@tiens.nguyen/gonext-local-worker 1.0.334 → 1.0.337

package/gonext-repl.mjs

@@ -688,15 +688,33 @@ let wsSync = { enabled: false, name: "", path: "" };
 const workspaceKeyFor = (cwd) =>
   createHash("sha256").update(cwd).digest("hex").slice(0, 32);
 
-// Task #104: agent CODE-model OUTPUT-token counters. The API keys the global total on
-// (userId + coding-model URL from settings) and the workspace total on workspaceKey; the
-// REPL just supplies workspaceKey + this turn's delta. All best-effort (never block a turn).
+// Task #104: agent CODE-model OUTPUT-token counters. The API keys the workspace total on
+// workspaceKey and the GLOBAL total on (userId + coding-model URL). Task #125: the REPL
+// now supplies the URL of the backend in EFFECT this session (activeCodingUrl) so the
+// global total follows a /model backend switch; empty → the API falls back to the
+// account-default URL from settings (older behaviour). All best-effort (never block a turn).
+const activeCodingUrl = () => sessionCodingUrl || defaultCodingUrl;
+
+// Task #126: the footer's global "↓" for the CURRENT backend. The saved-full-responses sum
+// (savedResponsesTotal) is authoritative per-URL — and correct even for turns before #125,
+// because each saved row carries its own url — so PREFER it when it has data. Fall back to
+// the #125 incremental counter (globalTotal) when Save-Full-Response is off (no saved rows,
+// so savedResponsesTotal is 0), which keeps a correct non-zero number instead of a
+// surprising 0. Returns null when the response had neither, so callers keep the old value.
+function pickGlobalTotal(t) {
+  if (!t) return null;
+  const saved = Number.isFinite(t.savedResponsesTotal) ? t.savedResponsesTotal : 0;
+  if (saved > 0) return saved;
+  return Number.isFinite(t.globalTotal) ? t.globalTotal : null;
+}
+
 async function fetchOutputTokenTotals(cwd) {
   try {
-    const r = await fetch(
-      `${apiBase}/api/worker/output-tokens?workspaceKey=${workspaceKeyFor(cwd)}`,
-      { headers: { "X-Worker-Key": workerKey } }
-    );
+    const qs = new URLSearchParams({ workspaceKey: workspaceKeyFor(cwd) });
+    if (activeCodingUrl()) qs.set("codingUrl", activeCodingUrl());
+    const r = await fetch(`${apiBase}/api/worker/output-tokens?${qs.toString()}`, {
+      headers: { "X-Worker-Key": workerKey },
+    });
     if (!r.ok) return null;
     return await r.json();
   } catch {
@@ -710,7 +728,11 @@ async function addOutputTokens(cwd, delta) {
     const r = await fetch(`${apiBase}/api/worker/output-tokens`, {
       method: "POST",
       headers: { "Content-Type": "application/json", "X-Worker-Key": workerKey },
-      body: JSON.stringify({ workspaceKey: workspaceKeyFor(cwd), delta }),
+      body: JSON.stringify({
+        workspaceKey: workspaceKeyFor(cwd),
+        delta,
+        codingUrl: activeCodingUrl(),
+      }),
     });
     if (!r.ok) return null;
     return await r.json();
@@ -724,6 +746,7 @@ async function resetGlobalOutputTokens() {
     const r = await fetch(`${apiBase}/api/worker/output-tokens/reset-global`, {
       method: "POST",
       headers: { "Content-Type": "application/json", "X-Worker-Key": workerKey },
+      body: JSON.stringify({ codingUrl: activeCodingUrl() }),
     });
     return r.ok;
   } catch {
@@ -918,10 +941,20 @@ async function chooseModel() {
   // Remember the friendly name for the token footer, which should never show a raw
   // "<kind>::<model>" wire id.
   sessionCodingLabel = chosen.id === def ? "" : chosen.model;
+  // Task #125: remember the chosen backend's coding URL so the GLOBAL output-token total is
+  // keyed to it. Default pick → "" so activeCodingUrl() falls back to defaultCodingUrl.
+  sessionCodingUrl = chosen.id === def ? "" : (chosen.url ?? "");
   const where = showKind ? ` (${KIND_LABEL[chosen.kind ?? ""] ?? chosen.kind})` : "";
   console.log(
     green(`  ✓ coding model → ${chosen.model}${where}${chosen.id === def ? " (default)" : ""}\n`)
   );
+  // Task #125: the GLOBAL lifetime total is per-backend, so re-fetch it for the newly
+  // selected backend NOW — otherwise the footer would keep showing the previous backend's
+  // total until the next turn. (The live per-turn tail, latestOutputTokens, is local to
+  // runAgentTurn and is already 0 here between turns — /model can't run mid-turn — so there
+  // is nothing to zero, and it isn't in scope from here.)
+  const t = await fetchOutputTokenTotals(resolve(process.cwd()));
+  { const g = pickGlobalTotal(t); if (g !== null) globalOutputTokens = g; }
 }
 
