loom-agent 1.3.3 → 1.6.0

package/src/core/update.js

@@ -1,6 +1,116 @@
-async function updateCheck() {
-  if (process.env.LOOM_NO_UPDATE_CHECK) return;
-  return;
-}
-
-module.exports = { updateCheck };
+// Startup update check — a quiet, non-blocking "is there a newer version of
+// loom-agent on npm?" probe that runs once per day and prints at most one line
+// to stderr before the TUI/REPL takes over the terminal.
+//
+// Design constraints:
+//   - Non-blocking: runs concurrently with `main()`, never gates startup, and
+//     any failure (offline, registry down, malformed body) is swallowed — this
+//     must never break the app or add latency to first render.
+//   - Throttled: one network probe per 24h, tracked by a stamp file so it
+//     survives across runs.
+//   - Silent when up-to-date; silent when disabled via LOOM_NO_UPDATE_CHECK.
+//     Only prints when a strictly newer version is actually available.
+//   - Compare is a plain semver-ish tuple compare (major.minor.patch), not a
+//     library — enough for the "newer than installed?" question without a dep.
+
+const fs = require('fs');
+const path = require('path');
+const os = require('os');
+
+const PKG = require('../../package.json');
+const CHECK_INTERVAL_MS = 24 * 60 * 60 * 1000; // 24h between probes
+const FETCH_TIMEOUT_MS = 4000; // fail fast — this is best-effort only
+
+function stampFile() {
+  const dir = process.env.LOOM_CONFIG_DIR || path.join(os.homedir(), '.loom');
+  return path.join(dir, 'update-check.json');
+}
+
+// Extract a comparable version tuple. Returns null for anything unparseable.
+function parseVersion(v) {
+  if (typeof v !== 'string') return null;
+  const m = v.trim().match(/^v?(\d+)\.(\d+)\.(\d+)/);
+  if (!m) return null;
+  return [Number(m[1]), Number(m[2]), Number(m[3])];
+}
+
+// True when `latest` is strictly newer than `current`. Pre-release / build
+// suffixes are ignored (treated as the base triple) — good enough here.
+function isNewer(latest, current) {
+  const a = parseVersion(latest);
+  const b = parseVersion(current);
+  if (!a || !b) return false;
+  for (let i = 0; i < 3; i++) {
+    if (a[i] > b[i]) return true;
+    if (a[i] < b[i]) return false;
+  }
+  return false;
+}
+
+// Read the last-check timestamp; returns 0 if absent/corrupt (so the very
+// first run after install always checks).
+function lastCheckAt() {
+  try {
+    const raw = fs.readFileSync(stampFile(), 'utf8');
+    const n = Number(JSON.parse(raw).at);
+    return Number.isFinite(n) ? n : 0;
+  } catch {
+    return 0;
+  }
+}
+
+function markChecked() {
+  try {
+    const file = stampFile();
+    fs.mkdirSync(path.dirname(file), { recursive: true });
+    fs.writeFileSync(file, JSON.stringify({ at: Date.now(), checked: PKG.version }));
+  } catch {
+    // Throttle stamp is best-effort — a failure just means we probe again
+    // next run, never breaks startup.
+  }
+}
+
+/**
+ * Probe npm for the latest `loom-agent` version and, if it's newer than the
+ * installed one, print a one-line upgrade hint to stderr. Throttled to one
+ * probe per 24h. Never throws, never blocks, and returns the newer version
+ * string (or null) so tests can assert the decision without inspecting
+ * stderr.
+ *
+ * @returns {Promise<string|null>} latest version when an upgrade exists, else null
+ */
+async function updateCheck() {
+  if (process.env.LOOM_NO_UPDATE_CHECK) return null;
+
+  const now = Date.now();
+  if (now - lastCheckAt() < CHECK_INTERVAL_MS) return null;
+
+  // Claim the slot BEFORE the fetch: even if the network call fails we don't
+  // want to retry on every startup when offline.
+  markChecked();
+
+  try {
+    const ctrl = new AbortController();
+    const timer = setTimeout(() => ctrl.abort(), FETCH_TIMEOUT_MS);
+    let res;
+    try {
+      res = await fetch('https://registry.npmjs.org/loom-agent/latest', { signal: ctrl.signal });
+    } finally {
+      clearTimeout(timer);
+    }
+    if (!res || !res.ok) return null;
+    const body = /** @type {{version?: unknown}} */ (await res.json());
+    const latest = typeof body?.version === 'string' ? body.version : null;
+    if (!isNewer(latest, PKG.version)) return null;
+
+    process.stderr.write(
+      `\n[loom] New version available: ${PKG.version} → ${latest}\n` +
+      `[loom] Upgrade with: npm install -g ${PKG.name}@latest\n\n`
+    );
+    return latest;
+  } catch {
+    return null;
+  }
+}
+
+module.exports = { updateCheck, isNewer, parseVersion };

