jeo-code 0.6.22 → 0.6.23

package/CHANGELOG.md

@@ -6,6 +6,21 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 The README mirrors the latest 5 entries — regenerate with `bun run changelog:sync`.
 
+## [0.6.23] - 2026-06-19
+_Live reasoning/thinking streams in the TUI across every provider, three new OpenAI-compatible backends (LM Studio, xAI, Kimi) join the auth/discovery/catalog surface, and Gemini gains native function-calling._
+
+### Added
+- **Multi-provider reasoning/thinking streaming in the TUI.** Native reasoning is surfaced live (dimmed) and committed to scrollback for Anthropic (`thinking` deltas), OpenAI Codex/Responses (`reasoning*` deltas), OpenAI-compatible chat (`reasoning_content`/`reasoning`), Gemini & Antigravity (`thought` parts), and Ollama (`message.thinking`). A provider-agnostic `<think>…</think>` splitter routes inline chain-of-thought (DeepSeek-R1/Qwen-style local models) to the reasoning channel so it never pollutes the answer or the tool-call parse.
+- **Three new OpenAI-compatible providers — LM Studio (keyless local), xAI/Grok (`XAI_API_KEY`), and Kimi/Moonshot (`KIMI_API_KEY`).** All route through a shared `makeOpenAICompatibleAdapter` factory and are wired into `/provider`, `jeo auth status/login`, model discovery, and the capability catalog.
+- **Native Gemini function-calling (gjc parity).** Gemini now declares `functionDeclarations` and parses `functionCall` parts instead of the JSON-in-prose protocol — capable models stop fighting the `done` format, cutting wasted steps and stray "apology" prose from replies (verified live: a trivial reply dropped from 3 steps/14s to 1 step/2s).
+- **Mid-turn `/command` and `$skill` dispatch** with a live command/skill preview while typing.
+
+### Changed
+- **API-key providers are first-class in the auth core.** `AuthProvider` now splits into the OAuth-capable subset (`OAuthProvider`) plus API-key-only providers (xai/kimi); these resolve through the standard `resolveCredential` path (`config.providers` / `<NAME>_API_KEY`) and model discovery now sends their key (a prior gap left discovery unauthenticated). `jeo auth login <xai|kimi> --token <key>` stores the API key.
+
+### Fixed
+- **Config-schema dropped a stored `xai` key on validation** (the providers schema was missing `xai`/`kimi`); both are now persisted.
+
 ## [0.6.22] - 2026-06-18
 _Extended-thinking activation is now consistent across providers: a `low` session thinking level enables reasoning everywhere._
 

package/README.ja.md

@@ -2,6 +2,10 @@
   <img src="assets/hero.png" alt="jeo-code 自律コーディングエージェントのヒーローイラスト" width="100%" />
 </p>
 
+<p align="center">
+  <img src="assets/icon.png" alt="jeo-code icon" width="96" />
+</p>
+
 <h1 align="center">jeo-code (jeo)</h1>
 
 <p align="center">
@@ -200,11 +204,11 @@ CI は `.github/workflows/npm-publish.yml` で公開します — GitHub リリ
 ## 変更履歴 (Changelog)
 
 <!-- CHANGELOG:START (auto-generated from CHANGELOG.md — run `bun run changelog:sync`) -->
+- **[0.6.23]** (2026-06-19) — Live reasoning/thinking streams in the TUI across every provider, three new OpenAI-compatible backends (LM Studio, xAI, Kimi) join the auth/discovery/catalog surface, and Gemini gains native function-calling.
 - **[0.6.22]** (2026-06-18) — Extended-thinking activation is now consistent across providers: a `low` session thinking level enables reasoning everywhere.
 - **[0.6.21]** (2026-06-18) — Session thinking level now reaches the provider's actual reasoning depth, not just the token ceiling.
 - **[0.6.20]** (2026-06-18) — Launch REPL internals decomposed into testable modules: `@mention` path completion, slash-command view renderers, and slash-command handlers extracted from the monolithic `launch.ts` into dedicated files with full unit-test coverage.
 - **[0.6.19]** (2026-06-18) — Post-turn hooks run once per batch (not per edit), local hook reads are mtime-cached, tool-result formatting is parallelized, and wrapped colored text keeps its tint.
-- **[0.6.18]** (2026-06-17) — Memory data-flow diagram and a README "Memory flow" section documenting the actual runtime behavior.
 
