Merge origin/main into provider-setup-wizard

2026-04-02 18:03:44 +05:30
parent 08f0b6030e 4260f5bcd7
commit 71a3f36e95
40 changed files with 1801 additions and 132 deletions
--- a/.github/workflows/pr-checks.yml
+++ b/.github/workflows/pr-checks.yml
@@ -12,15 +12,15 @@ jobs:

    steps:
      - name: Check out repository
-        uses: actions/checkout@v4
+        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2

      - name: Set up Node.js
-        uses: actions/setup-node@v4
+        uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4.4.0
        with:
          node-version: 22

      - name: Set up Bun
-        uses: oven-sh/setup-bun@v2
+        uses: oven-sh/setup-bun@4bc047ad259df6fc24a6c9b0f9a0cb08cf17fbe5 # v2.0.1
        with:
          bun-version: 1.3.11

--- a/README.md
+++ b/README.md
@@ -2,7 +2,7 @@

 Use Claude Code with **any LLM** — not just Claude.

-OpenClaude is a fork of the [Claude Code source leak](https://gitlawb.com/node/repos/z6MkgKkb/instructkr-claude-code) (exposed via npm source maps on March 31, 2026). We added an OpenAI-compatible provider shim so you can plug in GPT-4o, DeepSeek, Gemini, Llama, Mistral, or any model that speaks the OpenAI chat completions API. It now also supports the ChatGPT Codex backend for `codexplan` and `codexspark`.
+OpenClaude is a fork of the [Claude Code source leak](https://gitlawb.com/node/repos/z6MkgKkb/instructkr-claude-code) (exposed via npm source maps on March 31, 2026). We added an OpenAI-compatible provider shim so you can plug in GPT-4o, DeepSeek, Gemini, Llama, Mistral, or any model that speaks the OpenAI chat completions API. It now also supports the ChatGPT Codex backend for `codexplan` and `codexspark`, and local inference via [Atomic Chat](https://atomic.chat/) on Apple Silicon.

 All of Claude Code's tools work — bash, file read/write/edit, grep, glob, agents, tasks, MCP — just powered by whatever model you choose.

@@ -122,9 +122,13 @@ export OPENAI_MODEL=deepseek-chat
 export CLAUDE_CODE_USE_OPENAI=1
 export OPENAI_API_KEY=sk-or-...
 export OPENAI_BASE_URL=https://openrouter.ai/api/v1
-export OPENAI_MODEL=google/gemini-2.0-flash
+export OPENAI_MODEL=google/gemini-2.0-flash-001
 ```

+OpenRouter model availability changes over time. If a model stops working,
+pick another currently available OpenRouter model before assuming the
+OpenAI-compatible setup is broken.
+
 ### Ollama (local, free)

 ```bash
@@ -136,6 +140,23 @@ export OPENAI_MODEL=llama3.3:70b
 # no API key needed for local models
 ```

+### Atomic Chat (local, Apple Silicon)
+
+```bash
+export CLAUDE_CODE_USE_OPENAI=1
+export OPENAI_BASE_URL=http://127.0.0.1:1337/v1
+export OPENAI_MODEL=your-model-name
+# no API key needed for local models
+```
+
+Or use the profile launcher:
+
+```bash
+bun run dev:atomic-chat
+```
+
+Download Atomic Chat from [atomic.chat](https://atomic.chat/). The app must be running with a model loaded before launching.
+
 ### LM Studio (local)

 ```bash
@@ -187,7 +208,7 @@ export OPENAI_MODEL=gpt-4o
 | Variable | Required | Description |
 |----------|----------|-------------|
 | `CLAUDE_CODE_USE_OPENAI` | Yes | Set to `1` to enable the OpenAI provider |
-| `OPENAI_API_KEY` | Yes* | Your API key (*not needed for local models like Ollama) |
+| `OPENAI_API_KEY` | Yes* | Your API key (*not needed for local models like Ollama/Atomic Chat) |
 | `OPENAI_MODEL` | Yes | Model name (e.g. `gpt-4o`, `deepseek-chat`, `llama3.3:70b`) |
 | `OPENAI_BASE_URL` | No | API endpoint (defaults to `https://api.openai.com/v1`) |
 | `CODEX_API_KEY` | Codex only | Codex/ChatGPT access token override |
@@ -250,6 +271,9 @@ bun run profile:codex
 # openai bootstrap with explicit key
 bun run profile:init -- --provider openai --api-key sk-...

+# atomic-chat bootstrap (auto-detects running model)
+bun run profile:init -- --provider atomic-chat
+
 # ollama bootstrap with custom model
 bun run profile:init -- --provider ollama --model llama3.1:8b

@@ -270,6 +294,9 @@ bun run dev:openai

 # Ollama profile (defaults: localhost:11434, llama3.1:8b)
 bun run dev:ollama
+
+# Atomic Chat profile (Apple Silicon local LLMs at 127.0.0.1:1337)
+bun run dev:atomic-chat
 ```

 `profile:recommend` ranks installed Ollama models for `latency`, `balanced`, or `coding`, and `profile:auto` can persist the recommendation directly.
@@ -280,8 +307,9 @@ Goal-based Ollama selection only recommends among models that are already instal

 Use `profile:codex` or `--provider codex` when you want the ChatGPT Codex backend.

-`dev:openai`, `dev:ollama`, and `dev:codex` run `doctor:runtime` first and only launch the app if checks pass.
+`dev:openai`, `dev:ollama`, `dev:atomic-chat`, and `dev:codex` run `doctor:runtime` first and only launch the app if checks pass.
 For `dev:ollama`, make sure Ollama is running locally before launch.
+For `dev:atomic-chat`, make sure Atomic Chat is running with a model loaded.

 ---

--- a/atomic_chat_provider.py
+++ b/atomic_chat_provider.py
@@ -0,0 +1,146 @@
+"""
+atomic_chat_provider.py
+-----------------------
+Adds native Atomic Chat support to openclaude.
+Lets Claude Code route requests to any locally-running model via
+Atomic Chat (Apple Silicon only) at 127.0.0.1:1337.
+
+Atomic Chat exposes an OpenAI-compatible API, so messages are forwarded
+directly without translation.
+
+Usage (.env):
+    PREFERRED_PROVIDER=atomic-chat
+    ATOMIC_CHAT_BASE_URL=http://127.0.0.1:1337
+"""
+
+import httpx
+import json
+import logging
+import os
+from typing import AsyncIterator
+
+logger = logging.getLogger(__name__)
+ATOMIC_CHAT_BASE_URL = os.getenv("ATOMIC_CHAT_BASE_URL", "http://127.0.0.1:1337")
+
+
+def _api_url(path: str) -> str:
+    return f"{ATOMIC_CHAT_BASE_URL}/v1{path}"
+
+
+async def check_atomic_chat_running() -> bool:
+    try:
+        async with httpx.AsyncClient(timeout=3.0) as client:
+            resp = await client.get(_api_url("/models"))
+            return resp.status_code == 200
+    except Exception:
+        return False
+
+
+async def list_atomic_chat_models() -> list[str]:
+    try:
+        async with httpx.AsyncClient(timeout=5.0) as client:
+            resp = await client.get(_api_url("/models"))
+            resp.raise_for_status()
+            data = resp.json()
+            return [m["id"] for m in data.get("data", [])]
+    except Exception as e:
+        logger.warning(f"Could not list Atomic Chat models: {e}")
+        return []
+
+
+async def atomic_chat(
+    model: str,
+    messages: list[dict],
+    system: str | None = None,
+    max_tokens: int = 4096,
+    temperature: float = 1.0,
+) -> dict:
+    chat_messages = list(messages)
+    if system:
+        chat_messages.insert(0, {"role": "system", "content": system})
+
+    payload = {
+        "model": model,
+        "messages": chat_messages,
+        "max_tokens": max_tokens,
+        "temperature": temperature,
+        "stream": False,
+    }
+
+    async with httpx.AsyncClient(timeout=120.0) as client:
+        resp = await client.post(_api_url("/chat/completions"), json=payload)
+        resp.raise_for_status()
+        data = resp.json()
+
+    choice = data.get("choices", [{}])[0]
+    assistant_text = choice.get("message", {}).get("content", "")
+    usage = data.get("usage", {})
+
+    return {
+        "id": data.get("id", "msg_atomic_chat"),
+        "type": "message",
+        "role": "assistant",
+        "content": [{"type": "text", "text": assistant_text}],
+        "model": model,
+        "stop_reason": "end_turn",
+        "stop_sequence": None,
+        "usage": {
+            "input_tokens": usage.get("prompt_tokens", 0),
+            "output_tokens": usage.get("completion_tokens", 0),
+        },
+    }
+
+
+async def atomic_chat_stream(
+    model: str,
+    messages: list[dict],
+    system: str | None = None,
+    max_tokens: int = 4096,
+    temperature: float = 1.0,
+) -> AsyncIterator[str]:
+    chat_messages = list(messages)
+    if system:
+        chat_messages.insert(0, {"role": "system", "content": system})
+
+    payload = {
+        "model": model,
+        "messages": chat_messages,
+        "max_tokens": max_tokens,
+        "temperature": temperature,
+        "stream": True,
+    }
+
+    yield "event: message_start\n"
+    yield f'data: {json.dumps({"type": "message_start", "message": {"id": "msg_atomic_chat_stream", "type": "message", "role": "assistant", "content": [], "model": model, "stop_reason": None, "usage": {"input_tokens": 0, "output_tokens": 0}}})}\n\n'
+    yield "event: content_block_start\n"
+    yield f'data: {json.dumps({"type": "content_block_start", "index": 0, "content_block": {"type": "text", "text": ""}})}\n\n'
+
+    async with httpx.AsyncClient(timeout=120.0) as client:
+        async with client.stream("POST", _api_url("/chat/completions"), json=payload) as resp:
+            resp.raise_for_status()
+            async for line in resp.aiter_lines():
+                if not line or not line.startswith("data: "):
+                    continue
+                raw = line[len("data: "):]
+                if raw.strip() == "[DONE]":
+                    break
+                try:
+                    chunk = json.loads(raw)
+                    delta = chunk.get("choices", [{}])[0].get("delta", {})
+                    delta_text = delta.get("content", "")
+                    if delta_text:
+                        yield "event: content_block_delta\n"
+                        yield f'data: {json.dumps({"type": "content_block_delta", "index": 0, "delta": {"type": "text_delta", "text": delta_text}})}\n\n'
+
+                    finish_reason = chunk.get("choices", [{}])[0].get("finish_reason")
+                    if finish_reason:
+                        usage = chunk.get("usage", {})
+                        yield "event: content_block_stop\n"
+                        yield f'data: {json.dumps({"type": "content_block_stop", "index": 0})}\n\n'
+                        yield "event: message_delta\n"
+                        yield f'data: {json.dumps({"type": "message_delta", "delta": {"stop_reason": "end_turn", "stop_sequence": None}, "usage": {"output_tokens": usage.get("completion_tokens", 0)}})}\n\n'
+                        yield "event: message_stop\n"
+                        yield f'data: {json.dumps({"type": "message_stop"})}\n\n'
+                        break
+                except json.JSONDecodeError:
+                    continue
--- a/package.json
+++ b/package.json
@@ -21,6 +21,7 @@
    "dev:gemini": "bun run scripts/provider-launch.ts gemini",
    "dev:ollama": "bun run scripts/provider-launch.ts ollama",
    "dev:ollama:fast": "bun run scripts/provider-launch.ts ollama --fast --bare",
+    "dev:atomic-chat": "bun run scripts/provider-launch.ts atomic-chat",
    "profile:init": "bun run scripts/provider-bootstrap.ts",
    "profile:recommend": "bun run scripts/provider-recommend.ts",
    "profile:auto": "bun run scripts/provider-recommend.ts --apply",
@@ -30,7 +31,7 @@
    "dev:fast": "bun run profile:fast && bun run dev:ollama:fast",
    "dev:code": "bun run profile:code && bun run dev:profile",
    "start": "node dist/cli.mjs",
-    "test:provider-recommendation": "node --test --experimental-strip-types src/utils/providerRecommendation.test.ts src/utils/providerProfile.test.ts",
+    "test:provider-recommendation": "bun test src/utils/providerRecommendation.test.ts src/utils/providerProfile.test.ts",
    "typecheck": "tsc --noEmit",
    "smoke": "bun run build && node dist/cli.mjs --version",
    "test:provider": "bun test src/services/api/*.test.ts src/utils/context.test.ts",
--- a/scripts/provider-bootstrap.ts
+++ b/scripts/provider-bootstrap.ts
@@ -8,6 +8,7 @@ import {
  recommendOllamaModel,
 } from '../src/utils/providerRecommendation.ts'
 import {
+  buildAtomicChatProfileEnv,
  buildCodexProfileEnv,
  buildGeminiProfileEnv,
  buildOllamaProfileEnv,
@@ -19,8 +20,11 @@ import {
  type ProviderProfile,
 } from '../src/utils/providerProfile.ts'
 import {
+  getAtomicChatChatBaseUrl,
  getOllamaChatBaseUrl,
+  hasLocalAtomicChat,
  hasLocalOllama,
+  listAtomicChatModels,
  listOllamaModels,
 } from './provider-discovery.ts'

@@ -33,7 +37,7 @@ function parseArg(name: string): string | null {

 function parseProviderArg(): ProviderProfile | 'auto' {
  const p = parseArg('--provider')?.toLowerCase()
-  if (p === 'openai' || p === 'ollama' || p === 'codex' || p === 'gemini') return p
+  if (p === 'openai' || p === 'ollama' || p === 'codex' || p === 'gemini' || p === 'atomic-chat') return p
  return 'auto'
 }

@@ -101,6 +105,21 @@ async function main(): Promise<void> {
        getOllamaChatBaseUrl,
      },
    )
+  } else if (selected === 'atomic-chat') {
+    const model = argModel || (await listAtomicChatModels(argBaseUrl || undefined))[0]
+    if (!model) {
+      if (!(await hasLocalAtomicChat(argBaseUrl || undefined))) {
+        console.error('Atomic Chat is not running (could not connect to 127.0.0.1:1337).\n  Download from https://atomic.chat/ and launch the application.')