package/src/mcp/connector-presets.js

@@ -0,0 +1,105 @@
+// Hosting / cloud-service connector presets — platforms you "connect" your
+// project to rather than dev-tool servers. Surfaced under /connectors (and
+// `loom add connector`), not /mcp. Lives in the core layer so the CLI
+// (`loom add connector <preset> -e KEY=V`) and the TUI picker share one
+// catalog; the TUI re-exports this list from src/tui/mcp-presets.ts.
+//
+// Each preset: id, label, package (npx-run) or command (full override, e.g.
+// docker / npx mcp-remote), description, args (may reference $KEY env
+// placeholders resolved at add time), env defaults, prompts (guided key
+// entry + the required-key set for CLI expansion), and an optional
+// optionalArgsPrompt (non-secret trailing arg, e.g. --project-ref).
+
+/**
+ * @typedef {Object} ConnectorPrompt
+ * @property {string} key
+ * @property {string} label
+ * @property {boolean} [mask]
+ */
+
+/**
+ * @typedef {Object} OptionalArgsPrompt
+ * @property {string|null} flag
+ * @property {string} label
+ */
+
+/**
+ * @typedef {Object} ConnectorPreset
+ * @property {string} id
+ * @property {string} label
+ * @property {string} [package]
+ * @property {string} [command]
+ * @property {string} description
+ * @property {string[]} args
+ * @property {Record<string, string>} env
+ * @property {ConnectorPrompt[]} prompts
+ * @property {OptionalArgsPrompt} [optionalArgsPrompt]
+ */
+
+/** @type {ConnectorPreset[]} */
+const CONNECTOR_PRESETS = [
+  {
+    id: 'supabase',
+    label: 'Supabase',
+    package: '@supabase/mcp-server-supabase',
+    description: 'Query schema, run SQL, manage Supabase projects',
+    args: ['--access-token', '$SUPABASE_ACCESS_TOKEN'],
+    env: { SUPABASE_ACCESS_TOKEN: '' },
+    prompts: [
+      { key: 'SUPABASE_ACCESS_TOKEN', label: 'Personal access token (supabase.com/dashboard/account/tokens)', mask: true },
+    ],
+    optionalArgsPrompt: { flag: '--project-ref', label: 'Project ref (optional, e.g. abcdefghijklmnop) — Enter to skip' },
+  },
+  {
+    id: 'nextjs',
+    label: 'Next.js',
+    package: 'nextjs-mcp-server',
+    description: 'Next.js project introspection (routes, pages, config)',
+    args: [],
+    env: {},
+    prompts: [],
+  },
+  {
+    id: 'railway',
+    label: 'Railway',
+    package: '@railway/mcp-server',
+    description: 'Railway deploys, services, variables',
+    args: [],
+    env: { RAILWAY_API_TOKEN: '' },
+    prompts: [
+      { key: 'RAILWAY_API_TOKEN', label: 'API token (railway.app/account/tokens)', mask: true },
+    ],
+  },
+  {
+    id: 'vercel',
+    label: 'Vercel',
+    command: 'npx',
+    args: ['-y', 'mcp-remote', 'https://mcp.vercel.com'],
+    description: 'Vercel deployments and projects (remote MCP endpoint, OAuth)',
+    env: {},
+    prompts: [],
+  },
+  {
+    id: 'netlify',
+    label: 'Netlify',
+    command: 'npx',
+    args: ['-y', 'mcp-remote', 'https://api.netlify.com/mcp'],
+    description: 'Netlify sites, deploys, and environment vars (remote MCP endpoint, OAuth)',
+    env: {},
+    prompts: [],
+  },
+  {
+    id: 'cloudflare',
+    label: 'Cloudflare',
+    package: '@cloudflare/mcp-server-cloudflare',
+    description: 'Workers, DNS, and account management',
+    args: [],
+    env: { CLOUDFLARE_API_TOKEN: '', CLOUDFLARE_ACCOUNT_ID: '' },
+    prompts: [
+      { key: 'CLOUDFLARE_API_TOKEN', label: 'API token (dash.cloudflare.com/profile/api-tokens)', mask: true },
+      { key: 'CLOUDFLARE_ACCOUNT_ID', label: 'Account ID (dash.cloudflare.com — right sidebar)', mask: false },
+    ],
+  },
+];
+
+module.exports = { CONNECTOR_PRESETS };

package/src/providers/anthropic.js

@@ -54,9 +54,17 @@ function buildBody(messages, options) {
       // resumed/hand-edited sessions rather than failing the whole turn.
       const tcid = m.toolCallId || m.tool_call_id;
       if (tcid) {
+        // is_error marks a failed tool call so the model treats the string
+        // content as an error report, not a successful result (parity with
+        // the OpenAI-compat is_error/tool-result semantics).
         out.push({
           role: 'user',
-          content: [{ type: 'tool_result', tool_use_id: tcid, content: String(m.content) }],
+          content: [{
+            type: 'tool_result',
+            tool_use_id: tcid,
+            content: String(m.content),
+            ...(m.isError ? { is_error: true } : {}),
+          }],
         });
       }
       continue;
@@ -156,29 +164,69 @@ function normalizeBlocks(contentBlocks) {
   return { content: textParts.join(''), toolCalls };
 }
 
