@tiens.nguyen/gu-cli 1.0.873 → 1.0.875

package/agent-model-command.mjs

@@ -350,8 +350,8 @@ export function agentRecoveryOptions({ keyProblem = false, host = "" } = {}) {
   return [
     { action: "leave", label: "leave it" },
     ...(keyProblem ? [{ action: "key", label: `enter the API key for ${host || "this provider"}…` }] : []),
-    { action: "ollama", label: "use an Ollama server for the agent…" },
-    { action: "provider", label: "add a provider for the agent — Claude API, OpenAI, Kimi…" },
+    // #244: no separate Ollama row — Ollama is one of the providers on the add screen.
+    { action: "provider", label: "add a provider for the agent — Claude API, OpenAI, Kimi, Ollama…" },
   ];
 }
 
@@ -375,8 +375,9 @@ export function agentMenuOptions({ host = "", modelCount = 0, needsKey = false }
   const where = host || "this server";
   return [
     { action: "models", label: `pick another model on ${where}${modelCount ? `  (${modelCount})` : ""}` },
-    { action: "ollama", label: "use an Ollama server for the agent…" },
-    { action: "provider", label: "use another provider — Claude API, OpenAI, Kimi…" },
+    // #244: "use an Ollama server for the agent…" was a second way to the same add screen, left from
+    // when Ollama had its own setup (#41). Ollama, ollama.com included, is one provider among the rest.
+    { action: "provider", label: "use another provider — Claude API, OpenAI, Kimi, Ollama…" },
     // Task #245: the SAME key manager /model's ctrl+k and /embed-model open — switch, add, make
     // default, delete — for this server's keys only.
     ...(needsKey ? [{ action: "key", label: `API keys for ${where} — switch, add or remove…` }] : []),
@@ -388,12 +389,61 @@ export function agentMenuOptions({ host = "", modelCount = 0, needsKey = false }
  * rows, current one marked — then "add a new one". Switching to a provider you already set up must
  * not mean typing its URL again. Returns [{ action: "row" | "add", row?, label }].
  */
+/**
+ * EVERY MODEL ALREADY SET UP ON THIS MACHINE, for the agent (#246) — what /model offers the coder.
+ *
+ * Reported: /model listed several providers and models, and /agent-model showed none of them —
+ * it asked to add a provider or enter a key. They are the same servers with the same keys: a key
+ * belongs to its server, and the API finds the agent's key by matching agentModelUrl to a saved row
+ * (resolveKeyForRole). So a pick here saves that row's URL and the model, and its key comes along.
+ *
+ * Rows: OpenAI-compatible and Ollama, enabled, with a URL; their default model, then the extra
+ * models offered in /model. Embedding models are left out — as the agent they answer nonsense.
+ * → [{ rowId, kind, url, host, model, label, current }]
+ */
+export function savedAgentChoices(settings) {
+  const rows = Array.isArray(settings?.agentCodingBackends) ? settings.agentCodingBackends : [];
+  const currentUrl = ensureV1Root(String(settings?.agentModelUrl ?? "")).toLowerCase();
+  const currentModel = currentAgentModel(settings);
+  const out = [];
+  const seen = new Set();
+  const takenKinds = new Set();
+  for (const row of rows) {
+    if (!row || typeof row !== "object" || row.enabled === false) continue;
+    const kind = String(row.kind ?? "");
+    if (kind !== "openai" && kind !== "ollama") continue;
+    const url = ensureV1Root(String(row.url ?? "").trim());
+    if (!url) continue;
+    // The API's identity rule: an id-less row is known by its kind, the first of each only.
+    const rowId = String(row.id ?? "").trim() || (takenKinds.has(kind) ? "" : kind);
+    if (!rowId) continue;
+    takenKinds.add(kind);
+    let host = url;
+    try {
+      host = new URL(url).host;
+    } catch {
+      /* keep the url */
+    }
+    const models = [String(row.defaultModel ?? "").trim(), ...(Array.isArray(row.models) ? row.models.map((m) => String(m ?? "").trim()) : [])];
+    for (const model of models) {
+      if (!model || isLikelyEmbedder({ model })) continue;
+      const key = `${url.toLowerCase()}\n${model}`;
+      if (seen.has(key)) continue;
+      seen.add(key);
+      const current = url.toLowerCase() === currentUrl && model === currentModel;
+      out.push({ rowId, kind, url, host, model, label: `${host}/${model}`, current });
+    }
+  }
+  return out;
+}
+
 export function agentProviderChoices(settings) {
   const rows = Array.isArray(settings?.agentCodingBackends) ? settings.agentCodingBackends : [];
   const current = ensureV1Root(String(settings?.agentModelUrl ?? "")).toLowerCase();
   const out = [];
   for (const row of rows) {
-    if (!row || row.kind !== "openai" || !row.url) continue;
+    // Ollama rows too (#244): ollama.com or a box saved earlier is a provider like any other.
+    if (!row || (row.kind !== "openai" && row.kind !== "ollama") || !row.url) continue;
     let host = String(row.url);
     try {
       host = new URL(String(row.url)).host;
@@ -403,7 +453,7 @@ export function agentProviderChoices(settings) {
     const isCurrent = ensureV1Root(String(row.url)).toLowerCase() === current;
     out.push({ action: "row", row, label: `${host}${isCurrent ? "  (current)" : ""}` });
   }
-  out.push({ action: "add", label: "add a new provider — Claude API, OpenAI, Kimi…" });
+  out.push({ action: "add", label: "add a new provider — Claude API, OpenAI, Kimi, Ollama…" });
   return out;
 }
 