 See [CHANGELOG.md](CHANGELOG.md) for the full history.
 <!-- CHANGELOG:END -->

package/README.ko.md

@@ -2,6 +2,10 @@
   <img src="assets/hero.png" alt="jeo-code 자율 코딩 에이전트 히어로 일러스트" width="100%" />
 </p>
 
+<p align="center">
+  <img src="assets/icon.png" alt="jeo-code icon" width="96" />
+</p>
+
 <h1 align="center">jeo-code (jeo)</h1>
 
 <p align="center">
@@ -200,11 +204,11 @@ CI는 `.github/workflows/npm-publish.yml`로 배포합니다 — GitHub 릴리
 ## 변경 이력 (Changelog)
 
 <!-- CHANGELOG:START (auto-generated from CHANGELOG.md — run `bun run changelog:sync`) -->
+- **[0.6.23]** (2026-06-19) — Live reasoning/thinking streams in the TUI across every provider, three new OpenAI-compatible backends (LM Studio, xAI, Kimi) join the auth/discovery/catalog surface, and Gemini gains native function-calling.
 - **[0.6.22]** (2026-06-18) — Extended-thinking activation is now consistent across providers: a `low` session thinking level enables reasoning everywhere.
 - **[0.6.21]** (2026-06-18) — Session thinking level now reaches the provider's actual reasoning depth, not just the token ceiling.
 - **[0.6.20]** (2026-06-18) — Launch REPL internals decomposed into testable modules: `@mention` path completion, slash-command view renderers, and slash-command handlers extracted from the monolithic `launch.ts` into dedicated files with full unit-test coverage.
 - **[0.6.19]** (2026-06-18) — Post-turn hooks run once per batch (not per edit), local hook reads are mtime-cached, tool-result formatting is parallelized, and wrapped colored text keeps its tint.
-- **[0.6.18]** (2026-06-17) — Memory data-flow diagram and a README "Memory flow" section documenting the actual runtime behavior.
 
 See [CHANGELOG.md](CHANGELOG.md) for the full history.
 <!-- CHANGELOG:END -->

package/README.md

@@ -2,6 +2,10 @@
   <img src="assets/hero.png" alt="jeo-code autonomous coding-agent hero illustration" width="100%" />
 </p>
 
+<p align="center">
+  <img src="assets/icon.png" alt="jeo-code icon" width="96" />
+</p>
+
 <h1 align="center">jeo-code (jeo)</h1>
 
 <p align="center">
@@ -200,11 +204,11 @@ Required npm token permissions (repository secret `NPM_TOKEN`):
 ## Changelog
 
 <!-- CHANGELOG:START (auto-generated from CHANGELOG.md — run `bun run changelog:sync`) -->
+- **[0.6.23]** (2026-06-19) — Live reasoning/thinking streams in the TUI across every provider, three new OpenAI-compatible backends (LM Studio, xAI, Kimi) join the auth/discovery/catalog surface, and Gemini gains native function-calling.
 - **[0.6.22]** (2026-06-18) — Extended-thinking activation is now consistent across providers: a `low` session thinking level enables reasoning everywhere.
 - **[0.6.21]** (2026-06-18) — Session thinking level now reaches the provider's actual reasoning depth, not just the token ceiling.
 - **[0.6.20]** (2026-06-18) — Launch REPL internals decomposed into testable modules: `@mention` path completion, slash-command view renderers, and slash-command handlers extracted from the monolithic `launch.ts` into dedicated files with full unit-test coverage.
 - **[0.6.19]** (2026-06-18) — Post-turn hooks run once per batch (not per edit), local hook reads are mtime-cached, tool-result formatting is parallelized, and wrapped colored text keeps its tint.
-- **[0.6.18]** (2026-06-17) — Memory data-flow diagram and a README "Memory flow" section documenting the actual runtime behavior.
 
 See [CHANGELOG.md](CHANGELOG.md) for the full history.
 <!-- CHANGELOG:END -->

package/README.zh.md

@@ -2,6 +2,10 @@
   <img src="assets/hero.png" alt="jeo-code 自主编码代理主视觉插图" width="100%" />
 </p>
 
+<p align="center">
+  <img src="assets/icon.png" alt="jeo-code icon" width="96" />
+</p>
+
 <h1 align="center">jeo-code (jeo)</h1>
 
 <p align="center">
@@ -200,11 +204,11 @@ CI 通过 `.github/workflows/npm-publish.yml` 发布 — GitHub 发布 release
 ## 更新日志 (Changelog)
 