+// Retry-After header (seconds) or rate-limit reset timestamps → wait ms.
+function retryAfterMs(resp) {
+  const ra = resp && resp.headers && resp.headers.get('retry-after');
+  if (ra) {
+    const secs = Number(ra);
+    if (Number.isFinite(secs) && secs >= 0) return Math.min(secs * 1000, 60000);
+  }
+  return null;
+}
+
+// Retryable: rate limits (429) and transient server errors (5xx/529).
+function isRetryableStatus(status) {
+  return status === 429 || status === 529 || (status >= 500 && status < 600);
+}
+
 async function post(body, options) {
   const key = getKey();
   if (!key) throw new Error("Anthropic API key not set. Use /connect anthropic or set ANTHROPIC_API_KEY.");
   const apiUrl = getApiUrl();
-  const resp = await fetch(apiUrl, {
-    method: 'POST',
-    headers: {
-      'Content-Type': 'application/json',
-      'x-api-key': key,
-      'anthropic-version': '2023-06-01',
-    },
-    body: JSON.stringify(body),
-    signal: options.signal,
-  });
-  if (!resp.ok) {
-    /** @type {any} */
-    const err = await resp.json().catch(() => ({}));
-    const detail = err.error?.message || resp.statusText;
-    if (resp.status === 401) throw new Error("Anthropic 401 Unauthorized: the API key is invalid or expired. Run /connect anthropic and paste a new key.");
-    if (resp.status === 403) throw new Error("Anthropic 403 Forbidden: the API key is not authorized for this model. Check the model name and your account access.");
-    throw new Error(`Anthropic API error ${resp.status}: ${detail}`);
+  const maxTries = 4;
+  let resp = null;
+  for (let attempt = 0; attempt < maxTries; attempt++) {
+    resp = await fetch(apiUrl, {
+      method: 'POST',
+      headers: {
+        'Content-Type': 'application/json',
+        'x-api-key': key,
+        'anthropic-version': '2023-06-01',
+      },
+      body: JSON.stringify(body),
+      signal: options.signal,
+    });
+    if (resp.ok) return resp;
+    if (!isRetryableStatus(resp.status) || attempt === maxTries - 1) break;
+    // Honor the provider's Retry-After (seconds); else exponential backoff
+    // with jitter so parallel retries don't stampede.
+    const waitMs = retryAfterMs(resp) ?? Math.min(1000 * Math.pow(2, attempt), 30000);
+    const jitter = Math.floor(Math.random() * 250);
+    await new Promise((r) => setTimeout(r, waitMs + jitter));
+    if (options.signal && options.signal.aborted) {
+      const abortErr = new Error('Request aborted');
+      abortErr.name = 'AbortError';
+      throw abortErr;
+    }
   }
-  return resp;
+  if (!resp) throw new Error('Anthropic API error: no response');
+  /** @type {any} */
+  const err = await resp.json().catch(() => ({}));
+  const detail = err.error?.message || resp.statusText;
+  if (resp.status === 401) throw new Error("Anthropic 401 Unauthorized: the API key is invalid or expired. Run /connect anthropic and paste a new key.");
+  if (resp.status === 403) throw new Error("Anthropic 403 Forbidden: the API key is not authorized for this model. Check the model name and your account access.");
+  const e = new Error(`Anthropic API error ${resp.status}: ${detail}`);
+  /** @type {any} */ (e).status = resp.status;
+  /** @type {any} */ (e).retryAfterMs = retryAfterMs(resp);
+  throw e;
+}
+
+// Anthropic stop_reason → the finishReason vocabulary the session loop knows
+// ('length' drives truncation auto-continue; 'tool_use'/'end_turn' are no-ops).
+function mapStopReason(stopReason) {
+  if (stopReason === 'max_tokens') return 'length';
+  if (stopReason === 'tool_use') return 'tool_calls';
+  return stopReason || undefined;
 }
 
 async function chat(messages, options = {}) {
@@ -186,7 +234,11 @@ async function chat(messages, options = {}) {
   const resp = await post(body, options);
   /** @type {any} */
   const data = await resp.json();
-  return { ...normalizeBlocks(data.content || []), usage: data.usage };
+  return {
+    ...normalizeBlocks(data.content || []),
+    usage: data.usage,
+    finishReason: mapStopReason(data.stop_reason),
+  };
 }
 
 async function stream(messages, options = {}, onDelta, onReasoning) {
@@ -198,6 +250,8 @@ async function stream(messages, options = {}, onDelta, onReasoning) {
   let content = '';
   let reasoning = '';
   let usage = null;
+  let finishReason;
+  let sawEnd = false;
   const toolAcc = new Map();
 
   for await (const chunk of resp.body) {
@@ -210,7 +264,25 @@ async function stream(messages, options = {}, onDelta, onReasoning) {
       if (raw === '[DONE]') continue;
       let evt;
       try { evt = JSON.parse(raw); } catch { continue; }
-      if (evt.type === 'message_delta' && evt.usage) usage = evt.usage;
+      // Mid-stream error events (content filter, overloaded after start)
+      // previously vanished — the turn dead-ended with empty content. Throw
+      // so the session retry/error path handles it.
+      if (evt.type === 'error') {
+        const emsg = evt.error?.message || 'unknown stream error';
+        const e = new Error(`Anthropic stream error: ${emsg}`);
+        /** @type {any} */ (e).status = evt.error?.type === 'overloaded_error' ? 529 : undefined;
+        throw e;
+      }
+      if (evt.type === 'message_start' && evt.message?.usage) {
+        // message_start carries input_tokens (prompt) — without it, a turn
+        // that ends in message_delta only undercounts input to zero.
+        usage = { ...evt.message.usage };
+      }
+      if (evt.type === 'message_delta') {
+        if (evt.usage) usage = { ...(usage || {}), ...evt.usage };
+        if (evt.delta && evt.delta.stop_reason) finishReason = mapStopReason(evt.delta.stop_reason);
+        sawEnd = true;
+      }
       if (evt.type === 'content_block_start' && evt.content_block?.type === 'tool_use') {
         toolAcc.set(evt.index, { id: evt.content_block.id, name: evt.content_block.name, input: '' });
       } else if (evt.type === 'content_block_delta') {
@@ -229,6 +301,14 @@ async function stream(messages, options = {}, onDelta, onReasoning) {
     }
   }
 
+  // A stream that never delivered message_delta AND produced nothing was
+  // truncated (network reset mid-body). Surface it as an error instead of a
+  // silent empty reply. Partial content with no trailing message_delta is
+  // kept (the session can continue from it).
+  if (!sawEnd && !usage && !content && !toolAcc.size) {
+    throw new Error('Anthropic stream truncated: connection ended before message_delta');
+  }
+
   const toolCalls = [];
   for (const acc of toolAcc.values()) {
     let input = {};
@@ -236,13 +316,35 @@ async function stream(messages, options = {}, onDelta, onReasoning) {
     toolCalls.push({ id: acc.id, name: acc.name, input });
   }
 
-  return { content, reasoning, toolCalls, usage };
+  return { content, reasoning, toolCalls, usage, finishReason };
 }
 