package/embed-model-command.mjs

@@ -106,8 +106,8 @@ export function embedMenuOptions({ current = DEFAULT_EMBED_MODEL, host = "", key
   const where = host || "this server";
   return [
     { action: "model", label: `switch the embedder model  (currently ${current})` },
-    { action: "ollama", label: "use an Ollama server for embedding…" },
-    { action: "provider", label: "use another OpenAI-compatible provider…" },
+    // #244: one row for every server — Ollama (this machine, the network) is a choice on the add screen.
+    { action: "provider", label: "use another provider — OpenAI, Ollama…" },
     ...(keyRequired ? [{ action: "key", label: `enter a new API key for ${where}…` }] : []),
   ];
 }

package/gu-repl.mjs

@@ -543,6 +543,9 @@ const color256 = (n, s) => `\x1b[38;5;${n}m${s}\x1b[0m`;
 // WAIT_COLORS_256 colour only when this run was started with `gu --color`. White goes through
 // paint(), so NO_COLOR / TERM=dumb still turn it off.
 const waitPaint = (n, s) => (REPL_COLOR ? color256(n, s) : white(s));
+// The live thought line above it (#245) follows the same switch: white by default, the accent
+// cyan only with `gu --color`.
+const thoughtPaint = (s) => (REPL_COLOR ? accent(s) : white(s));
 const orange = (s) => color256(208, s); // peak-token label (task #84)
 // Compact token count: 950 → "950", 12300 → "12.3k", 1500000 → "1.5M" (task #83).
 const fmtTokens = (n) => {
@@ -636,7 +639,7 @@ if (cliVerdict.kind === "run") {
       "  -v, --verbose   show RAG index-file activity in the bottom bar (each shard written)\n" +
       "  --log           log this run in detail (~/.gonext/worker.log). Without it only starts,\n" +
       "                  stops and errors are logged. Ends when this gu closes.\n" +
-      "  --color         colour the thinking spinner and its playful word (cycling colours).\n" +
+      "  --color         colour the live thought, the thinking spinner and its playful word.\n" +
       "                  Without it they are plain white.\n" +
       "  --name <text>   show <text> instead of the gu logo in the startup banner\n" +
       "                  (this run only; letters, digits, space - _ are drawn as art)\n" +
@@ -6642,6 +6645,8 @@ async function addServer({ role = "coding", dialect = "openai", url: presetUrl =
   for (;;) {
     // --- 1. the server --------------------------------------------------------------------
     target = null;
+    // The dialect the URL prompt reads the address in: the caller's, or Ollama's when that row was picked.
+    let askDialect = dialect;
     if (presetUrl) {
       target = as.serverTarget(dialect, presetUrl);
       if (!target.ok) {
@@ -6653,7 +6658,14 @@ async function addServer({ role = "coding", dialect = "openai", url: presetUrl =
       // the one add flow, so every role that adds a provider offers the same list (the embedder
       // without Claude, which serves no embeddings).
       const presets = ps.providerPresets(role);
-      const rowsShown = [...presets.map((p) => `${p.name}${dim(`  · ${p.url}`)}`), "another URL…"];
+      // #244: an Ollama on this machine or the network is a row here, like the hosted ones — the
+      // menus no longer have a separate "use an Ollama server…" entrance. It asks for the address.
+      const OLLAMA_HERE = presets.length;
+      const rowsShown = [
+        ...presets.map((p) => `${p.name}${dim(`  · ${p.url}`)}`),
+        `Ollama — this machine or your network…${dim("  · e.g. 127.0.0.1:11434")}`,
+        "another URL…",
+      ];
       // A preset names its own dialect (Ollama cloud is "ollama"); the rest are OpenAI-compatible.
       console.log(dim(`   Which provider?  (${PICK_HINT})`));
       const pi = await pickFromList(rowsShown, 0, { back });
@@ -6663,11 +6675,12 @@ async function addServer({ role = "coding", dialect = "openai", url: presetUrl =
         return;
       }
       if (pi < presets.length) target = as.serverTarget(presets[pi].dialect ?? "openai", presets[pi].url);
+      else if (pi === OLLAMA_HERE) askDialect = "ollama";
     }
     if (!target && !presetUrl) {
-      const dflt = as.defaultServerUrl(role, dialect, settings, existing);
-      console.log(dim(dialect === "ollama"
-        ? "   An Ollama server: this machine, a box on your network, or https://ollama.com (needs a key)."
+      const dflt = as.defaultServerUrl(role, askDialect, settings, existing);
+      console.log(dim(askDialect === "ollama"
+        ? "   An Ollama server: this machine, a box on your network, or ollama.com (with a key)."
         : "   Any OpenAI-compatible endpoint: OpenRouter, Together, vLLM, LM Studio, …"));
       // The address goes ABOVE, in full, and Enter's meaning is said out loud (46d1524e).
       console.log(dim(dflt ? `  Enter uses ${dflt}  ·  Esc cancels\n` : "  Esc cancels.\n"));
@@ -6676,9 +6689,9 @@ async function addServer({ role = "coding", dialect = "openai", url: presetUrl =
         if (typed === null || as.isCancelWord(typed)) break;
         const raw = typed.trim() || dflt;
         if (!raw) break;
-        const t = as.serverTarget(dialect, raw);
+        const t = as.serverTarget(askDialect, raw);
         if (!t.ok) {
-          if (dialect === "openai") for (const l of ps.describeUrlProblem(t.why, raw)) console.log(GUTTER + dim(l));
+          if (askDialect === "openai") for (const l of ps.describeUrlProblem(t.why, raw)) console.log(GUTTER + dim(l));
           else console.log(GUTTER + dim(`"${raw}" is ${t.why} — try host:port, e.g. 127.0.0.1:11434`));
           continue;
         }
@@ -8805,8 +8818,8 @@ async function chooseEmbedModel() {
     await addServer({ role: "embed", dialect: "ollama", model: chosen, plan });
     return;
   }
-  if (action === "ollama" || action === "provider") {
-    if ((await addServer({ role: "embed", dialect: action === "ollama" ? "ollama" : "openai", back: true })) === BACK) return chooseEmbedModel();
+  if (action === "provider") {
+    if ((await addServer({ role: "embed", dialect: "openai", back: true })) === BACK) return chooseEmbedModel();
     return;
   }
 
