@tiens.nguyen/gonext-local-worker 1.0.205 → 1.0.206

package/gonext-local-worker.mjs

@@ -2020,6 +2020,71 @@ function normalizeBaseUrl(raw) {
   return typeof raw === "string" ? raw.trim().replace(/\/+$/, "") : "";
 }
 
+/**
+ * "Load Models" (task #44): probe the agent coding-model server's model list. Tries the
+ * OpenAI-compatible /v1/models (MLX + Ollama both serve it) and falls back to Ollama's
+ * /api/tags. The server is on the user's network, so only this worker can reach it. The
+ * job's resultText is JSON {models:[...]} the web parses into checkboxes.
+ */
+async function runListModelsJob(job) {
+  const { jobId, payload } = job;
+  const start = Date.now();
+  const runRes = await workerFetch(`/api/worker/jobs/${jobId}`, {
+    method: "PATCH",
+    body: JSON.stringify({ jobStatus: "running" }),
+  });
+  await ensureWorkerOk(runRes, `mark running list_models jobId=${jobId}`);
+  try {
+    const base = normalizeBaseUrl(payload?.url).replace(/\/v1$/i, "");
+    if (!base) throw new Error("No coding-model URL provided.");
+    const models = new Set();
+    // 1) OpenAI-compatible /v1/models → { data: [{ id }] }
+    try {
+      const r = await fetch(`${base}/v1/models`, { method: "GET" });
+      if (r.ok) {
+        const j = await r.json().catch(() => ({}));
+        for (const m of Array.isArray(j?.data) ? j.data : []) {
+          if (typeof m?.id === "string" && m.id.trim()) models.add(m.id.trim());
+        }
+      }
+    } catch {
+      /* try the Ollama shape next */
+    }
+    // 2) Ollama /api/tags → { models: [{ name }] }
+    if (models.size === 0) {
+      const r = await fetch(`${base}/api/tags`, { method: "GET" });
+      if (r.ok) {
+        const j = await r.json().catch(() => ({}));
+        for (const m of Array.isArray(j?.models) ? j.models : []) {
+          if (typeof m?.name === "string" && m.name.trim()) models.add(m.name.trim());
+        }
+      }
+    }
+    const list = Array.from(models).sort();
+    const doneRes = await workerFetch(`/api/worker/jobs/${jobId}`, {
+      method: "PATCH",
+      body: JSON.stringify({
+        jobStatus: "completed",
+        resultText: JSON.stringify({ models: list }),
+        totalTimeSeconds: (Date.now() - start) / 1000,
+      }),
+    });
+    await ensureWorkerOk(doneRes, `complete list_models jobId=${jobId}`);
+    console.log(`[gonext-worker] list_models ${jobId}: ${list.length} model(s) from ${base}`);
+  } catch (e) {
+    const message = e instanceof Error ? e.message : String(e);
+    await workerFetch(`/api/worker/jobs/${jobId}`, {
+      method: "PATCH",
+      body: JSON.stringify({
+        jobStatus: "failed",
+        errorMessage: `Could not list models: ${message}`,
+        totalTimeSeconds: (Date.now() - start) / 1000,
+      }),
+    }).catch(() => {});
+    console.error(`[gonext-worker] failed list_models ${jobId}:`, message);
+  }
+}
+
 function normalizeOpenAiV1Root(raw) {
   const base = normalizeBaseUrl(raw);
   if (!base) return "";
@@ -2601,6 +2666,10 @@ async function pollOnce() {
       await runHttpProbeJob(job);
       return;
     }
+    if (job.jobType === "list_models") {
+      await runListModelsJob(job);
+      return;
+    }
     if (job.jobType === "agent_chat") {
       await runAgentChatJob(job);
       return;

package/gonext-repl.mjs

@@ -21,7 +21,7 @@
  * interrupted investigation actually resumes it). /reset clears it.
  *
  * Requires the worker daemon to be running (it claims the job).
- * Commands: /exit  /reset  /revert [runId]  /help      Ctrl-C stops following the current run.
+ * Commands: /exit  /model  /reset  /revert [runId]  /help   Ctrl-C stops following the current run.
  */
 import { readFile, mkdir, writeFile, unlink } from "node:fs/promises";
 import { existsSync, statSync, readFileSync } from "node:fs";
@@ -119,7 +119,7 @@ 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 · /reset · /revert [runId] · /help — Ctrl-C stops following a turn.\n" +
+      "Inside the REPL: /exit · /model · /reset · /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.\n" +
       "Conversation history is saved per-folder in ~/.gonext/sessions and resumed next time\n" +
@@ -193,6 +193,19 @@ rl.on("line", (l) => {
     _lineQueue.push(clean);
   }
 });
+// Neutralize navigation keys (↑/↓ history recall, ←/→ cursor, Home/End, etc.) during a
+// turn: echo is already muted, but readline still recalls history into the invisible
+// buffer on ↑/↓ — reset the line state after every keypress so nothing accumulates or
+// moves. Ctrl+C is unaffected (readline emits its SIGINT before this handler runs).
+if (process.stdin.isTTY) {
+  process.stdin.on("keypress", () => {
+    if (following) {
+      rl.line = "";
+      rl.cursor = 0;
+      rl.historyIndex = -1;
+    }
+  });
+}
 rl.on("close", () => {
   _stdinClosed = true;
   if (_lineWaiter) {
@@ -338,6 +351,53 @@ async function clearSession(cwd) {
 }
 