+      } else {
+        console.error('Atomic Chat is running but no model is loaded. Open Atomic Chat and download or start a model first.')
+      }
+      process.exit(1)
+    }
+
+    env = buildAtomicChatProfileEnv(model, {
+      baseUrl: argBaseUrl,
+      getAtomicChatChatBaseUrl,
+    })
  } else if (selected === 'codex') {
    const builtEnv = buildCodexProfileEnv({
      model: argModel,
--- a/scripts/provider-discovery.ts
+++ b/scripts/provider-discovery.ts
@@ -1,8 +1,13 @@
 export {
  benchmarkOllamaModel,
+  DEFAULT_ATOMIC_CHAT_BASE_URL,
  DEFAULT_OLLAMA_BASE_URL,
+  getAtomicChatApiBaseUrl,
+  getAtomicChatChatBaseUrl,
  getOllamaApiBaseUrl,
  getOllamaChatBaseUrl,
+  hasLocalAtomicChat,
  hasLocalOllama,
+  listAtomicChatModels,
  listOllamaModels,
 } from '../src/utils/providerDiscovery.ts'
--- a/scripts/provider-launch.ts
+++ b/scripts/provider-launch.ts
@@ -15,8 +15,11 @@ import {
  type ProviderProfile,
 } from '../src/utils/providerProfile.ts'
 import {
+  getAtomicChatChatBaseUrl,
  getOllamaChatBaseUrl,
+  hasLocalAtomicChat,
  hasLocalOllama,
+  listAtomicChatModels,
  listOllamaModels,
 } from './provider-discovery.ts'

@@ -47,7 +50,7 @@ function parseLaunchOptions(argv: string[]): LaunchOptions {
      continue
    }

-    if ((lower === 'auto' || lower === 'openai' || lower === 'ollama' || lower === 'codex' || lower === 'gemini') && requestedProfile === 'auto') {
+    if ((lower === 'auto' || lower === 'openai' || lower === 'ollama' || lower === 'codex' || lower === 'gemini' || lower === 'atomic-chat') && requestedProfile === 'auto') {
      requestedProfile = lower as ProviderProfile | 'auto'
      continue
    }
@@ -85,6 +88,11 @@ async function resolveOllamaDefaultModel(
  return recommended?.name ?? null
 }

+async function resolveAtomicChatDefaultModel(): Promise<string | null> {
+  const models = await listAtomicChatModels()
+  return models[0] ?? null
+}
+
 function runCommand(command: string, env: NodeJS.ProcessEnv): Promise<number> {
  return runProcess(command, [], env)
 }
@@ -121,6 +129,10 @@ function printSummary(profile: ProviderProfile, env: NodeJS.ProcessEnv): void {
    console.log(`OPENAI_BASE_URL=${env.OPENAI_BASE_URL}`)
    console.log(`OPENAI_MODEL=${env.OPENAI_MODEL}`)
    console.log(`CODEX_API_KEY_SET=${Boolean(resolveCodexApiCredentials(env).apiKey)}`)
+  } else if (profile === 'atomic-chat') {
+    console.log(`OPENAI_BASE_URL=${env.OPENAI_BASE_URL}`)
+    console.log(`OPENAI_MODEL=${env.OPENAI_MODEL}`)
+    console.log('OPENAI_API_KEY_SET=false (local provider, no key required)')
  } else {
    console.log(`OPENAI_BASE_URL=${env.OPENAI_BASE_URL}`)
    console.log(`OPENAI_MODEL=${env.OPENAI_MODEL}`)
@@ -132,7 +144,7 @@ async function main(): Promise<void> {
  const options = parseLaunchOptions(process.argv.slice(2))
  const requestedProfile = options.requestedProfile
  if (!requestedProfile) {
-    console.error('Usage: bun run scripts/provider-launch.ts [openai|ollama|codex|gemini|auto] [--fast] [--goal <latency|balanced|coding>] [-- <cli args>]')
+    console.error('Usage: bun run scripts/provider-launch.ts [openai|ollama|codex|gemini|atomic-chat|auto] [--fast] [--goal <latency|balanced|coding>] [-- <cli args>]')
    process.exit(1)
  }

@@ -164,12 +176,30 @@ async function main(): Promise<void> {
    }
  }

+  let resolvedAtomicChatModel: string | null = null
+  if (
+    profile === 'atomic-chat' &&
+    (persisted?.profile !== 'atomic-chat' || !persisted?.env?.OPENAI_MODEL)
+  ) {
+    if (!(await hasLocalAtomicChat())) {
+      console.error('Atomic Chat is not running (could not connect to 127.0.0.1:1337).\n  Download from https://atomic.chat/ and launch the application.')
+      process.exit(1)
+    }
+    resolvedAtomicChatModel = await resolveAtomicChatDefaultModel()
+    if (!resolvedAtomicChatModel) {
+      console.error('Atomic Chat is running but no model is loaded. Open Atomic Chat and download or start a model first.')
+      process.exit(1)
+    }
+  }
+
  const env = await buildLaunchEnv({
    profile,
    persisted,
    goal: options.goal,
    getOllamaChatBaseUrl,
    resolveOllamaDefaultModel: async () => resolvedOllamaModel || 'llama3.1:8b',
+    getAtomicChatChatBaseUrl,
+    resolveAtomicChatDefaultModel: async () => resolvedAtomicChatModel,
  })
  if (options.fast) {
    applyFastFlags(env)
--- a/scripts/system-check.ts
+++ b/scripts/system-check.ts
@@ -93,11 +93,15 @@ function isLocalBaseUrl(baseUrl: string): boolean {
 }

 const GEMINI_DEFAULT_BASE_URL = 'https://generativelanguage.googleapis.com/v1beta/openai'
+const GITHUB_MODELS_DEFAULT_BASE = 'https://models.github.ai/inference'

 function currentBaseUrl(): string {
  if (isTruthy(process.env.CLAUDE_CODE_USE_GEMINI)) {
    return process.env.GEMINI_BASE_URL ?? GEMINI_DEFAULT_BASE_URL
  }
+  if (isTruthy(process.env.CLAUDE_CODE_USE_GITHUB)) {
+    return process.env.OPENAI_BASE_URL ?? GITHUB_MODELS_DEFAULT_BASE
+  }
  return process.env.OPENAI_BASE_URL ?? 'https://api.openai.com/v1'
 }

@@ -126,15 +130,47 @@ function checkGeminiEnv(): CheckResult[] {
  return results
 }

+function checkGithubEnv(): CheckResult[] {
+  const results: CheckResult[] = []
+  const baseUrl = process.env.OPENAI_BASE_URL ?? GITHUB_MODELS_DEFAULT_BASE
+  results.push(pass('Provider mode', 'GitHub Models provider enabled.'))
+
+  const token = process.env.GITHUB_TOKEN ?? process.env.GH_TOKEN
+  if (!token?.trim()) {
+    results.push(fail('GITHUB_TOKEN', 'Missing. Set GITHUB_TOKEN or GH_TOKEN.'))
+  } else {
+    results.push(pass('GITHUB_TOKEN', 'Configured.'))
+  }
+
+  if (!process.env.OPENAI_MODEL) {
+    results.push(
+      pass(
+        'OPENAI_MODEL',
+        'Not set. Default github:copilot → openai/gpt-4.1 at runtime.',
+      ),
+    )
+  } else {
+    results.push(pass('OPENAI_MODEL', process.env.OPENAI_MODEL))
+  }
+
+  results.push(pass('OPENAI_BASE_URL', baseUrl))
+  return results
+}
+
 function checkOpenAIEnv(): CheckResult[] {
  const results: CheckResult[] = []
  const useGemini = isTruthy(process.env.CLAUDE_CODE_USE_GEMINI)
+  const useGithub = isTruthy(process.env.CLAUDE_CODE_USE_GITHUB)
  const useOpenAI = isTruthy(process.env.CLAUDE_CODE_USE_OPENAI)

  if (useGemini) {
    return checkGeminiEnv()
  }

+  if (useGithub && !useOpenAI) {
+    return checkGithubEnv()
+  }
+
  if (!useOpenAI) {
    results.push(pass('Provider mode', 'Anthropic login flow enabled (CLAUDE_CODE_USE_OPENAI is off).'))
    return results
@@ -181,12 +217,21 @@ function checkOpenAIEnv(): CheckResult[] {
  }

  const key = process.env.OPENAI_API_KEY
+  const githubToken = process.env.GITHUB_TOKEN ?? process.env.GH_TOKEN
  if (key === 'SUA_CHAVE') {
    results.push(fail('OPENAI_API_KEY', 'Placeholder value detected: SUA_CHAVE.'))
-  } else if (!key && !isLocalBaseUrl(request.baseUrl)) {
+  } else if (
+    !key &&
+    !isLocalBaseUrl(request.baseUrl) &&
+    !(useGithub && githubToken?.trim())
+  ) {
    results.push(fail('OPENAI_API_KEY', 'Missing key for non-local provider URL.'))
+  } else if (!key && useGithub && githubToken?.trim()) {
+    results.push(
+      pass('OPENAI_API_KEY', 'Not set; GITHUB_TOKEN/GH_TOKEN will be used for GitHub Models.'),
+    )
  } else if (!key) {
-    results.push(pass('OPENAI_API_KEY', 'Not set (allowed for local providers like Ollama/LM Studio).'))
+    results.push(pass('OPENAI_API_KEY', 'Not set (allowed for local providers like Atomic Chat/Ollama/LM Studio).'))
  } else {
    results.push(pass('OPENAI_API_KEY', 'Configured.'))
  }
@@ -197,11 +242,19 @@ function checkOpenAIEnv(): CheckResult[] {
 async function checkBaseUrlReachability(): Promise<CheckResult> {
  const useGemini = isTruthy(process.env.CLAUDE_CODE_USE_GEMINI)
  const useOpenAI = isTruthy(process.env.CLAUDE_CODE_USE_OPENAI)
+  const useGithub = isTruthy(process.env.CLAUDE_CODE_USE_GITHUB)

-  if (!useGemini && !useOpenAI) {
+  if (!useGemini && !useOpenAI && !useGithub) {
    return pass('Provider reachability', 'Skipped (OpenAI-compatible mode disabled).')
  }

+  if (useGithub) {
+    return pass(
+      'Provider reachability',
+      'Skipped for GitHub Models (inference endpoint differs from OpenAI /models probe).',
+    )
+  }
+
  const geminiBaseUrl = 'https://generativelanguage.googleapis.com/v1beta/openai'
  const resolvedBaseUrl = useGemini
    ? (process.env.GEMINI_BASE_URL ?? geminiBaseUrl)
@@ -271,8 +324,21 @@ async function checkBaseUrlReachability(): Promise<CheckResult> {
  }
 }

+function isAtomicChatUrl(baseUrl: string): boolean {
+  try {
+    const parsed = new URL(baseUrl)
+    return parsed.port === '1337' && isLocalBaseUrl(baseUrl)
+  } catch {
+    return false
+  }
+}
+
 function checkOllamaProcessorMode(): CheckResult {
-  if (!isTruthy(process.env.CLAUDE_CODE_USE_OPENAI) || isTruthy(process.env.CLAUDE_CODE_USE_GEMINI)) {
+  if (
+    !isTruthy(process.env.CLAUDE_CODE_USE_OPENAI) ||
+    isTruthy(process.env.CLAUDE_CODE_USE_GEMINI) ||
+    isTruthy(process.env.CLAUDE_CODE_USE_GITHUB)
+  ) {
    return pass('Ollama processor mode', 'Skipped (OpenAI-compatible mode disabled).')
  }

@@ -281,6 +347,10 @@ function checkOllamaProcessorMode(): CheckResult {
    return pass('Ollama processor mode', 'Skipped (provider URL is not local).')
  }

+  if (isAtomicChatUrl(baseUrl)) {
+    return pass('Ollama processor mode', 'Skipped (Atomic Chat local provider detected, not Ollama).')