@@ -8944,6 +8957,78 @@ async function chooseAgentProvider({ back = false } = {}) {
  * The decisions live in agent-model-command.mjs; this is the terminal half only.
  */
 async function chooseAgentModel() {
+  const am = await import("./agent-model-command.mjs");
+  const settings = await withPendingStatus("reading your agent model", fetchWorkerSettings);
+  if (!settings) {
+    console.log(red("   couldn't reach the API to read your current agent model.\n"));
+    return;
+  }
+  // WHAT /model ALREADY HAS, FIRST (#246). The providers and models added for the coder are the
+  // same servers with the same keys; listing them here means switching the agent to one of them is
+  // one pick, not "add a provider" or "enter the API key" for something already set up.
+  const saved = am.savedAgentChoices(settings);
+  if (!saved.length) return agentModelOnServer();
+  const current = am.currentAgentModel(settings);
+  const curHost = settings.agentModelUrl ? hostOf(String(settings.agentModelUrl)) : "";
+  for (;;) {
+    const labels = saved.map((c) => c.label + (c.current ? accent("  (current)") : ""));
+    const MORE = labels.length;
+    labels.push("other models, providers and API keys…");
+    const ADD = labels.length;
+    labels.push("add a provider — Claude API, OpenAI, Kimi, Ollama…");
+    // The hint drops to its own row when the header would wrap (80 columns over SSH).
+    const head = `   Agent model — the one that TALKS (now ${current || "unset"}${curHost ? ` on ${curHost}` : ""})`;
+    const hint = `(${PICK_HINT}):`;
+    const cols = Math.max(20, process.stdout.columns || 80);
+    console.log(dim(head.length + 2 + hint.length < cols ? `${head}  ${hint}` : `${head}\n   ${hint}`));
+    const idx = await pickFromList(labels, Math.max(0, saved.findIndex((c) => c.current)));
+    if (idx < 0) {
+      console.log(dim("   (cancelled)\n"));
+      return;
+    }
+    if (idx === MORE) {
+      if ((await agentModelOnServer({ back: true })) === BACK) continue;
+      return;
+    }
+    if (idx === ADD) {
+      if ((await addServer({ role: "agent", dialect: "openai", back: true })) === BACK) continue;
+      return;
+    }
+    await useSavedAgentChoice(saved[idx]);
+    return;
+  }
+}
+
+/**
+ * Point the agent at a model already set up for /model (#246): that row's URL and the model. The
+ * API finds the key by the URL (resolveKeyForRole), so nothing is re-entered — unless the server
+ * needs a key and has none, which is asked here rather than failing on the next question.
+ */
+async function useSavedAgentChoice(c) {
+  if (c.current) {
+    console.log(dim(`   already using ${c.model} on ${c.host} — nothing changed.\n`));
+    return;
+  }
+  const as = await import("./add-server.mjs");
+  const target = as.serverTarget(c.kind === "ollama" ? "ollama" : "openai", c.url);
+  const keys = (await withPendingStatus(`reading the keys for ${c.host}`, () => fetchServerKeys(c.rowId))) ?? [];
+  if (target.ok && as.keyRequirement(target) === "required" && !keys.length) {
+    console.log(GUTTER + dim(`${c.host} needs an API key and none is saved for it — enter one.`));
+    await addServer({ role: "agent", dialect: target.dialect, url: c.url, newKey: true });
+    return;
