext` REPL was opened from) so files land in the
+    user's project; otherwise fall back to the per-URL cache dir under ~/.gonext."""
+    import os
+    if _WS_ACTIVE:
+        os.makedirs(_WS_ACTIVE, exist_ok=True)
+        return _WS_ACTIVE
+    return _rag_workdir(source_key)
+
+
 def _rag_download(url: str, dest_path: str = "") -> tuple:
     """Download a public URL with SSRF + size guards — IDEMPOTENT per URL. Every URL
     maps to a deterministic work dir (~/.gonext/rag-work/<urlHash>/); relative
@@ -1168,7 +1179,7 @@ def _rag_download(url: str, dest_path: str = "") -> tuple:
     src = _rag_gdrive_direct((url or "").strip())
     _rag_assert_safe_url(src)
     key = _rag_source_key(url)
-    workdir = _rag_workdir(key)
+    workdir = _rag_base_dir(key)
     dest_path = os.path.expanduser((dest_path or "").strip())
     if not dest_path:
         name = os.path.basename(src.split("?")[0]) or "download.bin"
@@ -1176,21 +1187,23 @@ def _rag_download(url: str, dest_path: str = "") -> tuple:
             name += ".zip"
         dest_path = os.path.join(workdir, name)
     elif not os.path.isabs(dest_path):
-        # Contain relative paths in the per-URL work dir — models pass bare names like
+        # Contain relative paths in the work dir — models pass bare names like
         # "project.zip" which would otherwise land in the worker's cwd (the home dir).
         dest_path = os.path.join(workdir, dest_path)
     # Cache hit 1: the exact target already exists.
     if os.path.isfile(dest_path) and os.path.getsize(dest_path) > 0:
         return dest_path, True
-    # Cache hit 2: this URL was downloaded before under a different name — the work dir
-    # is per-URL, so any archive already in it IS this URL's file.
-    try:
-        for fn in sorted(os.listdir(workdir)):
-            p = os.path.join(workdir, fn)
-            if os.path.isfile(p) and fn.lower().endswith(".zip") and os.path.getsize(p) > 0:
-                return p, True
-    except OSError:
-        pass
+    # Cache hit 2: this URL was downloaded before under a different name — ONLY safe in
+    # the per-URL cache dir (one URL per dir). In an active workspace the folder holds
+    # many unrelated files, so a bare "any zip = this URL" match would be wrong.
+    if not _WS_ACTIVE:
+        try:
+            for fn in sorted(os.listdir(workdir)):
+                p = os.path.join(workdir, fn)
+                if os.path.isfile(p) and fn.lower().endswith(".zip") and os.path.getsize(p) > 0:
+                    return p, True
+        except OSError:
+            pass
     os.makedirs(os.path.dirname(dest_path) or ".", exist_ok=True)
     req = urllib.request.Request(src, headers={"User-Agent": "gonext-rag/1.0"})
     total = 0
@@ -1220,8 +1233,15 @@ def _rag_resolve_zip(zip_path: str) -> str:
     if os.path.isfile(zp):
         return zp
     if not os.path.isabs(zp):
+        base = os.path.basename(zp)
+        # Look in the active terminal workspace first (that's where downloads land now),
+        # then fall back to the per-URL cache dirs.
+        if _WS_ACTIVE:
+            cand = os.path.join(_WS_ACTIVE, base)
+            if os.path.isfile(cand):
+                return cand
         hits = _glob.glob(os.path.join(
-            os.path.expanduser("~"), ".gonext", "rag-work", "*", os.path.basename(zp)))
+            os.path.expanduser("~"), ".gonext", "rag-work", "*", base))
         if hits:
             return hits[0]
     return zp
@@ -1424,6 +1444,12 @@ def _rag_parse_location(loc: str):
 # require _ws_write_allowed (workspace root + not inside .git/).
 _WS_ROOTS: list = []  # [{"name": str, "path": realpath str, "allowRun": bool}]
 
+# The active terminal workspace: the folder the `gonext` REPL was launched from
+# (passed through as payload.activeWorkspace). When set, download_file/unzip_file put
+# their output HERE instead of the shared ~/.gonext/rag-work cache, so files land in
+# the user's actual project folder. Only honoured if it's a registered workspace root.
+_WS_ACTIVE: str = ""
+
 
 def _rag_read_allowed(path: str) -> str:
     """Resolve `path` and ensure it stays within a safe root (worker cwd, temp dir,
@@ -1741,7 +1767,7 @@ def run_agent_chat(cfg):
     # Registered on the Mac via `gonext-local-worker workspace add <path>`; the worker
     # passes the registry through cfg. Only these roots are writable — the security
     # boundary for the coding tools.
-    global _WS_ROOTS
+    global _WS_ROOTS, _WS_ACTIVE
     _WS_ROOTS = []
     for w in (cfg.get("workspaces") or []):
         try:
@@ -1756,6 +1782,20 @@ def run_agent_chat(cfg):
         except Exception:  # noqa: BLE001
             continue
     _WS_AVAILABLE = bool(_WS_ROOTS)
+    # Active terminal workspace (the `gonext` REPL's launch folder) — download_file /
+    # unzip_file put their output here. Only honoured if it's a registered workspace
+    # root, so it can't redirect writes to an arbitrary folder.
+    _WS_ACTIVE = ""
+    try:
+        import os as _os
+        aw = _os.path.realpath(_os.path.expanduser(str(cfg.get("activeWorkspace") or "")))
+        if aw and _os.path.isdir(aw) and any(
+            aw == w["path"] or aw.startswith(w["path"] + _os.sep) for w in _WS_ROOTS
+        ):
+            _WS_ACTIVE = aw
+            _log(f"active workspace: {aw} (download/unzip output → here)")
+    except Exception:  # noqa: BLE001
+        _WS_ACTIVE = ""
     if _WS_AVAILABLE:
         _log(f"workspaces: {[(w['name'], w['path'], w['allowRun']) for w in _WS_ROOTS]}")
         # Coding tasks are edit→test→fix loops — 5 steps is too tight. Bump the default

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiens.nguyen/gonext-local-worker",
-  "version": "1.0.175",
+  "version": "1.0.178",
   "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",