+  }
+
  const result = spawnSync('ollama', ['ps'], {
    cwd: process.cwd(),
    encoding: 'utf8',
@@ -319,6 +389,22 @@ function serializeSafeEnvSummary(): Record<string, string | boolean> {
      GEMINI_API_KEY_SET: Boolean(process.env.GEMINI_API_KEY ?? process.env.GOOGLE_API_KEY),
    }
  }
+  if (
+    isTruthy(process.env.CLAUDE_CODE_USE_GITHUB) &&
+    !isTruthy(process.env.CLAUDE_CODE_USE_OPENAI)
+  ) {
+    return {
+      CLAUDE_CODE_USE_GITHUB: true,
+      OPENAI_MODEL:
+        process.env.OPENAI_MODEL ??
+        '(unset, default: github:copilot → openai/gpt-4.1)',
+      OPENAI_BASE_URL:
+        process.env.OPENAI_BASE_URL ?? GITHUB_MODELS_DEFAULT_BASE,
+      GITHUB_TOKEN_SET: Boolean(
+        process.env.GITHUB_TOKEN ?? process.env.GH_TOKEN,
+      ),
+    }
+  }
  const request = resolveProviderRequest({
    model: process.env.OPENAI_MODEL,
    baseUrl: process.env.OPENAI_BASE_URL,
@@ -374,6 +460,13 @@ async function main(): Promise<void> {
  const options = parseOptions(process.argv.slice(2))
  const results: CheckResult[] = []

+  const { enableConfigs } = await import('../src/utils/config.js')
+  enableConfigs()
+  const { applySafeConfigEnvironmentVariables } = await import('../src/utils/managedEnv.js')
+  applySafeConfigEnvironmentVariables()
+  const { hydrateGithubModelsTokenFromSecureStorage } = await import('../src/utils/githubModelsCredentials.js')
+  hydrateGithubModelsTokenFromSecureStorage()
+
  results.push(checkNodeVersion())
  results.push(checkBunRuntime())
  results.push(checkBuildArtifacts())
--- a/smart_router.py
+++ b/smart_router.py
@@ -57,8 +57,8 @@ class Provider:
    @property
    def is_configured(self) -> bool:
        """True if the provider has an API key set."""
-        if self.name == "ollama":
-            return True  # Ollama needs no API key
+        if self.name in ("ollama", "atomic-chat"):
+            return True  # Local providers need no API key
        return bool(self.api_key)

    @property
@@ -93,6 +93,7 @@ def build_default_providers() -> list[Provider]:
    big = os.getenv("BIG_MODEL", "gpt-4.1")
    small = os.getenv("SMALL_MODEL", "gpt-4.1-mini")
    ollama_url = os.getenv("OLLAMA_BASE_URL", "http://localhost:11434")
+    atomic_chat_url = os.getenv("ATOMIC_CHAT_BASE_URL", "http://127.0.0.1:1337")

    return [
        Provider(
@@ -119,6 +120,14 @@ def build_default_providers() -> list[Provider]:
            big_model=big if "gemini" not in big and "gpt" not in big else "llama3:8b",
            small_model=small if "gemini" not in small and "gpt" not in small else "llama3:8b",
        ),
+        Provider(
+            name="atomic-chat",
+            ping_url=f"{atomic_chat_url}/v1/models",
+            api_key_env="",
+            cost_per_1k_tokens=0.0,   # free — local (Apple Silicon)
+            big_model=big if "gemini" not in big and "gpt" not in big else "llama3:8b",
+            small_model=small if "gemini" not in small and "gpt" not in small else "llama3:8b",
+        ),
    ]


--- a/src/commands.ts
+++ b/src/commands.ts
@@ -19,6 +19,7 @@ import cost from './commands/cost/index.js'
 import diff from './commands/diff/index.js'
 import ctx_viz from './commands/ctx_viz/index.js'
 import doctor from './commands/doctor/index.js'
+import onboardGithub from './commands/onboard-github/index.js'
 import memory from './commands/memory/index.js'
 import help from './commands/help/index.js'
 import ide from './commands/ide/index.js'
@@ -289,6 +290,7 @@ const COMMANDS = memoize((): Command[] => [
  memory,
  mobile,
  model,
+  onboardGithub,
  outputStyle,
  remoteEnv,
  plugin,
--- a/src/commands/onboard-github/index.ts
+++ b/src/commands/onboard-github/index.ts
@@ -0,0 +1,11 @@
+import type { Command } from '../../commands.js'
+
+const onboardGithub: Command = {
+  name: 'onboard-github',
+  description:
+    'Interactive setup for GitHub Models: device login or PAT, saved to secure storage',
+  type: 'local-jsx',
+  load: () => import('./onboard-github.js'),
+}
+
+export default onboardGithub
--- a/src/commands/onboard-github/onboard-github.tsx
+++ b/src/commands/onboard-github/onboard-github.tsx
@@ -0,0 +1,237 @@
+import * as React from 'react'
+import { useCallback, useState } from 'react'
+import { Select } from '../../components/CustomSelect/select.js'
+import { Spinner } from '../../components/Spinner.js'
+import TextInput from '../../components/TextInput.js'
+import { Box, Text } from '../../ink.js'
+import {
+  openVerificationUri,
+  pollAccessToken,
+  requestDeviceCode,
+} from '../../services/github/deviceFlow.js'
+import type { LocalJSXCommandCall } from '../../types/command.js'
+import {
+  hydrateGithubModelsTokenFromSecureStorage,
+  saveGithubModelsToken,
+} from '../../utils/githubModelsCredentials.js'
+import { updateSettingsForSource } from '../../utils/settings/settings.js'
+
+const DEFAULT_MODEL = 'github:copilot'
+
+type Step =
+  | 'menu'
+  | 'device-busy'
+  | 'pat'
+  | 'error'
+
+function mergeUserSettingsEnv(model: string): { ok: boolean; detail?: string } {
+  const { error } = updateSettingsForSource('userSettings', {
+    env: {
+      CLAUDE_CODE_USE_GITHUB: '1',
+      OPENAI_MODEL: model,
+      CLAUDE_CODE_USE_OPENAI: undefined as any,
+      CLAUDE_CODE_USE_GEMINI: undefined as any,
+      CLAUDE_CODE_USE_BEDROCK: undefined as any,
+      CLAUDE_CODE_USE_VERTEX: undefined as any,
+      CLAUDE_CODE_USE_FOUNDRY: undefined as any,
+    },
+  })
+  if (error) {
+    return { ok: false, detail: error.message }
+  }
+  return { ok: true }
+}
+
+function OnboardGithub(props: {
+  onDone: Parameters<LocalJSXCommandCall>[0]
+  onChangeAPIKey: () => void
+}): React.ReactNode {
+  const { onDone, onChangeAPIKey } = props
+  const [step, setStep] = useState<Step>('menu')
+  const [errorMsg, setErrorMsg] = useState<string | null>(null)
+  const [deviceHint, setDeviceHint] = useState<{
+    user_code: string
+    verification_uri: string
+  } | null>(null)
+  const [patDraft, setPatDraft] = useState('')
+  const [cursorOffset, setCursorOffset] = useState(0)
+
+  const finalize = useCallback(
+    async (token: string, model: string = DEFAULT_MODEL) => {
+      const saved = saveGithubModelsToken(token)
+      if (!saved.success) {
+        setErrorMsg(saved.warning ?? 'Could not save token to secure storage.')
+        setStep('error')
+        return
+      }
+      const merged = mergeUserSettingsEnv(model.trim() || DEFAULT_MODEL)
+      if (!merged.ok) {
+        setErrorMsg(
+          `Token saved, but settings were not updated: ${merged.detail ?? 'unknown error'}. ` +
+            `Add env CLAUDE_CODE_USE_GITHUB=1 and OPENAI_MODEL to ~/.claude/settings.json manually.`,
+        )
+        setStep('error')
+        return
+      }
+      process.env.CLAUDE_CODE_USE_GITHUB = '1'
+      process.env.OPENAI_MODEL = model.trim() || DEFAULT_MODEL
+      hydrateGithubModelsTokenFromSecureStorage()
+      onChangeAPIKey()
+      onDone(
+        'GitHub Models onboard complete. Token stored in secure storage; user settings updated. Restart if the model does not switch.',
+        { display: 'user' },
+      )
+    },
+    [onChangeAPIKey, onDone],
+  )
+
+  const runDeviceFlow = useCallback(async () => {
+    setStep('device-busy')
+    setErrorMsg(null)
+    setDeviceHint(null)
+    try {
+      const device = await requestDeviceCode()
+      setDeviceHint({
+        user_code: device.user_code,
+        verification_uri: device.verification_uri,
+      })
+      await openVerificationUri(device.verification_uri)
+      const token = await pollAccessToken(device.device_code, {
+        initialInterval: device.interval,
+        timeoutSeconds: device.expires_in,
+      })
+      await finalize(token, DEFAULT_MODEL)
+    } catch (e) {
+      setErrorMsg(e instanceof Error ? e.message : String(e))
+      setStep('error')
+    }
+  }, [finalize])
+
+  if (step === 'error' && errorMsg) {
+    const options = [
+      {
+        label: 'Back to menu',
+        value: 'back' as const,
+      },
+      {
+        label: 'Exit',
+        value: 'exit' as const,
+      },
+    ]
+    return (
+      <Box flexDirection="column" gap={1}>
+        <Text color="red">{errorMsg}</Text>
+        <Select
+          options={options}
+          onChange={(v: string) => {
+            if (v === 'back') {
+              setStep('menu')
+              setErrorMsg(null)
+            } else {
+              onDone('GitHub onboard cancelled', { display: 'system' })
+            }
+          }}
+        />
+      </Box>
+    )
+  }
+
+  if (step === 'device-busy') {
+    return (
+      <Box flexDirection="column" gap={1}>
+        <Text>GitHub device login</Text>
+        {deviceHint ? (
+          <>
+            <Text>
+              Enter code <Text bold>{deviceHint.user_code}</Text> at{' '}
+              {deviceHint.verification_uri}
+            </Text>
+            <Text dimColor>
+              A browser window may have opened. Waiting for authorization…
+            </Text>
+          </>
+        ) : (
+          <Text dimColor>Requesting device code from GitHub…</Text>
+        )}
+        <Spinner />
+      </Box>
+    )
+  }
+
+  if (step === 'pat') {
+    return (
+      <Box flexDirection="column" gap={1}>
+        <Text>Paste a GitHub personal access token with access to GitHub Models.</Text>
+        <Text dimColor>Input is masked. Enter to submit; Esc to go back.</Text>
+        <TextInput
+          value={patDraft}
+          mask="*"
+          onChange={setPatDraft}
+          onSubmit={async (value: string) => {
+            const t = value.trim()
+            if (!t) {
+              return
+            }
+            await finalize(t, DEFAULT_MODEL)
+          }}
+          onExit={() => {
+            setStep('menu')
+            setPatDraft('')
+          }}
+          columns={80}
+          cursorOffset={cursorOffset}
+          onChangeCursorOffset={setCursorOffset}
+        />
+      </Box>
+    )
+  }
+
+  const menuOptions = [
+    {
+      label: 'Sign in with browser (device code)',
+      value: 'device' as const,
+    },
+    {
+      label: 'Paste personal access token',
+      value: 'pat' as const,
+    },
+    {
+      label: 'Cancel',
+      value: 'cancel' as const,
+    },
+  ]
+
+  return (
+    <Box flexDirection="column" gap={1}>
+      <Text bold>GitHub Models setup</Text>
+      <Text dimColor>
+        Stores your token in the OS credential store (macOS Keychain when available)
+        and enables CLAUDE_CODE_USE_GITHUB in your user settings — no export
+        GITHUB_TOKEN needed for future runs.
+      </Text>
+      <Select
+        options={menuOptions}
+        onChange={(v: string) => {
+          if (v === 'cancel') {
+            onDone('GitHub onboard cancelled', { display: 'system' })
+            return
+          }
+          if (v === 'pat') {
+            setStep('pat')
+            return
+          }
+          void runDeviceFlow()
+        }}
+      />
+    </Box>
+  )
+}
+
+export const call: LocalJSXCommandCall = async (onDone, context) => {
+  return (
+    <OnboardGithub
+      onDone={onDone}
+      onChangeAPIKey={context.onChangeAPIKey}
+    />
+  )
+}
--- a/src/components/CostThresholdDialog.tsx
+++ b/src/components/CostThresholdDialog.tsx
@@ -1,50 +1,53 @@
-import { c as _c } from "react-compiler-runtime";
-import React from 'react';
-import { Box, Link, Text } from '../ink.js';
-import { Select } from './CustomSelect/index.js';
-import { Dialog } from './design-system/Dialog.js';
+import React from 'react'
+import { Box, Link, Text } from '../ink.js'
+import { Select } from './CustomSelect/index.js'
+import { Dialog } from './design-system/Dialog.js'
+import { getAPIProvider } from '../utils/model/providers.js'
+
 type Props = {
-  onDone: () => void;
-};
-export function CostThresholdDialog(t0) {
-  const $ = _c(7);
-  const {
-    onDone
-  } = t0;
-  let t1;
-  if ($[0] === Symbol.for("react.memo_cache_sentinel")) {
-    t1 = <Box flexDirection="column"><Text>Learn more about how to monitor your spending:</Text><Link url="https://code.claude.com/docs/en/costs" /></Box>;
-    $[0] = t1;
-  } else {
-    t1 = $[0];
+  onDone: () => void
 }
-  let t2;
-  if ($[1] === Symbol.for("react.memo_cache_sentinel")) {
-    t2 = [{
-      value: "ok",
-      label: "Got it, thanks!"