 // Probe whether SSH KEY auth already works for a server (so we can tell the user to run
@@ -1282,6 +1315,15 @@ let sessionCodingLabel = "";
 // per-turn token footer next to the GLOBAL output total so it's clear which model that
 // lifetime count belongs to. The /model override (sessionCodingModel) wins when set.
 let defaultCodingModel = "";
+// Task #125: the coding-server URL of the backend in effect this session — the /model
+// pick's backend when one is chosen, else the account default. Sent to the output-token
+// endpoints so the GLOBAL lifetime total is keyed to the RIGHT backend (a /model switch to
+// another backend must not keep showing the previous backend's total). Empty = let the API
+// fall back to the account-default URL from settings (older behaviour).
+let sessionCodingUrl = "";
+// The account-default backend's URL (from the startup probe's codingChoices), used when no
+// /model override is active. Set once at startup alongside defaultCodingModel.
+let defaultCodingUrl = "";
 // Auto-test mode (/test-auto): when on, the agent verifies its work (build/run/curl/…)
 // and fixes → re-tests until it passes before finishing. Default off; remembered per
 // folder in the session file; sent to /agent-ask as autoTest so python drives the loop.
@@ -2008,7 +2050,7 @@ async function runAgentTurn(history) {
         if (latestOutputTokens > 0 && !outputTokensPosted) {
           outputTokensPosted = true;
           void addOutputTokens(resolve(process.cwd()), latestOutputTokens).then((t) => {
-            if (t && Number.isFinite(t.globalTotal)) globalOutputTokens = t.globalTotal;
+            { const g = pickGlobalTotal(t); if (g !== null) globalOutputTokens = g; }
             // This turn's total is now folded into globalOutputTokens — zero the live
             // per-turn tally so the ↓ (global + latest) doesn't double-count it.
             latestOutputTokens = 0;
@@ -2166,14 +2208,17 @@ async function main() {
   // Task #104: seed the ↓ global OUTPUT-token total (user + coding-URL) from the API.
   {
     const t = await fetchOutputTokenTotals(resolve(process.cwd()));
-    if (t && Number.isFinite(t.globalTotal)) globalOutputTokens = t.globalTotal;
+    { const g = pickGlobalTotal(t); if (g !== null) globalOutputTokens = g; }
   }
   // Show which models will answer, straight from user settings (also validates auth).
   try {
     const probe = await fetchAgentPayload([{ role: "user", content: "ping" }]);
     const p = probe?.payload ?? {};
-    // Remember the default coder id for the token footer's global-total label.
+    // Remember the default coder id for the token footer's global-total label, and the
+    // default backend's coding URL so the GLOBAL output-token total is keyed to it (task
+    // #125) when no /model override is active.
     if (p.codingModelId) defaultCodingModel = p.codingModelId;
+    if (p.codingBaseURL) defaultCodingUrl = p.codingBaseURL;
     // RAG storage choice (task #97): ask ONCE per folder, and only when RAG is actually
     // enabled in Settings. Default Yes = LOCAL (kept on this machine, no S3 needed).
     if (p.ragEnabled && sessionRagMode === null) {
@@ -2322,9 +2367,11 @@ async function main() {
     if (line === "/output-token") {
       // Task #104: print the OUTPUT-token total for THIS workspace (and the global one).
       const t = await fetchOutputTokenTotals(cwd);
-      if (t && Number.isFinite(t.globalTotal)) globalOutputTokens = t.globalTotal;
+      { const g = pickGlobalTotal(t); if (g !== null) globalOutputTokens = g; }
       const ws = t && Number.isFinite(t.workspaceTotal) ? t.workspaceTotal : 0;
-      const gl = t && Number.isFinite(t.globalTotal) ? t.globalTotal : globalOutputTokens;
+      // Task #126: same saved-else-counter rule as the footer (pickGlobalTotal), not the
+      // raw counter — so /output-token and the live ↓ footer agree.
+      const gl = pickGlobalTotal(t) ?? globalOutputTokens;
       console.log(
         cyan(`↓ output tokens`) +
           dim(` — this workspace: `) +

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiens.nguyen/gonext-local-worker",
-  "version": "1.0.334",
+  "version": "1.0.337",
   "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",