 <!-- CHANGELOG:START (auto-generated from CHANGELOG.md — run `bun run changelog:sync`) -->
+- **[0.6.23]** (2026-06-19) — Live reasoning/thinking streams in the TUI across every provider, three new OpenAI-compatible backends (LM Studio, xAI, Kimi) join the auth/discovery/catalog surface, and Gemini gains native function-calling.
 - **[0.6.22]** (2026-06-18) — Extended-thinking activation is now consistent across providers: a `low` session thinking level enables reasoning everywhere.
 - **[0.6.21]** (2026-06-18) — Session thinking level now reaches the provider's actual reasoning depth, not just the token ceiling.
 - **[0.6.20]** (2026-06-18) — Launch REPL internals decomposed into testable modules: `@mention` path completion, slash-command view renderers, and slash-command handlers extracted from the monolithic `launch.ts` into dedicated files with full unit-test coverage.
 - **[0.6.19]** (2026-06-18) — Post-turn hooks run once per batch (not per edit), local hook reads are mtime-cached, tool-result formatting is parallelized, and wrapped colored text keeps its tint.
-- **[0.6.18]** (2026-06-17) — Memory data-flow diagram and a README "Memory flow" section documenting the actual runtime behavior.
 
 See [CHANGELOG.md](CHANGELOG.md) for the full history.
 <!-- CHANGELOG:END -->

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "jeo-code",
-  "version": "0.6.22",
+  "version": "0.6.23",
   "description": "Clean, highly optimized AI coding agent using spec-first loop",
   "type": "module",
   "main": "src/cli.ts",

package/src/agent/config-schema.ts

@@ -1,4 +1,5 @@
 import { findCatalogEntry } from "../ai/model-catalog-compat";
+import { OPENAI_COMPAT_PROVIDERS } from "../ai/providers/openai-compatible-catalog";
 import { CODEX_MODELS } from "../ai/model-catalog";
 import { z } from "zod";
 
@@ -18,6 +19,11 @@ const StoredOAuthSchema = z.object({
 });
 
 const OAuthEntry = z.union([z.string(), StoredOAuthSchema]);
+// Catalog-driven OpenAI-compatible providers contribute their own apiKey + oauth-slot
+// schema keys (incl. hyphenated names like `alibaba-coding-plan`), so config-file keys
+// are validated/kept rather than stripped. Adding a provider = one catalog row.
+const compatKeySchema = Object.fromEntries(OPENAI_COMPAT_PROVIDERS.map(p => [p.name, z.string().optional()]));
+const compatOAuthSchema = Object.fromEntries(OPENAI_COMPAT_PROVIDERS.map(p => [p.name, OAuthEntry.optional()]));
 const HookConfigSchema = z.object({
   enabled: z.boolean().optional(),
   hooks: z
@@ -45,6 +51,9 @@ export const ConfigSchema = z
         openai: z.string().optional(),
         gemini: z.string().optional(),
         antigravity: z.string().optional(),
+        xai: z.string().optional(),
+        kimi: z.string().optional(),
+        ...compatKeySchema,
       })
       .default({}),
     oauth: z
@@ -53,6 +62,9 @@ export const ConfigSchema = z
         openai: OAuthEntry.optional(),
         gemini: OAuthEntry.optional(),
         antigravity: OAuthEntry.optional(),
+        xai: OAuthEntry.optional(),
+        kimi: OAuthEntry.optional(),
+        ...compatOAuthSchema,
       })
       .optional(),
     ollamaBaseUrl: z.string().optional(),

package/src/agent/session.ts

@@ -404,9 +404,16 @@ export async function exportSession(
     const role = m.role.charAt(0).toUpperCase() + m.role.slice(1);
     // Fence longer than the longest backtick run in the body (CommonMark) so message
     // content containing ``` doesn't prematurely close the code fence.
-    const longest = (m.content.match(/`+/g) ?? []).reduce((mx, r) => Math.max(mx, r.length), 0);
-    const fence = "`".repeat(Math.max(3, longest + 1));
-    lines.push(`## ${role}`, "", fence, m.content, fence, "");
+    const fenceFor = (s: string) => "`".repeat(Math.max(3, (s.match(/`+/g) ?? []).reduce((mx, r) => Math.max(mx, r.length), 0) + 1));
+    lines.push(`## ${role}`, "");
+    // Persisted thinking (gjc "think → answer"): include it in the durable export so the
+    // markdown record matches the in-app transcript and the JSON export.
+    if (m.role === "assistant" && m.reasoning?.trim()) {
+      const tf = fenceFor(m.reasoning);
+      lines.push("### Thinking", "", tf, m.reasoning, tf, "");
+    }
+    const fence = fenceFor(m.content);
+    lines.push(fence, m.content, fence, "");
   }
   return lines.join("\n");
 }

package/src/agent/state.ts

@@ -2,6 +2,8 @@ import * as fs from "node:fs/promises";
 import * as path from "node:path";
 import * as os from "node:os";
 import { parseConfig } from "./config-schema";
+import type { AuthProvider } from "../auth/storage";
+import { OPENAI_COMPAT_PROVIDERS } from "../ai/providers/openai-compatible-catalog";
 import { jeoEnv } from "../util/env";
 