+  }
+  // A folder pick of ANOTHER server's key would be sent to this one: this folder follows the
+  // server's own key from now on.
+  if (sessionAgentKeyId && !keys.some((k) => k.id === sessionAgentKeyId)) await saveFolderKeyPick("agent", "");
+  await saveAgentModel(c.url, c.model, c.host);
+}
+
+/**
+ * One server's view: its full model list, other providers, its keys — what /agent-model was before
+ * #246, now behind "other models, providers and API keys…" when something is already set up.
+ * `back`: opened from that list, so its first menu offers "← back".
+ */
+async function agentModelOnServer({ back = false } = {}) {
   const am = await import("./agent-model-command.mjs");
   const settings = await withPendingStatus("reading your agent model", fetchWorkerSettings);
   if (!settings) {
@@ -9020,11 +9105,12 @@ async function chooseAgentModel() {
     const options = am.agentRecoveryOptions({ keyProblem, host: listHost });
     const go = await pickFromList(
       options.map((o, i) => (i === 0 ? dim(o.label) : o.label)),
-      keyProblem ? 1 : 0
+      keyProblem ? 1 : 0,
+      { back }
     );
+    if (go === PICK_BACK) return BACK;
     const action = go >= 0 ? options[go].action : "leave";
     if (action === "key") await enterAgentProviderKey(source.row, listHost);
-    else if (action === "ollama") await addServer({ role: "agent", dialect: "ollama" });
     else if (action === "provider") await addServer({ role: "agent", dialect: "openai" });
     else console.log("");
     return;
@@ -9035,19 +9121,16 @@ async function chooseAgentModel() {
   // are all you can choose". A short menu of what to change comes before any list.
   console.log(dim(`   Agent model — the one that TALKS: ${current || "unset"} on ${listHost}`));
   const menu = am.agentMenuOptions({ host: listHost, modelCount: models.length, needsKey });
-  const first = await pickFromList(menu.map((o) => o.label), 0);
+  const first = await pickFromList(menu.map((o) => o.label), 0, { back });
+  if (first === PICK_BACK) return BACK;
   const firstAction = first >= 0 ? menu[first].action : "cancel";
   if (firstAction === "cancel") {
     console.log(dim("   (cancelled)\n"));
     return;
   }
   // "← back" in any of these returns to this menu (#242) — read again, as on a first open.
-  if (firstAction === "ollama") {
-    if ((await addServer({ role: "agent", dialect: "ollama", back: true })) === BACK) return chooseAgentModel();
-    return;
-  }
   if (firstAction === "provider") {
-    if ((await chooseAgentProvider({ back: true })) === BACK) return chooseAgentModel();
+    if ((await chooseAgentProvider({ back: true })) === BACK) return agentModelOnServer({ back });
     return;
   }
   if (firstAction === "key") {
@@ -9055,7 +9138,7 @@ async function chooseAgentModel() {
     const rowId = String(source.row?.id ?? "").trim() || String(source.row?.kind ?? "").trim();
     if (!rowId) return enterAgentProviderKey(source.row, listHost);
     const r = await manageServerKeys({ role: "agent", backendId: rowId, host: listHost, model: am.currentAgentModel(settings), back: true });
-    if (r === BACK) return chooseAgentModel();
+    if (r === BACK) return agentModelOnServer({ back });
     return;
   }
 