+// Current Anthropic lineup, verified against the models.dev registry
+// (2026-10-17). Dated snapshots kept for back-compat with saved configs.
 const models = [
-  { id: 'claude-sonnet-4-20250514', name: 'Claude Sonnet 4', provider: 'anthropic', context: 200000, priceIn: 3, priceOut: 15, tags: ['reasoning'] },
-  { id: 'claude-opus-4-20250514', name: 'Claude Opus 4', provider: 'anthropic', context: 200000, priceIn: 15, priceOut: 75, tags: ['reasoning'] },
-  { id: 'claude-3-5-haiku-20241022', name: 'Claude 3.5 Haiku', provider: 'anthropic', context: 200000, priceIn: 0.8, priceOut: 4 },
+  // ── Opus ──
+  { id: 'claude-opus-5-5', name: 'Claude Opus 5.5', provider: 'anthropic', context: 1000000, priceIn: 4, priceOut: 20, tags: ['frontier', 'reasoning'] },
+  { id: 'claude-opus-5', name: 'Claude Opus 5', provider: 'anthropic', context: 1000000, priceIn: 5, priceOut: 25, tags: ['frontier', 'reasoning'] },
+  { id: 'claude-opus-4-8', name: 'Claude Opus 4.8', provider: 'anthropic', context: 1000000, priceIn: 5, priceOut: 25, tags: ['frontier', 'reasoning'] },
+  { id: 'claude-opus-4-5', name: 'Claude Opus 4.5', provider: 'anthropic', context: 200000, priceIn: 5, priceOut: 25, tags: ['frontier', 'reasoning'] },
+  { id: 'claude-fable-5-1', name: 'Claude Fable 5.1', provider: 'anthropic', context: 1000000, priceIn: 10, priceOut: 50, tags: ['frontier', 'reasoning'] },
+  { id: 'claude-fable-5', name: 'Claude Fable 5', provider: 'anthropic', context: 1000000, priceIn: 10, priceOut: 50, tags: ['frontier', 'reasoning'] },
+  { id: 'claude-opus-4-7', name: 'Claude Opus 4.7', provider: 'anthropic', context: 1000000, priceIn: 5, priceOut: 25, tags: ['frontier', 'reasoning'] },
+  { id: 'claude-opus-4-6', name: 'Claude Opus 4.6', provider: 'anthropic', context: 1000000, priceIn: 5, priceOut: 25, tags: ['frontier', 'reasoning'] },
+  // ── Sonnet ──
+  { id: 'claude-sonnet-5-5', name: 'Claude Sonnet 5.5', provider: 'anthropic', context: 1000000, priceIn: 2, priceOut: 10, tags: ['reasoning'] },
+  { id: 'claude-sonnet-5', name: 'Claude Sonnet 5', provider: 'anthropic', context: 1000000, priceIn: 2, priceOut: 10, tags: ['reasoning'] },
+  { id: 'claude-sonnet-4-6', name: 'Claude Sonnet 4.6', provider: 'anthropic', context: 1000000, priceIn: 3, priceOut: 15, tags: ['reasoning'] },
+  { id: 'claude-sonnet-4-5', name: 'Claude Sonnet 4.5', provider: 'anthropic', context: 200000, priceIn: 3, priceOut: 15, tags: ['reasoning'] },
+  // ── Haiku ──
+  { id: 'claude-haiku-5-5', name: 'Claude Haiku 5.5', provider: 'anthropic', context: 1000000, priceIn: 0.1, priceOut: 0.5, tags: ['fast', 'cheap', 'reasoning'] },
+  { id: 'claude-haiku-4-5', name: 'Claude Haiku 4.5', provider: 'anthropic', context: 200000, priceIn: 1, priceOut: 5, tags: ['fast', 'reasoning'] },
+  // ── Legacy snapshots ──
+  // Retired models verified against Anthropic's model-deprecations page
+  // (2026-10-17): claude-sonnet-4-20250514 and claude-opus-4-20250514 were
+  // retired 2026-06-15, claude-3-5-haiku-20241022 retired 2026-02-19 — they
+  // 404 and are no longer listed. claude-mythos-5/5.1 exist but are limited
+  // to verified organizations, so they are not offered in the picker.
 ];
 
 module.exports = { chat, stream, models, buildBody, normalizeBlocks };