-    }];
-    $[1] = t2;
-  } else {
-    t2 = $[1];
+
+function getProviderLabel(): string {
+  const provider = getAPIProvider()
+  switch (provider) {
+    case 'firstParty':
+      return 'Anthropic API'
+    case 'bedrock':
+      return 'AWS Bedrock'
+    case 'vertex':
+      return 'Google Vertex'
+    case 'foundry':
+      return 'Azure Foundry'
+    case 'openai':
+      return 'OpenAI-compatible API'
+    case 'gemini':
+      return 'Gemini API'
+    default:
+      return 'API'
  }
-  let t3;
-  if ($[2] !== onDone) {
-    t3 = <Select options={t2} onChange={onDone} />;
-    $[2] = onDone;
-    $[3] = t3;
-  } else {
-    t3 = $[3];
 }
-  let t4;
-  if ($[4] !== onDone || $[5] !== t3) {
-    t4 = <Dialog title="You've spent $5 on the Anthropic API this session." onCancel={onDone}>{t1}{t3}</Dialog>;
-    $[4] = onDone;
-    $[5] = t3;
-    $[6] = t4;
-  } else {
-    t4 = $[6];
+
+export function CostThresholdDialog({ onDone }: Props): React.ReactNode {
+  const providerLabel = getProviderLabel()
+  return (
+    <Dialog
+      title={`You've spent $5 on the ${providerLabel} this session.`}
+      onCancel={onDone}
+    >
+      <Box flexDirection="column">
+        <Text>Learn more about how to monitor your spending:</Text>
+        <Link url="https://code.claude.com/docs/en/costs" />
+      </Box>
+      <Select
+        options={[
+          {
+            value: 'ok',
+            label: 'Got it, thanks!',
+          },
+        ]}
+        onChange={onDone}
+      />
+    </Dialog>
+  )
 }
-  return t4;
-}
-//# sourceMappingURL=data:application/json;charset=utf-8;base64,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
--- a/src/components/StartupScreen.ts
+++ b/src/components/StartupScreen.ts
@@ -80,6 +80,7 @@ const LOGO_CLAUDE = [

 function detectProvider(): { name: string; model: string; baseUrl: string; isLocal: boolean } {
  const useGemini = process.env.CLAUDE_CODE_USE_GEMINI === '1' || process.env.CLAUDE_CODE_USE_GEMINI === 'true'
+  const useGithub = process.env.CLAUDE_CODE_USE_GITHUB === '1' || process.env.CLAUDE_CODE_USE_GITHUB === 'true'
  const useOpenAI = process.env.CLAUDE_CODE_USE_OPENAI === '1' || process.env.CLAUDE_CODE_USE_OPENAI === 'true'

  if (useGemini) {
@@ -88,6 +89,13 @@ function detectProvider(): { name: string; model: string; baseUrl: string; isLoc
    return { name: 'Google Gemini', model, baseUrl, isLocal: false }
  }

+  if (useGithub) {
+    const model = process.env.OPENAI_MODEL || 'github:copilot'
+    const baseUrl =
+      process.env.OPENAI_BASE_URL || 'https://models.github.ai/inference'
+    return { name: 'GitHub Models', model, baseUrl, isLocal: false }
+  }
+
  if (useOpenAI) {
    const model = process.env.OPENAI_MODEL || 'gpt-4o'
    const baseUrl = process.env.OPENAI_BASE_URL || 'https://api.openai.com/v1'
--- a/src/entrypoints/agentSdkTypes.ts
+++ b/src/entrypoints/agentSdkTypes.ts
@@ -441,3 +441,8 @@ export async function connectRemoteControl(
 ): Promise<RemoteControlHandle | null> {
  throw new Error('not implemented')
 }
+
+// add exit reason types for removing the error within gracefulShutdown file 
+export type ExitReason = {
+  
+}
--- a/src/entrypoints/cli.tsx
+++ b/src/entrypoints/cli.tsx
@@ -53,6 +53,9 @@ function isLocalProviderUrl(baseUrl: string | undefined): boolean {
 function getProviderValidationError(
  env: NodeJS.ProcessEnv = process.env,
 ): string | null {
+  const useOpenAI = isEnvTruthy(env.CLAUDE_CODE_USE_OPENAI)
+  const useGithub = isEnvTruthy(env.CLAUDE_CODE_USE_GITHUB)
+
  if (isEnvTruthy(env.CLAUDE_CODE_USE_GEMINI)) {
    if (!(env.GEMINI_API_KEY ?? env.GOOGLE_API_KEY)) {
      return 'GEMINI_API_KEY is required when CLAUDE_CODE_USE_GEMINI=1.'
@@ -60,7 +63,15 @@ function getProviderValidationError(
    return null
  }

-  if (!isEnvTruthy(env.CLAUDE_CODE_USE_OPENAI)) {
+  if (useGithub && !useOpenAI) {
+    const token = (env.GITHUB_TOKEN?.trim() || env.GH_TOKEN?.trim()) ?? ''
+    if (!token) {
+      return 'GITHUB_TOKEN or GH_TOKEN is required when CLAUDE_CODE_USE_GITHUB=1.'
+    }
+    return null
+  }
+
+  if (!useOpenAI) {
    return null
  }

@@ -91,6 +102,10 @@ function getProviderValidationError(
  }

  if (!env.OPENAI_API_KEY && !isLocalProviderUrl(request.baseUrl)) {
+    const hasGithubToken = !!(env.GITHUB_TOKEN?.trim() || env.GH_TOKEN?.trim())
+    if (useGithub && hasGithubToken) {
+      return null
+    }
    return 'OPENAI_API_KEY is required when CLAUDE_CODE_USE_OPENAI=1 and OPENAI_BASE_URL is not local.'
  }

@@ -121,6 +136,15 @@ async function main(): Promise<void> {
    return;
  }

+  {
+    const { enableConfigs } = await import('../utils/config.js')
+    enableConfigs()
+    const { applySafeConfigEnvironmentVariables } = await import('../utils/managedEnv.js')
+    applySafeConfigEnvironmentVariables()
+    const { hydrateGithubModelsTokenFromSecureStorage } = await import('../utils/githubModelsCredentials.js')
+    hydrateGithubModelsTokenFromSecureStorage()
+  }
+
  const startupEnv = await buildStartupEnvFromProfile({
    processEnv: process.env,
  })
--- a/src/main.tsx
+++ b/src/main.tsx
@@ -2313,7 +2313,11 @@ async function run(): Promise<CommanderCommand> {
        errors
      } = getSettingsWithErrors();
      const nonMcpErrors = errors.filter(e => !e.mcpErrorMetadata);
-      if (nonMcpErrors.length > 0 && !isEnvTruthy(process.env.CLAUDE_CODE_USE_OPENAI)) {
+      if (
+        nonMcpErrors.length > 0 &&
+        !isEnvTruthy(process.env.CLAUDE_CODE_USE_OPENAI) &&
+        !isEnvTruthy(process.env.CLAUDE_CODE_USE_GITHUB)
+      ) {
        await launchInvalidSettingsDialog(root, {
          settingsErrors: nonMcpErrors,
          onExit: () => gracefulShutdownSync(1)
--- a/src/migrations/migrateSonnet1mToSonnet45.ts
+++ b/src/migrations/migrateSonnet1mToSonnet45.ts
@@ -3,6 +3,7 @@ import {
  setMainLoopModelOverride,
 } from '../bootstrap/state.js'
 import { getGlobalConfig, saveGlobalConfig } from '../utils/config.js'
+import { getAPIProvider } from '../utils/model/providers.js'
 import {
  getSettingsForSource,
  updateSettingsForSource,
@@ -23,6 +24,10 @@ import {
 * tracked by a completion flag in global config.
 */
 export function migrateSonnet1mToSonnet45(): void {
+  if (getAPIProvider() !== 'firstParty') {
+    return
+  }
+
  const config = getGlobalConfig()
  if (config.sonnet1m45MigrationComplete) {
    return
--- a/src/screens/REPL.tsx
+++ b/src/screens/REPL.tsx
@@ -137,7 +137,7 @@ import { generateSessionTitle } from '../utils/sessionTitle.js';
 import { BASH_INPUT_TAG, COMMAND_MESSAGE_TAG, COMMAND_NAME_TAG, LOCAL_COMMAND_STDOUT_TAG } from '../constants/xml.js';
 import { escapeXml } from '../utils/xml.js';
 import type { ThinkingConfig } from '../utils/thinking.js';
-import { gracefulShutdownSync } from '../utils/gracefulShutdown.js';
+import { gracefulShutdownSync, isShuttingDown } from '../utils/gracefulShutdown.js';
 import { handlePromptSubmit, type PromptInputHelpers } from '../utils/handlePromptSubmit.js';
 import { useQueueProcessor } from '../hooks/useQueueProcessor.js';
 import { useMailboxBridge } from '../hooks/useMailboxBridge.js';
@@ -4886,7 +4886,7 @@ export function REPL({

          {mrRender()}

-          {!toolJSX?.shouldHidePromptInput && !focusedInputDialog && !isExiting && !disabled && !cursor && <>
+          {!toolJSX?.shouldHidePromptInput && !focusedInputDialog && !isExiting && !disabled && !cursor && !isShuttingDown() && <>
            {autoRunIssueReason && <AutoRunIssueNotification onRun={handleAutoRunIssue} onCancel={handleCancelAutoRunIssue} reason={getAutoRunIssueReasonText(autoRunIssueReason)} />}
            {postCompactSurvey.state !== 'closed' ? <FeedbackSurvey state={postCompactSurvey.state} lastResponse={postCompactSurvey.lastResponse} handleSelect={postCompactSurvey.handleSelect} inputValue={inputValue} setInputValue={setInputValue} onRequestFeedback={handleSurveyRequestFeedback} /> : memorySurvey.state !== 'closed' ? <FeedbackSurvey state={memorySurvey.state} lastResponse={memorySurvey.lastResponse} handleSelect={memorySurvey.handleSelect} handleTranscriptSelect={memorySurvey.handleTranscriptSelect} inputValue={inputValue} setInputValue={setInputValue} onRequestFeedback={handleSurveyRequestFeedback} message="How well did Claude use its memory? (optional)" /> : <FeedbackSurvey state={feedbackSurvey.state} lastResponse={feedbackSurvey.lastResponse} handleSelect={feedbackSurvey.handleSelect} handleTranscriptSelect={feedbackSurvey.handleTranscriptSelect} inputValue={inputValue} setInputValue={setInputValue} onRequestFeedback={didAutoRunIssueRef.current ? undefined : handleSurveyRequestFeedback} />}
            {/* Frustration-triggered transcript sharing prompt */}
--- a/src/services/api/client.ts
+++ b/src/services/api/client.ts
@@ -154,7 +154,10 @@ export async function getAnthropicClient({
      fetch: resolvedFetch,
    }),
  }
-  if (isEnvTruthy(process.env.CLAUDE_CODE_USE_OPENAI)) {
+  if (
+    isEnvTruthy(process.env.CLAUDE_CODE_USE_OPENAI) ||
+    isEnvTruthy(process.env.CLAUDE_CODE_USE_GITHUB)
+  ) {
    const { createOpenAIShimClient } = await import('./openaiShim.js')
    return createOpenAIShimClient({
      defaultHeaders,
--- a/src/services/api/openaiShim.ts
+++ b/src/services/api/openaiShim.ts
@@ -14,8 +14,15 @@
 *   OPENAI_BASE_URL=http://...        — base URL (default: https://api.openai.com/v1)
 *   OPENAI_MODEL=gpt-4o              — default model override
 *   CODEX_API_KEY / ~/.codex/auth.json — Codex auth for codexplan/codexspark
+ *
+ * GitHub Models (models.github.ai), OpenAI-compatible:
+ *   CLAUDE_CODE_USE_GITHUB=1         — enable GitHub inference (no need for USE_OPENAI)
+ *   GITHUB_TOKEN or GH_TOKEN         — PAT with models access (mapped to Bearer auth)
+ *   OPENAI_MODEL                     — optional; use github:copilot or openai/gpt-4.1 style IDs
 */

+import { isEnvTruthy } from '../../utils/envUtils.js'
+import { hydrateGithubModelsTokenFromSecureStorage } from '../../utils/githubModelsCredentials.js'
 import {
  codexStreamToAnthropic,
  collectCodexCompletedResponse,
@@ -31,6 +38,25 @@ import {
 } from './providerConfig.js'
 import { redactSecretValueForDisplay } from '../../utils/providerProfile.js'

+const GITHUB_MODELS_DEFAULT_BASE = 'https://models.github.ai/inference'
+const GITHUB_API_VERSION = '2022-11-28'
+const GITHUB_429_MAX_RETRIES = 3
+const GITHUB_429_BASE_DELAY_SEC = 1
+const GITHUB_429_MAX_DELAY_SEC = 32
+
+function isGithubModelsMode(): boolean {
+  return isEnvTruthy(process.env.CLAUDE_CODE_USE_GITHUB)
+}
+
+function formatRetryAfterHint(response: Response): string {
+  const ra = response.headers.get('retry-after')
+  return ra ? ` (Retry-After: ${ra})` : ''
+}
+
+function sleepMs(ms: number): Promise<void> {
+  return new Promise(resolve => setTimeout(resolve, ms))
+}
+
 // ---------------------------------------------------------------------------
 // Types — minimal subset of Anthropic SDK types we need to produce
 // ---------------------------------------------------------------------------
@@ -236,28 +262,66 @@ function normalizeSchemaForOpenAI(
  schema: Record<string, unknown>,
  strict = true,
 ): Record<string, unknown> {
-  if (schema.type !== 'object' || !schema.properties) return schema
-  const properties = schema.properties as Record<string, unknown>
-  const existingRequired = Array.isArray(schema.required) ? schema.required as string[] : []
-  // OpenAI strict mode requires every property to be listed in required[].