@@ -9088,7 +9171,7 @@ async function chooseAgentModel() {
   // No other entries here any more (#44): the ways to change the provider are on the menu above,
   // where they can be seen, rather than after every model.
   const idx = await pickFromList(labels, Math.max(0, rows.findIndex((r) => r.current)), { back: true });
-  if (idx === PICK_BACK) return chooseAgentModel();
+  if (idx === PICK_BACK) return agentModelOnServer({ back });
   if (idx < 0) {
     console.log(dim("   (cancelled)\n"));
     return;
@@ -12493,7 +12576,8 @@ async function runAgentTurn(history, { attachJobId = "", ownJob = false } = {})
     // Only worth probing when the mouse is captured — thoughtRow exists for hover/click
     // hit-testing, and with `/hover` off nothing reads it (and nothing would consume the reply).
     if (hoverMouse && process.stdin.isTTY && now - lastDsrAt > 400) { lastDsrAt = now; ui("\x1b[6n"); }
-    // Line 1: cyan ● + cyan phase/seconds — the logo's colour, which is what this REPL uses for
+    // Line 1 (#245: white by default; the cyan below only with `gu --color`, like the spinner row):
+    // cyan ● + cyan phase/seconds — the logo's colour, which is what this REPL uses for
     // WHAT IS LIVE RIGHT NOW: the ">> " prompt is cyan while you are typing and goes dim once
     // submitted, and a selection is drawn on the same cyan. Green is left to mean SUCCEEDED (the
     // ✓ bullets), which is what it means everywhere else here — a running step is not a finished
@@ -12512,7 +12596,7 @@ async function runAgentTurn(history, { attachJobId = "", ownJob = false } = {})
     const clipRow = (r) => clipVisible(r, Math.max(20, process.stdout.columns || 80) - 1);
     const contentRows = [];
     for (let i = 0; i < LEAD_ROWS; i++) contentRows.push(""); // the separator — see LEAD_ROWS
-    contentRows.push(clipRow(`${markLine(accent(BULLET))}${hovering ? hoverThought(label) : accent(label)}`));
+    contentRows.push(clipRow(`${markLine(thoughtPaint(BULLET))}${hovering ? hoverThought(label) : thoughtPaint(label)}`));
     for (const dr of detailRows) contentRows.push(clipRow(dim(dr)));
     for (const wl of expLines) contentRows.push(clipRow(GUTTER + dim(wl)));
     for (let i = 0; i < WORD_GAP; i++) contentRows.push(""); // see WORD_GAP
@@ -14182,7 +14266,7 @@ async function main() {
           red("gu: this machine has no models set up yet.") +
             dim(
               "\n  Nothing can run until it knows which Ollama server to use." +
-                "\n  Start `gu`, then `/agent-model` → use an Ollama server, `/embed-model` for RAG," +
+                "\n  Start `gu`, then `/agent-model` → use another provider → Ollama, `/embed-model` for RAG," +
                 "\n  and `/model` → add a provider → another URL… for the coding model." +
                 `\n  The agent server needs a ${REQUIRED_AGENT_PREFIX} model (e.g. ${REQUIRED_AGENT_MODEL})` +
                 `\n  and any ONE of these embedders: ${ACCEPTED_EMBED_MODELS.map((m) => m.replace(/:latest$/, "")).join(", ")};` +

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tiens.nguyen/gu-cli",
-  "version": "1.0.873",
+  "version": "1.0.875",
   "description": "Gu CLI — the gu terminal plus the local worker that runs agent / OCR / PDF / embedding jobs on your Mac (Ollama / MLX / OpenAI-compatible).",
   "type": "module",
   "license": "MIT",