package/src/providers/google.js

@@ -12,17 +12,25 @@ const provider = {
     envKeyHint: 'GOOGLE',
   }),
   models: [
-  // ── Frontier ──
-  { id: 'gemini-2.5-pro', name: 'Gemini 2.5 Pro', provider: 'google', tags: ['frontier'], context: 1000000, priceIn: 1.25, priceOut: 10 },
-  { id: 'gemini-2.5-flash', name: 'Gemini 2.5 Flash', provider: 'google', tags: ['fast'], context: 1000000, priceIn: 0.3, priceOut: 2.5 },
-  // ── Flash ──
-  { id: 'gemini-2.0-flash', name: 'Gemini 2.0 Flash', provider: 'google', tags: [], context: 1000000, priceIn: 0.1, priceOut: 0.4 },
-  { id: 'gemini-2.0-flash-lite', name: 'Gemini 2.0 Flash Lite', provider: 'google', tags: ['free', 'fast'], context: 1000000, priceIn: 0.075, priceOut: 0.3 },
-  // ── Legacy ──
-  { id: 'gemini-1.5-flash', name: 'Gemini 1.5 Flash', provider: 'google', tags: [], context: 1000000, priceIn: 0.075, priceOut: 0.3 },
-  { id: 'gemini-1.5-pro', name: 'Gemini 1.5 Pro', provider: 'google', tags: [], context: 2000000, priceIn: 1.25, priceOut: 5 },
-  // ── Gemma (Local / On-device) ──
-  { id: 'gemma-4-31b-it', name: 'Gemma 4 31B IT', provider: 'google', tags: ['small'], context: 32768, priceIn: 0.2, priceOut: 0.8 },
+  // ── Gemini 4 / 3.x (current) ──
+  { id: 'gemini-4-argon', name: 'Gemini 4 Argon', provider: 'google', tags: ['frontier', 'reasoning'], context: 1048576, priceIn: 2, priceOut: 10 },
+  { id: 'gemini-3.1-pro-preview', name: 'Gemini 3.1 Pro Preview', provider: 'google', tags: ['frontier', 'reasoning'], context: 1048576, priceIn: 2, priceOut: 12 },
+  { id: 'gemini-3.8-flash', name: 'Gemini 3.8 Flash', provider: 'google', tags: ['fast', 'reasoning'], context: 1048576, priceIn: 0.75, priceOut: 3.75 },
+  { id: 'gemini-3.7-flash', name: 'Gemini 3.7 Flash', provider: 'google', tags: ['fast', 'reasoning'], context: 1048576, priceIn: 0.75, priceOut: 3.75 },
+  { id: 'gemini-3.6-flash', name: 'Gemini 3.6 Flash', provider: 'google', tags: ['fast', 'reasoning'], context: 1048576, priceIn: 0.75, priceOut: 3.75 },
+  { id: 'gemini-3.5-flash', name: 'Gemini 3.5 Flash', provider: 'google', tags: ['reasoning'], context: 1048576, priceIn: 1.5, priceOut: 9 },
+  { id: 'gemini-3.5-flash-lite', name: 'Gemini 3.5 Flash Lite', provider: 'google', tags: ['fast', 'cheap', 'reasoning'], context: 1048576, priceIn: 0.3, priceOut: 2.5 },
+  { id: 'gemini-3.1-flash-lite', name: 'Gemini 3.1 Flash Lite', provider: 'google', tags: ['fast', 'cheap', 'reasoning'], context: 1048576, priceIn: 0.25, priceOut: 1.5 },
+  // ── Aliases ──
+  { id: 'gemini-flash-latest', name: 'Gemini Flash (Latest)', provider: 'google', tags: ['fast', 'reasoning'], context: 1048576, priceIn: 0.75, priceOut: 3.75 },
+  { id: 'gemini-flash-lite-latest', name: 'Gemini Flash-Lite (Latest)', provider: 'google', tags: ['fast', 'cheap', 'reasoning'], context: 1048576, priceIn: 0.3, priceOut: 2.5 },
+  // ── Gemini 2.5 (stable) ──
+  { id: 'gemini-2.5-pro', name: 'Gemini 2.5 Pro', provider: 'google', tags: ['frontier'], context: 1048576, priceIn: 1.25, priceOut: 10 },
+  { id: 'gemini-2.5-flash', name: 'Gemini 2.5 Flash', provider: 'google', tags: ['fast'], context: 1048576, priceIn: 0.3, priceOut: 2.5 },
+  { id: 'gemini-2.5-flash-lite', name: 'Gemini 2.5 Flash-Lite', provider: 'google', tags: ['fast', 'cheap'], context: 1048576, priceIn: 0.1, priceOut: 0.4 },
+  // ── Gemma (open weights) ──
+  { id: 'gemma-4-31b-it', name: 'Gemma 4 31B IT', provider: 'google', tags: ['reasoning'], context: 262144, priceIn: 0.2, priceOut: 0.8 },
+  { id: 'gemma-4-26b-a4b-it', name: 'Gemma 4 26B A4B IT', provider: 'google', tags: ['fast', 'reasoning'], context: 262144, priceIn: 0.2, priceOut: 0.8 },
   ],
 };
 

package/src/providers/nvidia.js

@@ -5,10 +5,23 @@ function getKey() {
   return getApiKey('nvidia') || process.env.NVIDIA_API_KEY || process.env.NVIDIA_NIM_API_KEY;
 }
 
-// Model IDs below are a curated subset of NVIDIA's live GET /v1/models registry
-// (fetched 2026-08-13). Stale IDs — e.g. meta/llama-3.1-405b-instruct,
-// deepseek-ai/deepseek-v4-flash, minimax-ai/minimax-m2.7 — return 404/410 and
-// were deliberately removed. Re-verify against /v1/models before adding any ID.
+// Model IDs below are a curated subset of NVIDIA's live GET /v1/models
+// registry (fetched + cross-checked against models.dev 2026-10-17, 81 live
+// models; only chat/completions-capable IDs kept — embeddings, rerankers,
+// safety/guard, document-parse, video/vision-specialist and reward models
+// are excluded). Stale IDs — e.g. meta/llama-3.1-8b-instruct,
+// deepseek-ai/deepseek-v4-flash-0731, z-ai/glm-5.2, openai/gpt-oss-120b —
+// return 404/410 and were removed. Re-verify against /v1/models before
+// adding any ID.
+//
+// NOTE (probed 2026-10-17): GET /v1/models is the GLOBAL catalog — a chat
+// POST can still fail with `Function '<id>': Not found for account '<key>'`,
+// i.e. the model exists but the key's account hasn't enabled it. On one
+// sample key only 9/54 of these IDs answered a chat call (37 account-gated
+// 404s, ~6 hang to timeout, 2 return 503 capacity errors), so the picker
+// may offer models that 404 for a given account. Availability is
+// per-account, so the full curated list stays; NVIDIA's own errors carry
+// the account hint.
 const provider = {
   ...createOpenAICompatProvider({
     getKey,
@@ -17,38 +30,34 @@ const provider = {
   }),
   models: [
   // ── DeepSeek ──
-  { id: 'deepseek-ai/deepseek-v4-flash-0731', name: 'DeepSeek V4 Flash', provider: 'nvidia', tags: ['free', 'fast'], context: 200000, priceIn: 0, priceOut: 0 },
+  { id: 'deepseek-ai/deepseek-v4.1-flash', name: 'DeepSeek V4.1 Flash', provider: 'nvidia', tags: ['reasoning', 'fast'], context: 1000000, priceIn: 0.14, priceOut: 0.28 },
   { id: 'deepseek-ai/deepseek-coder-6.7b-instruct', name: 'DeepSeek Coder 6.7B', provider: 'nvidia', tags: ['coding'], context: 16384, priceIn: 0.10, priceOut: 0.10 },
 