 /** Persisted OAuth credential set (access + refresh + expiry) for a provider. */
@@ -26,27 +28,21 @@ export interface HookConfig {
 }
 
 export interface Config {
-  providers: {
-    anthropic?: string;
-    openai?: string;
-    gemini?: string;
-    antigravity?: string;
-  };
+  /** Per-provider API keys, keyed by AuthProvider (cloud keys + catalog OpenAI-compatible). */
+  providers: Partial<Record<AuthProvider, string>>;
   /**
    * OAuth credentials. `resolveCredential()` returns these before API keys so refresh
    * metadata is not lost, but provider execution/status applies the GJC parity rule:
-   * an API key is broader and wins whenever both key + OAuth exist.
+   * an API key is broader and wins whenever both key + OAuth exist. API-key-only
+   * providers never populate OAuth; the key exists for index-compatibility.
    */
-  oauth?: {
-    anthropic?: string | StoredOAuth;
-    openai?: string | StoredOAuth;
-    gemini?: string | StoredOAuth;
-    antigravity?: string | StoredOAuth;
-  };
+  oauth?: Partial<Record<AuthProvider, string | StoredOAuth>>;
   /** Base URL for the local Ollama server (keyless). */
   ollamaBaseUrl?: string;
-  /** Base URL override for OpenAI-compatible providers (LM Studio, vLLM, llama-cpp-server, ...). */
+  /** Base URL override for OpenAI-compatible providers (vLLM, llama-cpp-server, ...). */
   openaiBaseUrl?: string;
+  /** Base URL for the local LM Studio server (keyless, OpenAI-compatible). */
+  lmstudioBaseUrl?: string;
   defaultModel: string;
   theme?: string;
   thinkingLevel?: "minimal" | "low" | "medium" | "high" | "xhigh";
@@ -193,6 +189,13 @@ function withEnvOverlay(cfg: Config): Config {
   if (!providers.anthropic && process.env.ANTHROPIC_API_KEY) providers.anthropic = process.env.ANTHROPIC_API_KEY;
   if (!providers.openai && process.env.OPENAI_API_KEY) providers.openai = process.env.OPENAI_API_KEY;
   if (!providers.gemini && process.env.GEMINI_API_KEY) providers.gemini = process.env.GEMINI_API_KEY;
+  if (!providers.xai && process.env.XAI_API_KEY) providers.xai = process.env.XAI_API_KEY;
+  // Catalog-driven OpenAI-compatible providers: each provider's own `apiKeyEnv`
+  // (e.g. GROQ_API_KEY, HF_TOKEN, NANO_GPT_API_KEY) fills config.providers[name].
+  for (const def of OPENAI_COMPAT_PROVIDERS) {
+    const key = def.name as AuthProvider; // every catalog name is an AuthProvider
+    if (!providers[key] && process.env[def.apiKeyEnv]) providers[key] = process.env[def.apiKeyEnv];
+  }
   return {
     ...cfg,
     providers,
@@ -200,6 +203,7 @@ function withEnvOverlay(cfg: Config): Config {
     defaultModel: jeoEnv("DEFAULT_MODEL") || cfg.defaultModel,
     ollamaBaseUrl: cfg.ollamaBaseUrl || process.env.OLLAMA_HOST || "http://localhost:11434",
     openaiBaseUrl: cfg.openaiBaseUrl || process.env.OPENAI_BASE_URL,
+    lmstudioBaseUrl: cfg.lmstudioBaseUrl || process.env.LMSTUDIO_BASE_URL || "http://localhost:1234/v1",
     roles: {
       smol: cfg.roles?.smol || jeoEnv("SMOL_MODEL"),
       slow: cfg.roles?.slow || jeoEnv("SLOW_MODEL"),
@@ -214,6 +218,7 @@ function envDefaultConfig(): Config {
       anthropic: process.env.ANTHROPIC_API_KEY,
       openai: process.env.OPENAI_API_KEY,
       gemini: process.env.GEMINI_API_KEY,
+      xai: process.env.XAI_API_KEY,
     },
     defaultModel: jeoEnv("DEFAULT_MODEL") || DEFAULT_MODEL,
     thinkingLevel: "medium",