 // ---------- fetch the ready-to-run agent payload from user settings ----------
+// /model command: fetch the allowed coding models (default + web-curated whitelist),
+// show a numbered picker, and set sessionCodingModel to the choice for this session.
+async function chooseModel() {
+  let probe;
+  try {
+    probe = await fetchAgentPayload([{ role: "user", content: "ping" }]);
+  } catch (err) {
+    console.log(red(`  couldn't load models: ${err.message}\n`));
+    return;
+  }
+  const allowed = Array.isArray(probe?.codingAllowed) ? probe.codingAllowed : [];
+  const def = (probe?.codingDefault ?? "").trim();
+  if (allowed.length === 0) {
+    console.log(dim("  No coding model configured. Set one in the web app → Settings → Agent.\n"));
+    return;
+  }
+  if (allowed.length === 1) {
+    console.log(
+      dim(`  Only one coding model is available: ${allowed[0]}.\n`) +
+        dim("  Add more in the web app → Settings → Agent → Load Models (checkboxes).\n")
+    );
+    return;
+  }
+  const active = sessionCodingModel || def;
+  console.log(dim("  Choose the coding model for this session:"));
+  allowed.forEach((m, i) => {
+    const mark = m === active ? green(" ●") : "  ";
+    const tag = m === def ? dim(" (default)") : "";
+    console.log(`   ${mark} ${i + 1}. ${m}${tag}`);
+  });
+  const pick = (await ask(dim("  number (or Enter to keep current): "))).trim();
+  if (!pick) {
+    console.log("");
+    return;
+  }
+  const idx = Number.parseInt(pick, 10) - 1;
+  if (!Number.isInteger(idx) || idx < 0 || idx >= allowed.length) {
+    console.log(red("  invalid choice.\n"));
+    return;
+  }
+  const chosen = allowed[idx];
+  // Empty override means "use the account default" — so don't send an override when the
+  // user picks the default (keeps the payload clean and lets a later default change win).
+  sessionCodingModel = chosen === def ? "" : chosen;
+  console.log(green(`  ✓ coding model → ${chosen}${chosen === def ? " (default)" : ""}\n`));
+}
+
 async function fetchAgentPayload(messages) {
   const res = await fetch(`${apiBase}/api/worker/agent-payload`, {
     method: "POST",
@@ -370,6 +430,9 @@ let following = false;
 let followAborted = false;
 let currentJobId = null; // the running turn's jobId — so Ctrl+C can cancel it server-side
 let cancelRequested = false; // a cancel POST has been sent for the current turn
+// Per-session coding-model override chosen via /model (empty = use the account default).
+// Sent to /agent-ask as codingModelOverride; the API validates it against the whitelist.
+let sessionCodingModel = "";
 
 // While a turn runs, mute readline's own echo/redraws so keystrokes can't corrupt the
 // live status display and there's no way to type a new question — only Ctrl+C works.
@@ -381,6 +444,15 @@ rl._writeToOutput = (s) => {
   if (_origWriteToOutput) _origWriteToOutput(s);
   else process.stdout.write(s);
 };
+// _writeToOutput only mutes echoed TEXT; readline's line refresh writes cursor-position
+// + erase escapes (e.g. ESC[1G, ESC[0J) straight to output on every keypress, which
+// would corrupt our live status display. No-op it during a turn so a keypress produces
+// ZERO terminal output.
+const _origRefreshLine = rl._refreshLine ? rl._refreshLine.bind(rl) : null;
+rl._refreshLine = () => {
+  if (following) return;
+  if (_origRefreshLine) _origRefreshLine();
+};
 
 const sleep = (ms) => new Promise((r) => setTimeout(r, ms));
 
@@ -397,8 +469,12 @@ async function runAgentTurn(history) {
     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()) }),
+    // a registered workspace). codingModelOverride = the /model choice for this session.
+    body: JSON.stringify({
+      messages: history,
+      cwd: resolve(process.cwd()),
+      ...(sessionCodingModel ? { codingModelOverride: sessionCodingModel } : {}),
+    }),
   });
   const data = await res.json().catch(() => ({}));
   if (!res.ok) {
@@ -813,12 +889,17 @@ async function main() {
     if (line === "/help") {
       console.log(
         dim(
-          "  /exit — quit · /revert [runId] — undo the agent's file edits (latest run) · " +
+          "  /exit — quit · /model — switch the coding model for this session · " +
+            "/revert [runId] — undo the agent's file edits (latest run) · " +
             "/reset — forget this folder's saved conversation and start fresh"
         )
       );
       continue;
     }
+    if (line === "/model") {
+      await chooseModel();
+      continue;
+    }
     if (line === "/reset" || line === "/new") {
       history.length = 0;
       await clearSession(cwd);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiens.nguyen/gonext-local-worker",
-  "version": "1.0.205",
+  "version": "1.0.206",
   "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",