-  // Gemini rejects schemas where required[] contains keys absent from properties,
-  // so only promote keys that actually exist in properties.
-  if (strict) {
-    const allKeys = Object.keys(properties)
-    const required = Array.from(new Set([...existingRequired, ...allKeys]))
-    return { ...schema, required }
+  if (!schema || typeof schema !== 'object' || Array.isArray(schema)) {
+    return (schema ?? {}) as Record<string, unknown>
  }
+
+  const record = { ...schema }
+
+  if (record.type === 'object' && record.properties) {
+    const properties = record.properties as Record<string, Record<string, unknown>>
+    const existingRequired = Array.isArray(record.required) ? record.required as string[] : []
+
+    // Recurse into each property
+    const normalizedProps: Record<string, unknown> = {}
+    for (const [key, value] of Object.entries(properties)) {
+      normalizedProps[key] = normalizeSchemaForOpenAI(
+        value as Record<string, unknown>,
+        strict,
+      )
+    }
+    record.properties = normalizedProps
+
+    if (strict) {
+      // OpenAI strict mode requires every property to be listed in required[]
+      const allKeys = Object.keys(normalizedProps)
+      record.required = Array.from(new Set([...existingRequired, ...allKeys]))
+      // OpenAI strict mode requires additionalProperties: false on all object
+      // schemas — override unconditionally to ensure nested objects comply.
+      record.additionalProperties = false
+    } else {
      // For Gemini: keep only existing required keys that are present in properties
-  const required = existingRequired.filter(k => k in properties)
-  return { ...schema, required }
+      record.required = existingRequired.filter(k => k in normalizedProps)
+    }
+  }
+
+  // Recurse into array items
+  if ('items' in record) {
+    if (Array.isArray(record.items)) {
+      record.items = (record.items as unknown[]).map(
+        item => normalizeSchemaForOpenAI(item as Record<string, unknown>, strict),
+      )
+    } else {
+      record.items = normalizeSchemaForOpenAI(record.items as Record<string, unknown>, strict)
+    }
+  }
+
+  // Recurse into combinators
+  for (const key of ['anyOf', 'oneOf', 'allOf'] as const) {
+    if (key in record && Array.isArray(record[key])) {
+      record[key] = (record[key] as unknown[]).map(
+        item => normalizeSchemaForOpenAI(item as Record<string, unknown>, strict),
+      )
+    }
+  }
+
+  return record
 }

 function convertTools(
  tools: Array<{ name: string; description?: string; input_schema?: Record<string, unknown> }>,
 ): OpenAITool[] {
-  const isGemini =
-    process.env.CLAUDE_CODE_USE_GEMINI === '1' ||
-    process.env.CLAUDE_CODE_USE_GEMINI === 'true'
+  const isGemini = isEnvTruthy(process.env.CLAUDE_CODE_USE_GEMINI)

  return tools
    .filter(t => t.name !== 'ToolSearchTool') // Not relevant for OpenAI
@@ -343,7 +407,7 @@ async function* openaiStreamToAnthropic(
 ): AsyncGenerator<AnthropicStreamEvent> {
  const messageId = makeMessageId()
  let contentBlockIndex = 0
-  const activeToolCalls = new Map<number, { id: string; name: string; index: number }>()
+  const activeToolCalls = new Map<number, { id: string; name: string; index: number; jsonBuffer: string }>()
  let hasEmittedContentStart = false
  let lastStopReason: 'tool_use' | 'max_tokens' | 'end_turn' | null = null
  let hasEmittedFinalUsage = false
@@ -375,6 +439,7 @@ async function* openaiStreamToAnthropic(
  const decoder = new TextDecoder()
  let buffer = ''

+  try {
    while (true) {
      const { done, value } = await reader.read()
      if (done) break
@@ -437,6 +502,7 @@ async function* openaiStreamToAnthropic(
                id: tc.id,
                name: tc.function.name,
                index: toolBlockIndex,
+                jsonBuffer: tc.function.arguments ?? '',
              })

              yield {
@@ -467,6 +533,9 @@ async function* openaiStreamToAnthropic(
              // Continuation of existing tool call
              const active = activeToolCalls.get(tc.index)
              if (active) {
+                if (tc.function.arguments) {
+                  active.jsonBuffer += tc.function.arguments
+                }
                yield {
                  type: 'content_block_delta',
                  index: active.index,
@@ -494,6 +563,36 @@ async function* openaiStreamToAnthropic(
          }
          // Close active tool calls
          for (const [, tc] of activeToolCalls) {
+            let suffixToAdd = ''
+            if (tc.jsonBuffer) {
+              try {
+                JSON.parse(tc.jsonBuffer)
+              } catch {
+                const str = tc.jsonBuffer.trimEnd()
+                const combinations = [
+                  '}', '"}', ']}', '"]}', '}}', '"}}', ']}}', '"]}}', '"]}]}', '}]}'
+                ]
+                for (const combo of combinations) {
+                  try {
+                    JSON.parse(str + combo)
+                    suffixToAdd = combo
+                    break
+                  } catch {}
+                }
+              }
+            }
+
+            if (suffixToAdd) {
+              yield {
+                type: 'content_block_delta',
+                index: tc.index,
+                delta: {
+                  type: 'input_json_delta',
+                  partial_json: suffixToAdd,
+                },
+              }
+            }
+
            yield { type: 'content_block_stop', index: tc.index }
          }

@@ -530,6 +629,9 @@ async function* openaiStreamToAnthropic(
      }
    }
    }
+  } finally {
+    reader.releaseLock()
+  }

  yield { type: 'message_stop' }
 }
@@ -680,6 +782,12 @@ class OpenAIShimMessages {
      body.stream_options = { include_usage: true }
    }

+    const isGithub = isGithubModelsMode()
+    if (isGithub && body.max_completion_tokens !== undefined) {
+      body.max_tokens = body.max_completion_tokens
+      delete body.max_completion_tokens
+    }
+
    if (params.temperature !== undefined) body.temperature = params.temperature
    if (params.top_p !== undefined) body.top_p = params.top_p

@@ -729,6 +837,11 @@ class OpenAIShimMessages {
      }
    }

+    if (isGithub) {
+      headers.Accept = 'application/vnd.github.v3+json'
+      headers['X-GitHub-Api-Version'] = GITHUB_API_VERSION
+    }
+
    // Build the chat completions URL
    // Azure Cognitive Services / Azure OpenAI require a deployment-specific path
    // and an api-version query parameter.
@@ -751,20 +864,43 @@ class OpenAIShimMessages {
      chatCompletionsUrl = `${request.baseUrl}/chat/completions`
    }

-    const response = await fetch(chatCompletionsUrl, {
-      method: 'POST',
+    const fetchInit = {
+      method: 'POST' as const,
      headers,
      body: JSON.stringify(body),
      signal: options?.signal,
-    })
-
-    if (!response.ok) {
-      const errorBody = await response.text().catch(() => 'unknown error')
-      throw new Error(`OpenAI API error ${response.status}: ${errorBody}`)
    }

+    const maxAttempts = isGithub ? GITHUB_429_MAX_RETRIES : 1
+    let response: Response | undefined
+    for (let attempt = 0; attempt < maxAttempts; attempt++) {
+      response = await fetch(chatCompletionsUrl, fetchInit)
+      if (response.ok) {
        return response
      }
+      if (
+        isGithub &&
+        response.status === 429 &&
+        attempt < maxAttempts - 1
+      ) {
+        await response.text().catch(() => {})
+        const delaySec = Math.min(
+          GITHUB_429_BASE_DELAY_SEC * 2 ** attempt,
+          GITHUB_429_MAX_DELAY_SEC,
+        )
+        await sleepMs(delaySec * 1000)
+        continue
+      }
+      const errorBody = await response.text().catch(() => 'unknown error')
+      const rateHint =
+        isGithub && response.status === 429 ? formatRetryAfterHint(response) : ''
+      throw new Error(
+        `OpenAI API error ${response.status}: ${errorBody}${rateHint}`,
+      )
+    }
+
+    throw new Error('OpenAI shim: request loop exited unexpectedly')
+  }

  private _convertNonStreamingResponse(
    data: {
@@ -773,7 +909,10 @@ class OpenAIShimMessages {
      choices?: Array<{
        message?: {
          role?: string
-          content?: string | null
+          content?:
+            | string
+            | null
+            | Array<{ type?: string; text?: string }>
          tool_calls?: Array<{
            id: string
            function: { name: string; arguments: string }
@@ -792,8 +931,25 @@ class OpenAIShimMessages {
    const choice = data.choices?.[0]
    const content: Array<Record<string, unknown>> = []

-    if (choice?.message?.content) {
-      content.push({ type: 'text', text: choice.message.content })
+    const rawContent = choice?.message?.content
+    if (typeof rawContent === 'string' && rawContent) {
+      content.push({ type: 'text', text: rawContent })
+    } else if (Array.isArray(rawContent) && rawContent.length > 0) {
+      const parts: string[] = []
+      for (const part of rawContent) {
+        if (
+          part &&
+          typeof part === 'object' &&
+          part.type === 'text' &&
+          typeof part.text === 'string'
+        ) {
+          parts.push(part.text)
+        }
+      }
+      const joined = parts.join('\n')
+      if (joined) {
+        content.push({ type: 'text', text: joined })
+      }
    }

    if (choice?.message?.tool_calls) {
@@ -852,12 +1008,11 @@ export function createOpenAIShimClient(options: {
  maxRetries?: number
  timeout?: number
 }): unknown {
+  hydrateGithubModelsTokenFromSecureStorage()
+
  // When Gemini provider is active, map Gemini env vars to OpenAI-compatible ones
  // so the existing providerConfig.ts infrastructure picks them up correctly.
-  if (
-    process.env.CLAUDE_CODE_USE_GEMINI === '1' ||
-    process.env.CLAUDE_CODE_USE_GEMINI === 'true'
-  ) {
+  if (isEnvTruthy(process.env.CLAUDE_CODE_USE_GEMINI)) {
    process.env.OPENAI_BASE_URL ??=
      process.env.GEMINI_BASE_URL ??