package/src/ai/index.ts

@@ -10,3 +10,4 @@ export { openaiAdapter } from "./providers/openai";
 export { geminiAdapter } from "./providers/gemini";
 export { antigravityAdapter } from "./providers/antigravity";
 export { ollamaAdapter } from "./providers/ollama";
+export { lmstudioAdapter } from "./providers/lmstudio";

package/src/ai/model-catalog.ts

@@ -7,6 +7,7 @@
  * catalog annotates known ids with capabilities.
  */
 import type { ProviderName } from "./types";
+import { openaiCompatDef } from "./providers/openai-compatible-catalog";
 
 export type ThinkLevel = "minimal" | "low" | "medium" | "high" | "xhigh";
 
@@ -36,6 +37,8 @@ const STD: ThinkLevel[] = ["minimal", "low", "medium", "high"];
 export const ANTIGRAVITY_MODELS = [
   "claude-opus-4-5-thinking",
   "claude-opus-4-6-thinking",
+  "claude-opus-4-8",
+  "claude-opus-4-8-thinking",
   "claude-sonnet-4-5",
   "claude-sonnet-4-5-thinking",
   "claude-sonnet-4-6",
@@ -59,6 +62,10 @@ export const MODEL_CATALOG: readonly CatalogModel[] = [
   { canonical: "claude-sonnet-4-5", provider: "anthropic", providerModel: "claude-sonnet-4-5-20250929", contextTokens: 200_000, maxOutputTokens: 64_000, thinking: FULL, images: true },
   { canonical: "claude-opus-4-1", provider: "anthropic", providerModel: "claude-opus-4-1-20250805", contextTokens: 200_000, maxOutputTokens: 32_000, thinking: FULL, images: true },
   { canonical: "claude-opus-4-5", provider: "anthropic", providerModel: "claude-opus-4-5-20251101", contextTokens: 200_000, maxOutputTokens: 64_000, thinking: FULL, images: true },
+  // NOTE: confirm exact dated provider ids when these ship publicly; the family
+  // heuristic in `catalogMetadata` keeps reasoning working even before that.
+  { canonical: "claude-opus-4-6", provider: "anthropic", providerModel: "claude-opus-4-6", contextTokens: 200_000, maxOutputTokens: 64_000, thinking: FULL, images: true },
+  { canonical: "claude-opus-4-8", provider: "anthropic", providerModel: "claude-opus-4-8", contextTokens: 200_000, maxOutputTokens: 64_000, thinking: FULL, images: true },
   // OpenAI
   { canonical: "gpt-4o", provider: "openai", providerModel: "gpt-4o", contextTokens: 128_000, maxOutputTokens: 16_384, thinking: [], images: true },
   { canonical: "gpt-4o-mini", provider: "openai", providerModel: "gpt-4o-mini", contextTokens: 128_000, maxOutputTokens: 16_384, thinking: [], images: true },
@@ -68,6 +75,16 @@ export const MODEL_CATALOG: readonly CatalogModel[] = [
   { canonical: "o4-mini", provider: "openai", providerModel: "o4-mini", contextTokens: 200_000, maxOutputTokens: 100_000, thinking: STD, images: true },
   { canonical: "gpt-5.5", provider: "openai", providerModel: "gpt-5.5", contextTokens: 400_000, maxOutputTokens: 128_000, thinking: FULL, images: true },
   { canonical: "gpt-5.4", provider: "openai", providerModel: "gpt-5.4", contextTokens: 400_000, maxOutputTokens: 128_000, thinking: FULL, images: true },
+  // xAI (Grok) — OpenAI-compatible at https://api.x.ai/v1 (XAI_API_KEY)
+  { canonical: "grok-4.3", provider: "xai", providerModel: "grok-4.3", contextTokens: 256_000, maxOutputTokens: 64_000, thinking: FULL, images: true },
+  { canonical: "grok-4-fast-reasoning", provider: "xai", providerModel: "grok-4-fast-reasoning", contextTokens: 2_000_000, maxOutputTokens: 64_000, thinking: FULL, images: true },
+  { canonical: "grok-4-fast-non-reasoning", provider: "xai", providerModel: "grok-4-fast-non-reasoning", contextTokens: 2_000_000, maxOutputTokens: 64_000, thinking: [], images: true },
+  { canonical: "grok-code-fast-1", provider: "xai", providerModel: "grok-code-fast-1", contextTokens: 256_000, maxOutputTokens: 64_000, thinking: FULL, images: false },
+  // Kimi (Moonshot) — OpenAI-compatible at https://api.moonshot.ai/v1 (KIMI_API_KEY)
+  { canonical: "kimi-k2-0711-preview", provider: "kimi", providerModel: "kimi-k2-0711-preview", contextTokens: 128_000, maxOutputTokens: 16_384, thinking: [], images: false },
+  { canonical: "kimi-thinking-preview", provider: "kimi", providerModel: "kimi-thinking-preview", contextTokens: 128_000, maxOutputTokens: 32_000, thinking: FULL, images: true },
+  { canonical: "kimi-latest", provider: "kimi", providerModel: "kimi-latest", contextTokens: 128_000, maxOutputTokens: 16_384, thinking: [], images: true },
+  { canonical: "moonshot-v1-128k", provider: "kimi", providerModel: "moonshot-v1-128k", contextTokens: 128_000, maxOutputTokens: 16_384, thinking: [], images: false },
   // Google
   { canonical: "gemini-1.5-pro", provider: "gemini", providerModel: "gemini-1.5-pro", contextTokens: 1_000_000, maxOutputTokens: 8_192, thinking: [], images: true },
   { canonical: "gemini-2.0-flash", provider: "gemini", providerModel: "gemini-2.0-flash", contextTokens: 1_000_000, maxOutputTokens: 8_192, thinking: [], images: true },
@@ -134,13 +151,111 @@ export function catalogByProvider(provider: ProviderName): CatalogModel[] {
   return MODEL_CATALOG.filter(m => m.provider === provider);
 }
 