   // ── Meta Llama ──
-  { id: 'meta/llama-3.1-8b-instruct', name: 'Llama 3.1 8B Instruct', provider: 'nvidia', tags: ['fast'], context: 128000, priceIn: 0.05, priceOut: 0.08 },
-  { id: 'meta/llama-3.1-70b-instruct', name: 'Llama 3.1 70B Instruct', provider: 'nvidia', tags: ['general'], context: 128000, priceIn: 0.35, priceOut: 0.40 },
-  { id: 'meta/llama-3.2-3b-instruct', name: 'Llama 3.2 3B Instruct', provider: 'nvidia', tags: ['fast'], context: 128000, priceIn: 0.02, priceOut: 0.02 },
   { id: 'meta/llama-3.2-11b-vision-instruct', name: 'Llama 3.2 11B Vision', provider: 'nvidia', tags: ['vision', 'fast'], context: 128000, priceIn: 0.055, priceOut: 0.055 },
   { id: 'meta/llama-3.2-90b-vision-instruct', name: 'Llama 3.2 90B Vision', provider: 'nvidia', tags: ['vision'], context: 128000, priceIn: 0.35, priceOut: 0.40 },
-  { id: 'meta/llama-3.3-70b-instruct', name: 'Llama 3.3 70B Instruct', provider: 'nvidia', tags: ['general'], context: 128000, priceIn: 0.35, priceOut: 0.40 },
+  { id: 'meta/muse-glimmer-30b', name: 'Llama Muse Glimmer 30B', provider: 'nvidia', tags: ['creative', 'reasoning'], context: 131072, priceIn: 0.20, priceOut: 0.20 },
   { id: 'meta/codellama-70b', name: 'CodeLlama 70B', provider: 'nvidia', tags: ['coding'], context: 16384, priceIn: 0.30, priceOut: 0.30 },
-  { id: 'meta/muse-glimmer-30b', name: 'Llama Muse Glimmer 30B', provider: 'nvidia', tags: ['creative'], context: 32768, priceIn: 0.20, priceOut: 0.20 },
+  { id: 'meta/llama2-70b', name: 'Llama 2 70B', provider: 'nvidia', tags: ['legacy'], context: 4096, priceIn: 0.20, priceOut: 0.20 },
 
   // ── NVIDIA Nemotron ──
-  { id: 'nvidia/llama-3.1-nemotron-70b-instruct', name: 'Llama 3.1 Nemotron 70B', provider: 'nvidia', tags: ['general'], context: 128000, priceIn: 0.35, priceOut: 0.40 },
-  { id: 'nvidia/llama-3.1-nemotron-ultra-253b-v1', name: 'Nemotron Ultra 253B', provider: 'nvidia', tags: ['frontier'], context: 128000, priceIn: 1.50, priceOut: 6.00 },
-  { id: 'nvidia/llama-3.3-nemotron-super-49b-v1', name: 'Nemotron Super 49B', provider: 'nvidia', tags: ['reasoning'], context: 128000, priceIn: 0.25, priceOut: 0.90 },
-  { id: 'nvidia/llama-3.3-nemotron-super-49b-v1.5', name: 'Nemotron Super 49B v1.5', provider: 'nvidia', tags: ['reasoning'], context: 128000, priceIn: 0.25, priceOut: 0.90 },
-  { id: 'nvidia/nemotron-3-ultra-550b-a55b', name: 'Nemotron 3 Ultra 550B', provider: 'nvidia', tags: ['frontier'], context: 256000, priceIn: 2.00, priceOut: 8.00 },
-  { id: 'nvidia/nemotron-3-super-120b-a12b', name: 'Nemotron 3 Super 120B', provider: 'nvidia', tags: ['frontier'], context: 256000, priceIn: 1.00, priceOut: 4.00 },
-  { id: 'nvidia/nemotron-3-nano-30b-a3b', name: 'Nemotron 3 Nano 30B', provider: 'nvidia', tags: ['fast'], context: 128000, priceIn: 0.10, priceOut: 0.40 },
-  { id: 'nvidia/nemotron-3-nano-omni-30b-a3b-reasoning', name: 'Nemotron 3 Nano Omni (Reasoning)', provider: 'nvidia', tags: ['reasoning', 'small'], context: 128000, priceIn: 0.10, priceOut: 0.40 },
-  { id: 'nvidia/nemotron-4-340b-instruct', name: 'Nemotron-4 340B Instruct', provider: 'nvidia', tags: ['frontier'], context: 4096, priceIn: 1.00, priceOut: 3.00 },
-  { id: 'nvidia/nemotron-mini-4b-instruct', name: 'Nemotron Mini 4B', provider: 'nvidia', tags: ['fast'], context: 32768, priceIn: 0.03, priceOut: 0.03 },
+  { id: 'nvidia/nemotron-3-ultra-550b-a55b', name: 'Nemotron 3 Ultra 550B', provider: 'nvidia', tags: ['frontier', 'reasoning'], context: 1000000, priceIn: 0.50, priceOut: 2.50 },
+  { id: 'nvidia/nemotron-3-super-120b-a12b', name: 'Nemotron 3 Super 120B', provider: 'nvidia', tags: ['reasoning'], context: 262144, priceIn: 0.25, priceOut: 1.00 },
+  { id: 'nvidia/nemotron-3-nano-omni-30b-a3b-reasoning', name: 'Nemotron 3 Nano Omni (Reasoning)', provider: 'nvidia', tags: ['reasoning', 'fast'], context: 256000, priceIn: 0.10, priceOut: 0.40 },
   { id: 'nvidia/nemotron-nano-3-30b-a3b', name: 'Nemotron Nano 3 30B', provider: 'nvidia', tags: ['fast'], context: 128000, priceIn: 0.10, priceOut: 0.40 },
-  { id: 'nvidia/nemotron-nano-12b-v2-vl', name: 'Nemotron Nano 12B V2 VL', provider: 'nvidia', tags: ['vision'], context: 128000, priceIn: 0.10, priceOut: 0.40 },
-  { id: 'nvidia/nvidia-nemotron-nano-9b-v2', name: 'Nemotron Nano 9B V2', provider: 'nvidia', tags: ['fast'], context: 128000, priceIn: 0.05, priceOut: 0.20 },
+  { id: 'nvidia/nemotron-3.5-lightning-30b-a3b', name: 'Nemotron 3.5 Lightning 30B', provider: 'nvidia', tags: ['reasoning', 'fast'], context: 262144, priceIn: 0.10, priceOut: 0.40 },
+  { id: 'nvidia/nemotron-3.5-super-vl-120b-a12b', name: 'Nemotron 3.5 Super VL 120B', provider: 'nvidia', tags: ['vision', 'reasoning'], context: 1000000, priceIn: 0.25, priceOut: 1.00 },
+  { id: 'nvidia/llama-3.1-nemotron-ultra-253b-v1', name: 'Nemotron Ultra 253B', provider: 'nvidia', tags: ['frontier', 'reasoning'], context: 128000, priceIn: 1.50, priceOut: 6.00 },
+  { id: 'nvidia/llama-3.1-nemotron-70b-instruct', name: 'Llama 3.1 Nemotron 70B', provider: 'nvidia', tags: ['general'], context: 128000, priceIn: 0.35, priceOut: 0.40 },
+  { id: 'nvidia/llama-3.1-nemotron-51b-instruct', name: 'Llama 3.1 Nemotron 51B', provider: 'nvidia', tags: ['general'], context: 128000, priceIn: 0.20, priceOut: 0.80 },
+  { id: 'nvidia/llama3-chatqa-1.5-70b', name: 'Llama3 ChatQA 1.5 70B', provider: 'nvidia', tags: ['general'], context: 8192, priceIn: 0.30, priceOut: 0.30 },
+  { id: 'nvidia/mistral-nemo-minitron-8b-8k-instruct', name: 'Mistral NeMo Minitron 8B', provider: 'nvidia', tags: ['fast'], context: 8192, priceIn: 0.05, priceOut: 0.05 },
+  { id: 'nvidia/nemotron-4-340b-instruct', name: 'Nemotron-4 340B Instruct', provider: 'nvidia', tags: ['frontier', 'legacy'], context: 4096, priceIn: 1.00, priceOut: 3.00 },
 