      'https://generativelanguage.googleapis.com/v1beta/openai'
@@ -866,6 +1021,10 @@ export function createOpenAIShimClient(options: {
    if (process.env.GEMINI_MODEL && !process.env.OPENAI_MODEL) {
      process.env.OPENAI_MODEL = process.env.GEMINI_MODEL
    }
+  } else if (isEnvTruthy(process.env.CLAUDE_CODE_USE_GITHUB)) {
+    process.env.OPENAI_BASE_URL ??= GITHUB_MODELS_DEFAULT_BASE
+    process.env.OPENAI_API_KEY ??=
+      process.env.GITHUB_TOKEN ?? process.env.GH_TOKEN ?? ''
  }

  const beta = new OpenAIShimBeta({
--- a/src/services/api/providerConfig.github.test.ts
+++ b/src/services/api/providerConfig.github.test.ts
@@ -0,0 +1,41 @@
+import { afterEach, expect, test } from 'bun:test'
+
+import {
+  DEFAULT_GITHUB_MODELS_API_MODEL,
+  normalizeGithubModelsApiModel,
+  resolveProviderRequest,
+} from './providerConfig.js'
+
+const originalUseGithub = process.env.CLAUDE_CODE_USE_GITHUB
+
+afterEach(() => {
+  if (originalUseGithub === undefined) {
+    delete process.env.CLAUDE_CODE_USE_GITHUB
+  } else {
+    process.env.CLAUDE_CODE_USE_GITHUB = originalUseGithub
+  }
+})
+
+test.each([
+  ['copilot', DEFAULT_GITHUB_MODELS_API_MODEL],
+  ['github:copilot', DEFAULT_GITHUB_MODELS_API_MODEL],
+  ['', DEFAULT_GITHUB_MODELS_API_MODEL],
+  ['github:gpt-4o', 'gpt-4o'],
+  ['gpt-4o', 'gpt-4o'],
+  ['github:copilot?reasoning=high', DEFAULT_GITHUB_MODELS_API_MODEL],
+] as const)('normalizeGithubModelsApiModel(%s) -> %s', (input, expected) => {
+  expect(normalizeGithubModelsApiModel(input)).toBe(expected)
+})
+
+test('resolveProviderRequest applies GitHub normalization when CLAUDE_CODE_USE_GITHUB=1', () => {
+  process.env.CLAUDE_CODE_USE_GITHUB = '1'
+  const r = resolveProviderRequest({ model: 'github:gpt-4o' })
+  expect(r.resolvedModel).toBe('gpt-4o')
+  expect(r.transport).toBe('chat_completions')
+})
+
+test('resolveProviderRequest leaves model unchanged without GitHub flag', () => {
+  delete process.env.CLAUDE_CODE_USE_GITHUB
+  const r = resolveProviderRequest({ model: 'github:gpt-4o' })
+  expect(r.resolvedModel).toBe('github:gpt-4o')
+})
--- a/src/services/api/providerConfig.ts
+++ b/src/services/api/providerConfig.ts
@@ -2,8 +2,12 @@ import { existsSync, readFileSync } from 'node:fs'
 import { homedir } from 'node:os'
 import { join } from 'node:path'

+import { isEnvTruthy } from '../../utils/envUtils.js'
+
 export const DEFAULT_OPENAI_BASE_URL = 'https://api.openai.com/v1'
 export const DEFAULT_CODEX_BASE_URL = 'https://chatgpt.com/backend-api/codex'
+/** Default GitHub Models API model when user selects copilot / github:copilot */
+export const DEFAULT_GITHUB_MODELS_API_MODEL = 'openai/gpt-4.1'

 const CODEX_ALIAS_MODELS: Record<
  string,
@@ -171,16 +175,31 @@ export function isCodexBaseUrl(baseUrl: string | undefined): boolean {
  }
 }

+/**
+ * Normalize user model string for GitHub Models inference (models.github.ai).
+ * Mirrors runtime devsper `github._normalize_model_id`.
+ */
+export function normalizeGithubModelsApiModel(requestedModel: string): string {
+  const noQuery = requestedModel.split('?', 1)[0] ?? requestedModel
+  const segment =
+    noQuery.includes(':') ? noQuery.split(':', 2)[1]!.trim() : noQuery.trim()
+  if (!segment || segment.toLowerCase() === 'copilot') {
+    return DEFAULT_GITHUB_MODELS_API_MODEL
+  }
+  return segment
+}
+
 export function resolveProviderRequest(options?: {
  model?: string
  baseUrl?: string
  fallbackModel?: string
 }): ResolvedProviderRequest {
+  const isGithubMode = isEnvTruthy(process.env.CLAUDE_CODE_USE_GITHUB)
  const requestedModel =
    options?.model?.trim() ||
    process.env.OPENAI_MODEL?.trim() ||
    options?.fallbackModel?.trim() ||
-    'gpt-4o'
+    (isGithubMode ? 'github:copilot' : 'gpt-4o')
  const descriptor = parseModelDescriptor(requestedModel)
  const rawBaseUrl =
    options?.baseUrl ??
@@ -192,10 +211,16 @@ export function resolveProviderRequest(options?: {
      ? 'codex_responses'
      : 'chat_completions'

+  const resolvedModel =
+    transport === 'chat_completions' &&
+    isEnvTruthy(process.env.CLAUDE_CODE_USE_GITHUB)
+      ? normalizeGithubModelsApiModel(requestedModel)
+      : descriptor.baseModel
+
  return {
    transport,
    requestedModel,
-    resolvedModel: descriptor.baseModel,
+    resolvedModel,
    baseUrl:
      (rawBaseUrl ??
        (transport === 'codex_responses'
--- a/src/services/github/deviceFlow.test.ts
+++ b/src/services/github/deviceFlow.test.ts
@@ -0,0 +1,94 @@
+import { afterEach, describe, expect, mock, test } from 'bun:test'
+
+import {
+  GitHubDeviceFlowError,
+  pollAccessToken,
+  requestDeviceCode,
+} from './deviceFlow.js'
+
+describe('requestDeviceCode', () => {
+  const originalFetch = globalThis.fetch
+
+  afterEach(() => {
+    globalThis.fetch = originalFetch
+  })
+
+  test('parses successful device code response', async () => {
+    globalThis.fetch = mock(() =>
+      Promise.resolve(
+        new Response(
+          JSON.stringify({
+            device_code: 'abc',
+            user_code: 'ABCD-1234',
+            verification_uri: 'https://github.com/login/device',
+            expires_in: 600,
+            interval: 5,
+          }),
+          { status: 200 },
+        ),
+      ),
+    )
+
+    const r = await requestDeviceCode({
+      clientId: 'test-client',
+      fetchImpl: globalThis.fetch,
+    })
+    expect(r.device_code).toBe('abc')
+    expect(r.user_code).toBe('ABCD-1234')
+    expect(r.verification_uri).toBe('https://github.com/login/device')
+    expect(r.expires_in).toBe(600)
+    expect(r.interval).toBe(5)
+  })
+
+  test('throws on HTTP error', async () => {
+    globalThis.fetch = mock(() =>
+      Promise.resolve(new Response('bad', { status: 500 })),
+    )
+    await expect(
+      requestDeviceCode({ clientId: 'x', fetchImpl: globalThis.fetch }),
+    ).rejects.toThrow(GitHubDeviceFlowError)
+  })
+})
+
+describe('pollAccessToken', () => {
+  const originalFetch = globalThis.fetch
+
+  afterEach(() => {
+    globalThis.fetch = originalFetch
+  })
+
+  test('returns token when GitHub responds with access_token immediately', async () => {
+    let calls = 0
+    globalThis.fetch = mock(() => {
+      calls++
+      return Promise.resolve(
+        new Response(JSON.stringify({ access_token: 'tok-xyz' }), {
+          status: 200,
+        }),
+      )
+    })
+
+    const token = await pollAccessToken('dev-code', {
+      clientId: 'cid',
+      fetchImpl: globalThis.fetch,
+    })
+    expect(token).toBe('tok-xyz')
+    expect(calls).toBe(1)
+  })
+
+  test('throws on access_denied', async () => {
+    globalThis.fetch = mock(() =>
+      Promise.resolve(
+        new Response(JSON.stringify({ error: 'access_denied' }), {
+          status: 200,
+        }),
+      ),
+    )
+    await expect(
+      pollAccessToken('dc', {
+        clientId: 'c',
+        fetchImpl: globalThis.fetch,
+      }),
+    ).rejects.toThrow(/denied/)
+  })
+})
--- a/src/services/github/deviceFlow.ts
+++ b/src/services/github/deviceFlow.ts
@@ -0,0 +1,174 @@
+/**
+ * GitHub OAuth device flow for CLI login (https://docs.github.com/en/apps/oauth-apps/building-oauth-apps/authorizing-oauth-apps#device-flow).
+ */
+
+import { execFileNoThrow } from '../../utils/execFileNoThrow.js'
+
+export const DEFAULT_GITHUB_DEVICE_FLOW_CLIENT_ID = 'Ov23liXjWSSui6QIahPl'
+
+export const GITHUB_DEVICE_CODE_URL = 'https://github.com/login/device/code'
+export const GITHUB_DEVICE_ACCESS_TOKEN_URL =
+  'https://github.com/login/oauth/access_token'
+
+/** Match runtime devsper github_oauth DEFAULT_SCOPE */
+export const DEFAULT_GITHUB_DEVICE_SCOPE = 'read:user,models:read'
+
+export class GitHubDeviceFlowError extends Error {
+  constructor(message: string) {
+    super(message)
+    this.name = 'GitHubDeviceFlowError'
+  }
+}
+
+export type DeviceCodeResult = {
+  device_code: string
+  user_code: string
+  verification_uri: string
+  expires_in: number
+  interval: number
+}
+
+export function getGithubDeviceFlowClientId(): string {
+  return (
+    process.env.GITHUB_DEVICE_FLOW_CLIENT_ID?.trim() ||
+    DEFAULT_GITHUB_DEVICE_FLOW_CLIENT_ID
+  )
+}
+
+function sleep(ms: number): Promise<void> {
+  return new Promise(resolve => setTimeout(resolve, ms))
+}
+
+export async function requestDeviceCode(options?: {
+  clientId?: string
+  scope?: string
+  fetchImpl?: typeof fetch
+}): Promise<DeviceCodeResult> {
+  const clientId = options?.clientId ?? getGithubDeviceFlowClientId()
+  if (!clientId) {
+    throw new GitHubDeviceFlowError(
+      'No OAuth client ID: set GITHUB_DEVICE_FLOW_CLIENT_ID or paste a PAT instead.',
+    )
+  }
+  const fetchFn = options?.fetchImpl ?? fetch
+  const res = await fetchFn(GITHUB_DEVICE_CODE_URL, {
+    method: 'POST',
+    headers: { Accept: 'application/json' },
+    body: new URLSearchParams({
+      client_id: clientId,
+      scope: options?.scope ?? DEFAULT_GITHUB_DEVICE_SCOPE,
+    }),
+  })
+  if (!res.ok) {
+    const text = await res.text().catch(() => '')
+    throw new GitHubDeviceFlowError(
+      `Device code request failed: ${res.status} ${text}`,
+    )
+  }
+  const data = (await res.json()) as Record<string, unknown>
+  const device_code = data.device_code
+  const user_code = data.user_code
+  const verification_uri = data.verification_uri
+  if (
+    typeof device_code !== 'string' ||
+    typeof user_code !== 'string' ||
+    typeof verification_uri !== 'string'
+  ) {
+    throw new GitHubDeviceFlowError('Malformed device code response from GitHub')
+  }
+  return {
+    device_code,
+    user_code,
+    verification_uri,
+    expires_in: typeof data.expires_in === 'number' ? data.expires_in : 900,
+    interval: typeof data.interval === 'number' ? data.interval : 5,
+  }
+}
+
+export type PollOptions = {
+  clientId?: string
+  initialInterval?: number
+  timeoutSeconds?: number
+  fetchImpl?: typeof fetch
+}
+
+export async function pollAccessToken(
+  deviceCode: string,
+  options?: PollOptions,
+): Promise<string> {
+  const clientId = options?.clientId ?? getGithubDeviceFlowClientId()
+  if (!clientId) {
+    throw new GitHubDeviceFlowError('client_id required for polling')
+  }
+  let interval = Math.max(1, options?.initialInterval ?? 5)
+  const timeoutSeconds = options?.timeoutSeconds ?? 900
+  const fetchFn = options?.fetchImpl ?? fetch
+  const start = Date.now()
+
+  while ((Date.now() - start) / 1000 < timeoutSeconds) {
+    const res = await fetchFn(GITHUB_DEVICE_ACCESS_TOKEN_URL, {
+      method: 'POST',
+      headers: { Accept: 'application/json' },
+      body: new URLSearchParams({
+        client_id: clientId,
+        device_code: deviceCode,
+        grant_type: 'urn:ietf:params:oauth:grant-type:device_code',
+      }),
+    })
+    if (!res.ok) {
+      const text = await res.text().catch(() => '')
+      throw new GitHubDeviceFlowError(
+        `Token request failed: ${res.status} ${text}`,
+      )
+    }
+    const data = (await res.json()) as Record<string, unknown>
+    const err = data.error as string | undefined
+    if (err == null) {
+      const token = data.access_token
+      if (typeof token === 'string' && token) {
+        return token
+      }
+      throw new GitHubDeviceFlowError('No access_token in response')
+    }
+    if (err === 'authorization_pending') {
+      await sleep(interval * 1000)
+      continue
+    }
+    if (err === 'slow_down') {
+      interval =
+        typeof data.interval === 'number' ? data.interval : interval + 5
+      await sleep(interval * 1000)
+      continue
+    }
+    if (err === 'expired_token') {
+      throw new GitHubDeviceFlowError(
+        'Device code expired. Start the login flow again.',
+      )
+    }
+    if (err === 'access_denied') {
+      throw new GitHubDeviceFlowError('Authorization was denied or cancelled.')
+    }
+    throw new GitHubDeviceFlowError(`GitHub OAuth error: ${err}`)
+  }
+  throw new GitHubDeviceFlowError('Timed out waiting for authorization.')
+}
+
+/**
+ * Best-effort open browser / OS handler for the verification URL.