+
+/**
+ * Heuristic capability inference for ids the static catalog does not list yet.
+ *
+ * New model revisions ship faster than this file is edited (e.g. a fresh
+ * `claude-opus-4-8` before its entry is added). Rather than treating every
+ * uncatalogued id as "no reasoning" — which silently hides the thinking TUI —
+ * we map the id to its model family and version and synthesize metadata so a
+ * brand-new revision behaves like its catalogued siblings (e.g. opus-4-6).
+ *
+ * Conservative by design: returns `undefined` for ids that do not match a known
+ * reasoning-capable family, so random/unknown ids stay "unknown caps".
+ */
+export function inferCatalogMetadata(modelId: string): CatalogModel | undefined {
+  const raw = modelId.trim();
+  if (!raw) return undefined;
+  const antigravity = /^antigravity\//i.test(raw);
+  const id = raw.replace(/^antigravity\//i, "").toLowerCase();
+
+  // Anthropic Claude: opus/sonnet/haiku. Major version >= 4 ships extended
+  // thinking (mirrors every catalogued claude-4-x entry); claude-3-x does not.
+  const claude = id.match(/^claude-(opus|sonnet|haiku)-(\d+)(?:[-.](\d+))?/);
+  if (claude) {
+    const major = Number(claude[2]);
+    const thinking = major >= 4 ? FULL : [];
+    return {
+      canonical: raw,
+      provider: antigravity ? "antigravity" : "anthropic",
+      providerModel: id,
+      contextTokens: 200_000,
+      maxOutputTokens: claude[1] === "haiku" ? 64_000 : 64_000,
+      thinking,
+      images: true,
+      company: antigravity ? "Anthropic via Antigravity" : "Anthropic",
+    };
+  }
+
+  // OpenAI reasoning families: the o-series (o1, o3, … any major incl. o10+) and
+  // gpt-5+ (digit-count agnostic so gpt-6/o10 never silently lose reasoning the way
+  // opus-4-8 did). gpt-4 and earlier are non-reasoning. Mirrors the openai.ts gate.
+  const gptMajor = id.match(/^gpt-(\d+)/);
+  const openaiReasoner = /^o\d+(-|$)/.test(id) || (gptMajor ? Number(gptMajor[1]) >= 5 : false);
+  if (openaiReasoner) {
+    const wide = gptMajor ? Number(gptMajor[1]) >= 5 : false;
+    return {
+      canonical: raw,
+      provider: antigravity ? "antigravity" : "openai",
+      providerModel: id,
+      contextTokens: wide ? 400_000 : 200_000,
+      maxOutputTokens: wide ? 128_000 : 100_000,
+      thinking: wide ? FULL : STD,
+      images: !id.includes("mini") || id.includes("o4-mini") || id.includes("o3"),
+      company: antigravity ? "OpenAI via Antigravity" : "OpenAI",
+    };
+  }
+
+  // Google Gemini: 2.5+ and 3.x expose thinking; 1.5/2.0 do not.
+  const gemini = id.match(/^gemini-(\d+)(?:\.(\d+))?/);
+  if (gemini) {
+    const major = Number(gemini[1]);
+    const minor = Number(gemini[2] ?? 0);
+    const reasons = major >= 3 || (major === 2 && minor >= 5);
+    const big3 = major >= 3;
+    return {
+      canonical: raw,
+      provider: antigravity ? "antigravity" : "gemini",
+      providerModel: id,
+      contextTokens: 1_000_000,
+      maxOutputTokens: 65_536,
+      thinking: !reasons ? [] : big3 || id.includes("thinking") || id.includes("-high") || id.includes("-low") ? FULL : STD,
+      images: true,
+      company: antigravity ? "Google Antigravity" : "Google",
+    };
+  }
+
+  // xAI Grok 4+ reasoning variants.
+  const grok = id.match(/^grok-(\d+)/);
+  if (grok && Number(grok[1]) >= 4) {
+    const nonReasoning = id.includes("non-reasoning");
+    return {
+      canonical: raw,
+      provider: "xai",
+      providerModel: id,
+      contextTokens: id.includes("fast") ? 2_000_000 : 256_000,
+      maxOutputTokens: 64_000,
+      thinking: nonReasoning ? [] : FULL,
+      images: !id.includes("code"),
+      company: "xAI",
+    };
+  }
+
+  return undefined;
+}
+
 /** Annotate a discovered/raw model id with catalog metadata, when known. */
 export function catalogMetadata(modelId: string): CatalogModel | undefined {
   const direct = findCatalogModel(modelId);
   if (direct) return direct;
   // Tolerate provider-prefixed or bare provider model ids.
   const bare = modelId.replace(/^[a-z-]+\//, "");
-  return MODEL_CATALOG.find(m => m.providerModel === bare || m.providerModel.endsWith(`/${bare}`) || m.canonical === bare);
+  const hit = MODEL_CATALOG.find(m => m.providerModel === bare || m.providerModel.endsWith(`/${bare}`) || m.canonical === bare);
+  if (hit) return hit;
+  // Last resort: infer capabilities from the model family so a brand-new
+  // revision still surfaces reasoning/thinking like its catalogued siblings.
+  return inferCatalogMetadata(modelId);
 }
 