   // ── Mistral ──
-  { id: 'mistralai/mistral-7b-instruct-v0.3', name: 'Mistral 7B Instruct', provider: 'nvidia', tags: ['fast'], context: 32768, priceIn: 0.05, priceOut: 0.05 },
   { id: 'mistralai/mistral-large', name: 'Mistral Large', provider: 'nvidia', tags: ['frontier'], context: 128000, priceIn: 2.00, priceOut: 6.00 },
   { id: 'mistralai/mistral-large-2-instruct', name: 'Mistral Large 2 Instruct', provider: 'nvidia', tags: ['frontier'], context: 128000, priceIn: 2.00, priceOut: 6.00 },
+  { id: 'mistralai/mistral-7b-instruct-v0.3', name: 'Mistral 7B Instruct', provider: 'nvidia', tags: ['fast'], context: 65536, priceIn: 0.05, priceOut: 0.05 },
   { id: 'mistralai/codestral-22b-instruct-v0.1', name: 'Codestral 22B', provider: 'nvidia', tags: ['coding'], context: 32768, priceIn: 0.30, priceOut: 0.60 },
   { id: 'mistralai/mixtral-8x22b-v0.1', name: 'Mixtral 8x22B', provider: 'nvidia', tags: [], context: 65536, priceIn: 0.60, priceOut: 0.60 },
   { id: 'nv-mistralai/mistral-nemo-12b-instruct', name: 'Mistral NeMo 12B', provider: 'nvidia', tags: ['fast'], context: 128000, priceIn: 0.05, priceOut: 0.05 },
@@ -58,26 +67,38 @@ const provider = {
   { id: 'microsoft/phi-3-vision-128k-instruct', name: 'Phi 3 Vision 128K', provider: 'nvidia', tags: ['vision'], context: 128000, priceIn: 0.05, priceOut: 0.05 },
 
   // ── Google ──
-  { id: 'google/gemma-3-4b-it', name: 'Gemma 3 4B IT', provider: 'nvidia', tags: ['fast'], context: 32768, priceIn: 0.03, priceOut: 0.03 },
-  { id: 'google/gemma-3-12b-it', name: 'Gemma 3 12B IT', provider: 'nvidia', tags: ['general'], context: 128000, priceIn: 0.10, priceOut: 0.10 },
-  { id: 'google/gemma-4-31b-it', name: 'Gemma 4 31B IT', provider: 'nvidia', tags: ['general'], context: 32768, priceIn: 0.20, priceOut: 0.80 },
+  { id: 'google/gemma-4-31b-it', name: 'Gemma 4 31B IT', provider: 'nvidia', tags: ['general', 'reasoning'], context: 256000, priceIn: 0.20, priceOut: 0.80 },
+  { id: 'google/gemma-3-12b-it', name: 'Gemma 3 12B IT', provider: 'nvidia', tags: ['general'], context: 131072, priceIn: 0.10, priceOut: 0.10 },
+  { id: 'google/gemma-3-4b-it', name: 'Gemma 3 4B IT', provider: 'nvidia', tags: ['fast'], context: 131072, priceIn: 0.03, priceOut: 0.03 },
+  { id: 'google/diffusiongemma-26b-a4b-it', name: 'DiffusionGemma 26B', provider: 'nvidia', tags: ['reasoning'], context: 250000, priceIn: 0.20, priceOut: 0.80 },
   { id: 'google/codegemma-1.1-7b', name: 'CodeGemma 1.1 7B', provider: 'nvidia', tags: ['coding', 'fast'], context: 16384, priceIn: 0.05, priceOut: 0.05 },
+  { id: 'google/codegemma-7b', name: 'CodeGemma 7B', provider: 'nvidia', tags: ['coding'], context: 16384, priceIn: 0.05, priceOut: 0.05 },
 