+ */
+export async function openVerificationUri(uri: string): Promise<void> {
+  try {
+    if (process.platform === 'darwin') {
+      await execFileNoThrow('open', [uri], { useCwd: false, timeout: 5000 })
+    } else if (process.platform === 'win32') {
+      await execFileNoThrow('cmd', ['/c', 'start', '', uri], {
+        useCwd: false,
+        timeout: 5000,
+      })
+    } else {
+      await execFileNoThrow('xdg-open', [uri], { useCwd: false, timeout: 5000 })
+    }
+  } catch {
+    // User can open the URL manually
+  }
+}
--- a/src/utils/auth.ts
+++ b/src/utils/auth.ts
@@ -117,7 +117,8 @@ export function isAnthropicAuthEnabled(): boolean {
    isEnvTruthy(process.env.CLAUDE_CODE_USE_VERTEX) ||
    isEnvTruthy(process.env.CLAUDE_CODE_USE_FOUNDRY) ||
    isEnvTruthy(process.env.CLAUDE_CODE_USE_OPENAI) ||
-    isEnvTruthy(process.env.CLAUDE_CODE_USE_GEMINI)
+    isEnvTruthy(process.env.CLAUDE_CODE_USE_GEMINI) ||
+    isEnvTruthy(process.env.CLAUDE_CODE_USE_GITHUB)

  // Check if user has configured an external API key source
  // This allows externally-provided API keys to work (without requiring proxy configuration)
@@ -1731,14 +1732,15 @@ export function getSubscriptionName(): string {
  }
 }

-/** Check if using third-party services (Bedrock or Vertex or Foundry or OpenAI-compatible or Gemini) */
+/** Check if using third-party services (Bedrock or Vertex or Foundry or OpenAI-compatible or Gemini or GitHub Models) */
 export function isUsing3PServices(): boolean {
  return !!(
    isEnvTruthy(process.env.CLAUDE_CODE_USE_BEDROCK) ||
    isEnvTruthy(process.env.CLAUDE_CODE_USE_VERTEX) ||
    isEnvTruthy(process.env.CLAUDE_CODE_USE_FOUNDRY) ||
    isEnvTruthy(process.env.CLAUDE_CODE_USE_OPENAI) ||
-    isEnvTruthy(process.env.CLAUDE_CODE_USE_GEMINI)
+    isEnvTruthy(process.env.CLAUDE_CODE_USE_GEMINI) ||
+    isEnvTruthy(process.env.CLAUDE_CODE_USE_GITHUB)
  )
 }

--- a/src/utils/autoUpdater.ts
+++ b/src/utils/autoUpdater.ts
@@ -9,6 +9,7 @@ import {
  logEvent,
 } from 'src/services/analytics/index.js'
 import { type ReleaseChannel, saveGlobalConfig } from './config.js'
+import { getAPIProvider } from './model/providers.js'
 import { logForDebugging } from './debug.js'
 import { env } from './env.js'
 import { getClaudeConfigHomeDir } from './envUtils.js'
@@ -72,6 +73,12 @@ export async function assertMinVersion(): Promise<void> {
    return
  }

+  // Skip version check for third-party providers — the min version
+  // kill-switch is Anthropic-specific and should not block 3P users
+  if (getAPIProvider() !== 'firstParty') {
+    return
+  }
+
  try {
    const versionConfig = await getDynamicConfig_BLOCKS_ON_INIT<{
      minVersion: string
--- a/src/utils/context.ts
+++ b/src/utils/context.ts
@@ -77,7 +77,9 @@ export function getContextWindowForModel(
    process.env.CLAUDE_CODE_USE_OPENAI === '1' ||
    process.env.CLAUDE_CODE_USE_OPENAI === 'true' ||
    process.env.CLAUDE_CODE_USE_GEMINI === '1' ||
-    process.env.CLAUDE_CODE_USE_GEMINI === 'true'
+    process.env.CLAUDE_CODE_USE_GEMINI === 'true' ||
+    process.env.CLAUDE_CODE_USE_GITHUB === '1' ||
+    process.env.CLAUDE_CODE_USE_GITHUB === 'true'
  ) {
    const openaiWindow = getOpenAIContextWindow(model)
    if (openaiWindow !== undefined) {
@@ -181,7 +183,9 @@ export function getModelMaxOutputTokens(model: string): {
    process.env.CLAUDE_CODE_USE_OPENAI === '1' ||
    process.env.CLAUDE_CODE_USE_OPENAI === 'true' ||
    process.env.CLAUDE_CODE_USE_GEMINI === '1' ||
-    process.env.CLAUDE_CODE_USE_GEMINI === 'true'
+    process.env.CLAUDE_CODE_USE_GEMINI === 'true' ||
+    process.env.CLAUDE_CODE_USE_GITHUB === '1' ||
+    process.env.CLAUDE_CODE_USE_GITHUB === 'true'
  ) {
    const openaiMax = getOpenAIMaxOutputTokens(model)
    if (openaiMax !== undefined) {
--- a/src/utils/githubModelsCredentials.hydrate.test.ts
+++ b/src/utils/githubModelsCredentials.hydrate.test.ts
@@ -0,0 +1,66 @@
+/**
+ * Hydrate tests live in a separate file with no static import of
+ * githubModelsCredentials so Bun's mock.module can replace secureStorage
+ * before that module is first loaded.
+ */
+import { afterEach, describe, expect, mock, test } from 'bun:test'
+
+describe('hydrateGithubModelsTokenFromSecureStorage', () => {
+  const orig = {
+    CLAUDE_CODE_USE_GITHUB: process.env.CLAUDE_CODE_USE_GITHUB,
+    GITHUB_TOKEN: process.env.GITHUB_TOKEN,
+    GH_TOKEN: process.env.GH_TOKEN,
+    CLAUDE_CODE_SIMPLE: process.env.CLAUDE_CODE_SIMPLE,
+  }
+
+  afterEach(() => {
+    mock.restore()
+    for (const [k, v] of Object.entries(orig)) {
+      if (v === undefined) {
+        delete process.env[k as keyof typeof orig]
+      } else {
+        process.env[k as keyof typeof orig] = v
+      }
+    }
+  })
+
+  test('sets GITHUB_TOKEN from secure storage when USE_GITHUB and env token empty', async () => {
+    process.env.CLAUDE_CODE_USE_GITHUB = '1'
+    delete process.env.GITHUB_TOKEN
+    delete process.env.GH_TOKEN
+    delete process.env.CLAUDE_CODE_SIMPLE
+
+    mock.module('./secureStorage/index.js', () => ({
+      getSecureStorage: () => ({
+        read: () => ({
+          githubModels: { accessToken: 'stored-secret' },
+        }),
+      }),
+    }))
+
+    const { hydrateGithubModelsTokenFromSecureStorage } = await import(
+      './githubModelsCredentials.js'
+    )
+    hydrateGithubModelsTokenFromSecureStorage()
+    expect(process.env.GITHUB_TOKEN).toBe('stored-secret')
+  })
+
+  test('does not override existing GITHUB_TOKEN', async () => {
+    process.env.CLAUDE_CODE_USE_GITHUB = '1'
+    process.env.GITHUB_TOKEN = 'already'
+
+    mock.module('./secureStorage/index.js', () => ({
+      getSecureStorage: () => ({
+        read: () => ({
+          githubModels: { accessToken: 'stored-secret' },
+        }),
+      }),
+    }))
+
+    const { hydrateGithubModelsTokenFromSecureStorage } = await import(
+      './githubModelsCredentials.js'
+    )
+    hydrateGithubModelsTokenFromSecureStorage()
+    expect(process.env.GITHUB_TOKEN).toBe('already')
+  })
+})
--- a/src/utils/githubModelsCredentials.test.ts
+++ b/src/utils/githubModelsCredentials.test.ts
@@ -0,0 +1,47 @@
+import { describe, expect, test } from 'bun:test'
+
+import {
+  clearGithubModelsToken,
+  readGithubModelsToken,
+  saveGithubModelsToken,
+} from './githubModelsCredentials.js'
+
+describe('readGithubModelsToken', () => {
+  test('returns undefined in bare mode', () => {
+    const prev = process.env.CLAUDE_CODE_SIMPLE
+    process.env.CLAUDE_CODE_SIMPLE = '1'
+    expect(readGithubModelsToken()).toBeUndefined()
+    if (prev === undefined) {
+      delete process.env.CLAUDE_CODE_SIMPLE
+    } else {
+      process.env.CLAUDE_CODE_SIMPLE = prev
+    }
+  })
+})
+
+describe('saveGithubModelsToken / clearGithubModelsToken', () => {
+  test('save returns failure in bare mode', () => {
+    const prev = process.env.CLAUDE_CODE_SIMPLE
+    process.env.CLAUDE_CODE_SIMPLE = '1'
+    const r = saveGithubModelsToken('abc')
+    expect(r.success).toBe(false)
+    expect(r.warning).toContain('Bare mode')
+    if (prev === undefined) {
+      delete process.env.CLAUDE_CODE_SIMPLE
+    } else {
+      process.env.CLAUDE_CODE_SIMPLE = prev
+    }
+  })
+
+  test('clear succeeds in bare mode', () => {
+    const prev = process.env.CLAUDE_CODE_SIMPLE
+    process.env.CLAUDE_CODE_SIMPLE = '1'
+    expect(clearGithubModelsToken().success).toBe(true)
+    if (prev === undefined) {
+      delete process.env.CLAUDE_CODE_SIMPLE
+    } else {
+      process.env.CLAUDE_CODE_SIMPLE = prev
+    }
+  })
+})
+
--- a/src/utils/githubModelsCredentials.ts
+++ b/src/utils/githubModelsCredentials.ts
@@ -0,0 +1,73 @@
+import { isBareMode, isEnvTruthy } from './envUtils.js'
+import { getSecureStorage } from './secureStorage/index.js'
+
+/** JSON key in the shared OpenClaude secure storage blob. */
+export const GITHUB_MODELS_STORAGE_KEY = 'githubModels' as const
+
+export type GithubModelsCredentialBlob = {
+  accessToken: string
+}
+
+export function readGithubModelsToken(): string | undefined {
+  if (isBareMode()) return undefined
+  try {
+    const data = getSecureStorage().read() as
+      | ({ githubModels?: GithubModelsCredentialBlob } & Record<string, unknown>)
+      | null
+    const t = data?.githubModels?.accessToken?.trim()
+    return t || undefined
+  } catch {
+    return undefined
+  }
+}
+
+/**
+ * If GitHub Models mode is on and no token is in the environment, copy the
+ * stored token into process.env so the OpenAI shim and validation see it.
+ */
+export function hydrateGithubModelsTokenFromSecureStorage(): void {
+  if (!isEnvTruthy(process.env.CLAUDE_CODE_USE_GITHUB)) {
+    return
+  }
+  if (process.env.GITHUB_TOKEN?.trim() || process.env.GH_TOKEN?.trim()) {
+    return
+  }
+  if (isBareMode()) {
+    return
+  }
+  const t = readGithubModelsToken()
+  if (t) {
+    process.env.GITHUB_TOKEN = t
+  }
+}
+
+export function saveGithubModelsToken(token: string): {
+  success: boolean
+  warning?: string
+} {
+  if (isBareMode()) {
+    return { success: false, warning: 'Bare mode: secure storage is disabled.' }
+  }
+  const trimmed = token.trim()
+  if (!trimmed) {
+    return { success: false, warning: 'Token is empty.' }
+  }
+  const secureStorage = getSecureStorage()
+  const prev = secureStorage.read() || {}
+  const merged = {
+    ...(prev as Record<string, unknown>),
+    [GITHUB_MODELS_STORAGE_KEY]: { accessToken: trimmed },
+  }
+  return secureStorage.update(merged as typeof prev)
+}
+
+export function clearGithubModelsToken(): { success: boolean; warning?: string } {
+  if (isBareMode()) {
+    return { success: true }
+  }
+  const secureStorage = getSecureStorage()
+  const prev = secureStorage.read() || {}
+  const next = { ...(prev as Record<string, unknown>) }
+  delete next[GITHUB_MODELS_STORAGE_KEY]
+  return secureStorage.update(next as typeof prev)
+}
--- a/src/utils/gracefulShutdown.ts
+++ b/src/utils/gracefulShutdown.ts
@@ -56,7 +56,7 @@ import { profileReport } from './startupProfiler.js'
 * 3. Failing to disable leaves the terminal in a broken state
 */
 /* eslint-disable custom-rules/no-sync-fs -- must be sync to flush before process.exit */
-function cleanupTerminalModes(): void {
+function cleanupTerminalModes(skipUnmount: boolean = false): void {
  if (!process.stdout.isTTY) {
    return
  }
@@ -84,7 +84,7 @@ function cleanupTerminalModes(): void {
    // Calling unmount() now does the final render on the alt buffer,
    // unsubscribes from signal-exit, and writes 1049l exactly once.
    const inst = instances.get(process.stdout)
-    if (inst?.isAltScreenActive) {
+    if (!skipUnmount && inst?.isAltScreenActive) {
      try {
        inst.unmount()
      } catch {
@@ -92,6 +92,11 @@ function cleanupTerminalModes(): void {
        // so printResumeHint still hits the main buffer.
        writeSync(1, EXIT_ALT_SCREEN)
      }
+    } else if (skipUnmount && inst?.isAltScreenActive) {
+      // We already unmounted asynchronously in gracefulShutdown, but if we 
+      // fallback to manual alt-screen exit here just in case Ink didn't write it or is dead.