 /** Whether a model supports a given thinking level (per the catalog). */
@@ -157,6 +272,11 @@ export function companyLabel(provider: string, entry?: { company?: string }): st
   if (low === "openai") return "OpenAI";
   if (low === "gemini") return "Google";
   if (low === "ollama") return "Ollama";
+  if (low === "lmstudio") return "LM Studio";
+  if (low === "xai") return "xAI";
+  if (low === "kimi") return "Moonshot";
+  const compat = openaiCompatDef(low);
+  if (compat) return compat.label;
   if (low === "antigravity") return "Antigravity";
   return provider.charAt(0).toUpperCase() + provider.slice(1);
 }

package/src/ai/model-discovery.ts

@@ -13,6 +13,7 @@ import type { ProviderName } from "./types";
 import { PROVIDER_NAMES } from "./provider-status";
 import { catalogByProvider, CODEX_MODELS } from "./model-catalog";
 import { extractChatgptAccountId } from "./providers/openai-responses";
+import { openaiCompatDef } from "./providers/openai-compatible-catalog";
 
 export interface ProviderModelsResult {
   provider: ProviderName;
@@ -68,7 +69,9 @@ function anthropicHeaders(cred: Credential): Record<string, string> {
 }
 
 function authProviderFor(provider: ProviderName): AuthProvider | undefined {
-  if (provider === "ollama") return undefined;
+  // Local providers (ollama/lmstudio) are keyless and do not resolve through the
+  // auth core. API-key providers (incl. xai/kimi) DO — so discovery sends their key.
+  if (provider === "ollama" || provider === "lmstudio") return undefined;
   return provider;
 }
 