   // ── OpenAI ──
-  { id: 'openai/gpt-oss-120b', name: 'GPT-OSS 120B', provider: 'nvidia', tags: ['openai', 'general'], context: 131072, priceIn: 0.25, priceOut: 1.00 },
-  { id: 'openai/gpt-oss-20b', name: 'GPT-OSS 20B', provider: 'nvidia', tags: ['openai', 'fast'], context: 131072, priceIn: 0.10, priceOut: 0.30 },
+  { id: 'openai/gpt-oss-20b', name: 'GPT-OSS 20B', provider: 'nvidia', tags: ['openai', 'reasoning', 'fast'], context: 131072, priceIn: 0.10, priceOut: 0.30 },
+
+  // ── Z.ai / Moonshot ──
+  { id: 'z-ai/glm-5.3', name: 'GLM-5.3', provider: 'nvidia', tags: ['reasoning'], context: 1000000, priceIn: 0.30, priceOut: 1.20 },
+  { id: 'z-ai/glm-5.3-flash', name: 'GLM-5.3 Flash', provider: 'nvidia', tags: ['fast', 'reasoning'], context: 1000000, priceIn: 0.05, priceOut: 0.20 },
+  { id: 'moonshotai/kimi-k2.6', name: 'Kimi K2.6', provider: 'nvidia', tags: ['reasoning'], context: 262144, priceIn: 0.20, priceOut: 0.80 },
 
   // ── Other ──
-  { id: 'z-ai/glm-5.2', name: 'GLM-5.2', provider: 'nvidia', tags: [], context: 131072, priceIn: 0.30, priceOut: 1.20 },
-  { id: 'moonshotai/kimi-k2.6', name: 'Kimi K2.6', provider: 'nvidia', tags: [], context: 128000, priceIn: 0.20, priceOut: 0.80 },
-  { id: 'minimaxai/minimax-m3', name: 'MiniMax M3', provider: 'nvidia', tags: [], context: 200000, priceIn: 0.40, priceOut: 1.60 },
-  { id: 'poolside/laguna-xs-2.1', name: 'Laguna XS 2.1', provider: 'nvidia', tags: ['free', 'coding'], context: 131072, priceIn: 0, priceOut: 0 },
-  { id: 'stepfun-ai/step-3.7-flash', name: 'Step 3.7 Flash', provider: 'nvidia', tags: ['fast'], context: 128000, priceIn: 0.15, priceOut: 0.40 },
+  { id: 'poolside/laguna-xs-2.1', name: 'Laguna XS 2.1', provider: 'nvidia', tags: ['free', 'coding', 'reasoning'], context: 262144, priceIn: 0, priceOut: 0 },
+  { id: 'stepfun-ai/step-5-preview', name: 'Step 5 Preview', provider: 'nvidia', tags: ['reasoning'], context: 256000, priceIn: 0.30, priceOut: 1.20 },
+  { id: 'writer/palmyra-creative-122b', name: 'Palmyra Creative 122B', provider: 'nvidia', tags: ['creative'], context: 128000, priceIn: 0.90, priceOut: 0.90 },
+  { id: 'writer/palmyra-fin-70b-32k', name: 'Palmyra Fin 70B', provider: 'nvidia', tags: ['finance'], context: 32768, priceIn: 0.90, priceOut: 0.90 },
+  { id: 'writer/palmyra-med-70b', name: 'Palmyra Med 70B', provider: 'nvidia', tags: ['medical'], context: 128000, priceIn: 0.90, priceOut: 0.90 },
+  { id: 'writer/palmyra-med-70b-32k', name: 'Palmyra Med 70B 32K', provider: 'nvidia', tags: ['medical'], context: 32768, priceIn: 0.90, priceOut: 0.90 },
   { id: '01-ai/yi-large', name: 'Yi Large', provider: 'nvidia', tags: [], context: 32768, priceIn: 0.30, priceOut: 0.90 },
   { id: 'ai21labs/jamba-1.5-large-instruct', name: 'Jamba 1.5 Large', provider: 'nvidia', tags: [], context: 256000, priceIn: 0.20, priceOut: 0.80 },
+  { id: 'aisingapore/sea-lion-7b-instruct', name: 'Sea-Lion 7B Instruct', provider: 'nvidia', tags: ['fast'], context: 8192, priceIn: 0.05, priceOut: 0.05 },
   { id: 'databricks/dbrx-instruct', name: 'DBRX Instruct', provider: 'nvidia', tags: [], context: 32768, priceIn: 0.60, priceOut: 0.60 },
   { id: 'ibm/granite-3.0-8b-instruct', name: 'Granite 3.0 8B', provider: 'nvidia', tags: ['fast'], context: 32768, priceIn: 0.05, priceOut: 0.05 },
+  { id: 'ibm/granite-3.0-3b-a800m-instruct', name: 'Granite 3.0 3B MoE', provider: 'nvidia', tags: ['fast'], context: 32768, priceIn: 0.03, priceOut: 0.03 },
   { id: 'ibm/granite-34b-code-instruct', name: 'Granite 34B Code', provider: 'nvidia', tags: ['coding'], context: 16384, priceIn: 0.20, priceOut: 0.20 },
+  { id: 'ibm/granite-8b-code-instruct', name: 'Granite 8B Code', provider: 'nvidia', tags: ['coding', 'fast'], context: 16384, priceIn: 0.10, priceOut: 0.10 },
+  { id: 'nvidia/riva-translate-4b-instruct', name: 'Riva Translate 4B', provider: 'nvidia', tags: ['fast', 'translation'], context: 32768, priceIn: 0.05, priceOut: 0.05 },
+  { id: 'nvidia/riva-translate-4b-instruct-v2', name: 'Riva Translate 4B v2', provider: 'nvidia', tags: ['fast', 'translation'], context: 32768, priceIn: 0.05, priceOut: 0.05 },
   { id: 'zyphra/zamba2-7b-instruct', name: 'Zamba2 7B', provider: 'nvidia', tags: ['fast'], context: 32768, priceIn: 0.05, priceOut: 0.05 },
 ],
 };