@@ -78,6 +81,17 @@ export function discoveryRequest(
   cred: Credential | undefined,
   baseUrl?: string,
 ): { url: string; headers: Record<string, string>; method?: "GET" | "POST"; body?: string } {
+  // Catalog-driven compat providers: OpenAI `${base}/models` (Bearer) or Anthropic
+  // `${base}/v1/models` (x-api-key). Both return { data: [{ id }] }.
+  const compat = openaiCompatDef(provider);
+  if (compat) {
+    const base = (baseUrl ?? compat.baseUrl).replace(/\/$/, "");
+    const token = cred?.kind === "api_key" || cred?.kind === "oauth" ? cred.token : "";
+    if (compat.protocol === "anthropic") {
+      return { url: `${base}/v1/models`, headers: token ? { "x-api-key": token, "anthropic-version": "2023-06-01" } : {} };
+    }
+    return { url: `${base}/models`, headers: token ? { Authorization: `Bearer ${token}` } : {} };
+  }
   switch (provider) {
     case "anthropic":
       return { url: "https://api.anthropic.com/v1/models", headers: anthropicHeaders(cred!) };
@@ -120,7 +134,21 @@ export function discoveryRequest(
       const base = (baseUrl ?? "http://localhost:11434").replace(/\/$/, "");
       return { url: `${base}/api/tags`, headers: {} };
     }
+    case "lmstudio": {
+      const base = (baseUrl ?? "http://localhost:1234/v1").replace(/\/$/, "");
+      return { url: `${base}/models`, headers: {} };
+    }
+    case "xai": {
+      const token = cred?.kind === "api_key" ? cred.token : "";
+      return { url: "https://api.x.ai/v1/models", headers: token ? { Authorization: `Bearer ${token}` } : {} };
+    }
+    case "kimi": {
+      const token = cred?.kind === "api_key" ? cred.token : "";
+      return { url: "https://api.moonshot.ai/v1/models", headers: token ? { Authorization: `Bearer ${token}` } : {} };
+    }
   }
+  // Unreachable: every ProviderName is a switch case or catalog-handled above.
+  throw new Error(`discoveryRequest: unhandled provider '${provider}'`);
 }
 
 /**
@@ -149,9 +177,26 @@ export function parseModelsBody(provider: ProviderName, body: unknown): string[]
     data?: { id?: string }[];
     models?: ({ name?: string; supportedGenerationMethods?: string[] } & CodexModelRow)[];
   };
+  if (openaiCompatDef(provider)) {
+    // Catalog OpenAI-compatible: { data: [{ id }] }. Prefix-qualify so the router
+    // maps the id back to this provider (the ids alone don't heuristically route).
+    return (data.data ?? []).map(m => (m.id ? `${provider}/${m.id}` : "")).filter(Boolean);
+  }
   if (provider === "ollama") {
     return (data.models ?? []).map(m => `ollama/${m.name ?? ""}`).filter(s => s !== "ollama/");
   }
+  if (provider === "lmstudio") {
+    // LM Studio is OpenAI-compatible: { data: [{ id }] }. Qualify with the routing prefix.
+    return (data.data ?? []).map(m => `lmstudio/${m.id ?? ""}`).filter(s => s !== "lmstudio/");
+  }
+  if (provider === "xai") {
+    // xAI is OpenAI-compatible: { data: [{ id }] }. Grok ids route to xai by name, so no prefix.
+    return (data.data ?? []).map(m => m.id ?? "").filter(Boolean);
+  }
+  if (provider === "kimi") {
+    // Moonshot is OpenAI-compatible: { data: [{ id }] }. kimi/moonshot ids route by name.
+    return (data.data ?? []).map(m => m.id ?? "").filter(Boolean);
+  }
   if (provider === "antigravity") {
     // fetchAvailableModels keys the map by the CALLABLE model id (e.g.
     // "gemini-3-flash"); the entry's `model` field is an internal enum
@@ -252,7 +297,14 @@ export async function listProviderModels(
 
   let cred: Credential | undefined;
   let source: ProviderModelsResult["source"] = "keyless";
-  if (provider !== "ollama") {
+  if (provider === "xai") {
+    // xAI (Grok) is API-key only and not an OAuth AuthProvider: resolve its key
+    // directly from config/env instead of the AuthProvider credential store.
+    const key = (opts.config ?? (await readGlobalConfig())).providers?.xai;
+    if (!key) return { provider, models: [], ok: false, source: "none", error: "not logged in" };
+    cred = { kind: "api_key", provider: "openai", token: key };
+    source = "api_key";
+  } else if (provider !== "ollama" && provider !== "lmstudio") {
     const authProvider = authProviderFor(provider);
     const raw = await resolveCredential(authProvider!);
     cred = raw;
@@ -338,7 +390,7 @@ export async function discoverModels(
       listProviderModels(p, {
         ...opts,
         config: cfg,
-        baseUrl: p === "ollama" ? (cfg.ollamaBaseUrl ?? opts.baseUrl) : p === "openai" ? (cfg.openaiBaseUrl ?? opts.baseUrl) : opts.baseUrl,
+        baseUrl: p === "ollama" ? (cfg.ollamaBaseUrl ?? opts.baseUrl) : p === "lmstudio" ? (cfg.lmstudioBaseUrl ?? opts.baseUrl) : p === "openai" ? (cfg.openaiBaseUrl ?? opts.baseUrl) : opts.baseUrl,
       }),
     ),
   );