+      // Actually, AlternateScreen unmount writes EXIT_ALT_SCREEN, so if we awaited unmount,
+      // we shouldn't emit it again. So we just do nothing here.
    }
    // Catches events that arrived during the unmount tree-walk.
    // detachForShutdown() below also drains.
@@ -411,12 +416,17 @@ export async function gracefulShutdown(
  )
  const sessionEndTimeoutMs = getSessionEndHookTimeoutMs()

+  // Await one tick so React can flush pending updates from commands (e.g. hiding
+  // the autocomplete menu on /exit) before we detach Ink. This lets log-update
+  // erase floating UI elements from the terminal so they don't linger after exit.
+  await new Promise(r => setTimeout(r, 20))
+
  // Failsafe: guarantee process exits even if cleanup hangs (e.g., MCP connections).
  // Runs cleanupTerminalModes first so a hung cleanup doesn't leave the terminal dirty.
  // Budget = max(5s, hook budget + 3.5s headroom for cleanup + analytics flush).
  failsafeTimer = setTimeout(
    code => {
-      cleanupTerminalModes()
+      cleanupTerminalModes(true)
      printResumeHint()
      forceExit(code)
    },
@@ -433,7 +443,7 @@ export async function gracefulShutdown(
  // cleanup (e.g., SIGKILL during macOS reboot). Without this, the resume
  // hint would only appear after cleanup functions, hooks, and analytics
  // flush — which can take several seconds.
-  cleanupTerminalModes()
+  cleanupTerminalModes(true)
  printResumeHint()

  // Flush session data first — this is the most critical cleanup. If the
--- a/src/utils/managedEnvConstants.ts
+++ b/src/utils/managedEnvConstants.ts
@@ -18,6 +18,7 @@ const PROVIDER_MANAGED_ENV_VARS = new Set([
  'CLAUDE_CODE_USE_BEDROCK',
  'CLAUDE_CODE_USE_VERTEX',
  'CLAUDE_CODE_USE_FOUNDRY',
+  'CLAUDE_CODE_USE_GITHUB',
  // Endpoint config (base URLs, project/resource identifiers)
  'ANTHROPIC_BASE_URL',
  'ANTHROPIC_BEDROCK_BASE_URL',
@@ -147,6 +148,7 @@ export const SAFE_ENV_VARS = new Set([
  'CLAUDE_CODE_SUBAGENT_MODEL',
  'CLAUDE_CODE_USE_BEDROCK',
  'CLAUDE_CODE_USE_FOUNDRY',
+  'CLAUDE_CODE_USE_GITHUB',
  'CLAUDE_CODE_USE_VERTEX',
  'DISABLE_AUTOUPDATER',
  'DISABLE_BUG_COMMAND',
--- a/src/utils/model/providers.test.ts
+++ b/src/utils/model/providers.test.ts
@@ -7,6 +7,7 @@ import {

 const originalEnv = {
  CLAUDE_CODE_USE_GEMINI: process.env.CLAUDE_CODE_USE_GEMINI,
+  CLAUDE_CODE_USE_GITHUB: process.env.CLAUDE_CODE_USE_GITHUB,
  CLAUDE_CODE_USE_OPENAI: process.env.CLAUDE_CODE_USE_OPENAI,
  CLAUDE_CODE_USE_BEDROCK: process.env.CLAUDE_CODE_USE_BEDROCK,
  CLAUDE_CODE_USE_VERTEX: process.env.CLAUDE_CODE_USE_VERTEX,
@@ -15,6 +16,7 @@ const originalEnv = {

 afterEach(() => {
  process.env.CLAUDE_CODE_USE_GEMINI = originalEnv.CLAUDE_CODE_USE_GEMINI
+  process.env.CLAUDE_CODE_USE_GITHUB = originalEnv.CLAUDE_CODE_USE_GITHUB
  process.env.CLAUDE_CODE_USE_OPENAI = originalEnv.CLAUDE_CODE_USE_OPENAI
  process.env.CLAUDE_CODE_USE_BEDROCK = originalEnv.CLAUDE_CODE_USE_BEDROCK
  process.env.CLAUDE_CODE_USE_VERTEX = originalEnv.CLAUDE_CODE_USE_VERTEX
@@ -23,6 +25,7 @@ afterEach(() => {

 function clearProviderEnv(): void {
  delete process.env.CLAUDE_CODE_USE_GEMINI
+  delete process.env.CLAUDE_CODE_USE_GITHUB
  delete process.env.CLAUDE_CODE_USE_OPENAI
  delete process.env.CLAUDE_CODE_USE_BEDROCK
  delete process.env.CLAUDE_CODE_USE_VERTEX
@@ -38,6 +41,7 @@ test('first-party provider keeps Anthropic account setup flow enabled', () => {

 test.each([
  ['CLAUDE_CODE_USE_OPENAI', 'openai'],
+  ['CLAUDE_CODE_USE_GITHUB', 'github'],
  ['CLAUDE_CODE_USE_GEMINI', 'gemini'],
  ['CLAUDE_CODE_USE_BEDROCK', 'bedrock'],
  ['CLAUDE_CODE_USE_VERTEX', 'vertex'],
@@ -52,3 +56,11 @@ test.each([
    expect(usesAnthropicAccountFlow()).toBe(false)
  },
 )
+
+test('GEMINI takes precedence over GitHub when both are set', () => {
+  clearProviderEnv()
+  process.env.CLAUDE_CODE_USE_GEMINI = '1'
+  process.env.CLAUDE_CODE_USE_GITHUB = '1'
+
+  expect(getAPIProvider()).toBe('gemini')
+})
--- a/src/utils/model/providers.ts
+++ b/src/utils/model/providers.ts
@@ -1,11 +1,20 @@
 import type { AnalyticsMetadata_I_VERIFIED_THIS_IS_NOT_CODE_OR_FILEPATHS } from '../../services/analytics/index.js'
 import { isEnvTruthy } from '../envUtils.js'

-export type APIProvider = 'firstParty' | 'bedrock' | 'vertex' | 'foundry' | 'openai' | 'gemini'
+export type APIProvider =
+  | 'firstParty'
+  | 'bedrock'
+  | 'vertex'
+  | 'foundry'
+  | 'openai'
+  | 'gemini'
+  | 'github'

 export function getAPIProvider(): APIProvider {
  return isEnvTruthy(process.env.CLAUDE_CODE_USE_GEMINI)
    ? 'gemini'
+    : isEnvTruthy(process.env.CLAUDE_CODE_USE_GITHUB)
+      ? 'github'
      : isEnvTruthy(process.env.CLAUDE_CODE_USE_OPENAI)
        ? 'openai'
        : isEnvTruthy(process.env.CLAUDE_CODE_USE_BEDROCK)
--- a/src/utils/providerDiscovery.ts
+++ b/src/utils/providerDiscovery.ts
@@ -1,6 +1,7 @@
 import type { OllamaModelDescriptor } from './providerRecommendation.ts'

 export const DEFAULT_OLLAMA_BASE_URL = 'http://localhost:11434'
+export const DEFAULT_ATOMIC_CHAT_BASE_URL = 'http://127.0.0.1:1337'

 function withTimeoutSignal(timeoutMs: number): {
  signal: AbortSignal
@@ -35,6 +36,23 @@ export function getOllamaChatBaseUrl(baseUrl?: string): string {
  return `${getOllamaApiBaseUrl(baseUrl)}/v1`
 }

+export function getAtomicChatApiBaseUrl(baseUrl?: string): string {
+  const parsed = new URL(
+    baseUrl || process.env.ATOMIC_CHAT_BASE_URL || DEFAULT_ATOMIC_CHAT_BASE_URL,
+  )
+  const pathname = trimTrailingSlash(parsed.pathname)
+  parsed.pathname = pathname.endsWith('/v1')
+    ? pathname.slice(0, -3) || '/'
+    : pathname || '/'
+  parsed.search = ''
+  parsed.hash = ''
+  return trimTrailingSlash(parsed.toString())
+}
+
+export function getAtomicChatChatBaseUrl(baseUrl?: string): string {
+  return `${getAtomicChatApiBaseUrl(baseUrl)}/v1`
+}
+
 export async function hasLocalOllama(baseUrl?: string): Promise<boolean> {
  const { signal, clear } = withTimeoutSignal(1200)
  try {
@@ -93,6 +111,48 @@ export async function listOllamaModels(
  }
 }

+export async function hasLocalAtomicChat(baseUrl?: string): Promise<boolean> {
+  const { signal, clear } = withTimeoutSignal(1200)
+  try {
+    const response = await fetch(`${getAtomicChatChatBaseUrl(baseUrl)}/models`, {
+      method: 'GET',
+      signal,
+    })
+    return response.ok
+  } catch {
+    return false
+  } finally {
+    clear()
+  }
+}
+
+export async function listAtomicChatModels(
+  baseUrl?: string,
+): Promise<string[]> {
+  const { signal, clear } = withTimeoutSignal(5000)
+  try {
+    const response = await fetch(`${getAtomicChatChatBaseUrl(baseUrl)}/models`, {
+      method: 'GET',
+      signal,
+    })
+    if (!response.ok) {
+      return []
+    }
+
+    const data = (await response.json()) as {
+      data?: Array<{ id?: string }>
+    }
+
+    return (data.data ?? [])
+      .filter(model => Boolean(model.id))
+      .map(model => model.id!)
+  } catch {
+    return []
+  } finally {
+    clear()
+  }
+}
+
 export async function benchmarkOllamaModel(
  modelName: string,
  baseUrl?: string,
--- a/src/utils/providerProfile.test.ts
+++ b/src/utils/providerProfile.test.ts
@@ -6,6 +6,7 @@ import test from 'node:test'

 import {
  buildStartupEnvFromProfile,
+  buildAtomicChatProfileEnv,
  buildCodexProfileEnv,
  buildGeminiProfileEnv,
  buildLaunchEnv,
@@ -529,3 +530,72 @@ test('auto profile falls back to openai when no viable ollama model exists', ()
  assert.equal(selectAutoProfile(null), 'openai')
  assert.equal(selectAutoProfile('qwen2.5-coder:7b'), 'ollama')
 })
+
+// ── Atomic Chat profile tests ────────────────────────────────────────────────
+
+test('atomic-chat profiles never persist openai api keys', () => {
+  const env = buildAtomicChatProfileEnv('some-local-model', {
+    getAtomicChatChatBaseUrl: () => 'http://127.0.0.1:1337/v1',
+  })
+
+  assert.deepEqual(env, {
+    OPENAI_BASE_URL: 'http://127.0.0.1:1337/v1',
+    OPENAI_MODEL: 'some-local-model',
+  })
+  assert.equal('OPENAI_API_KEY' in env, false)
+})
+
+test('atomic-chat profiles respect custom base url', () => {
+  const env = buildAtomicChatProfileEnv('my-model', {
+    baseUrl: 'http://192.168.1.100:1337',
+    getAtomicChatChatBaseUrl: (baseUrl?: string) =>
+      baseUrl ? `${baseUrl}/v1` : 'http://127.0.0.1:1337/v1',
+  })
+
+  assert.equal(env.OPENAI_BASE_URL, 'http://192.168.1.100:1337/v1')
+  assert.equal(env.OPENAI_MODEL, 'my-model')
+})
+
+test('matching persisted atomic-chat env is reused for atomic-chat launch', async () => {
+  const env = await buildLaunchEnv({
+    profile: 'atomic-chat',
+    persisted: profile('atomic-chat', {
+      OPENAI_BASE_URL: 'http://127.0.0.1:1337/v1',
+      OPENAI_MODEL: 'llama-3.1-8b',
+    }),
+    goal: 'balanced',
+    processEnv: {},
+    getAtomicChatChatBaseUrl: () => 'http://127.0.0.1:1337/v1',
+    resolveAtomicChatDefaultModel: async () => 'other-model',
+  })
+
+  assert.equal(env.OPENAI_BASE_URL, 'http://127.0.0.1:1337/v1')
+  assert.equal(env.OPENAI_MODEL, 'llama-3.1-8b')
+  assert.equal(env.OPENAI_API_KEY, undefined)
+  assert.equal(env.CODEX_API_KEY, undefined)
+})
+
+test('atomic-chat launch ignores mismatched persisted openai env', async () => {
+  const env = await buildLaunchEnv({
+    profile: 'atomic-chat',
+    persisted: profile('openai', {
+      OPENAI_BASE_URL: 'https://api.openai.com/v1',
+      OPENAI_MODEL: 'gpt-4o',
+      OPENAI_API_KEY: 'sk-persisted',
+    }),
+    goal: 'balanced',
+    processEnv: {
+      OPENAI_API_KEY: 'sk-live',
+      CODEX_API_KEY: 'codex-live',
+      CHATGPT_ACCOUNT_ID: 'acct_live',
+    },
+    getAtomicChatChatBaseUrl: () => 'http://127.0.0.1:1337/v1',
+    resolveAtomicChatDefaultModel: async () => 'local-model',
+  })
+
+  assert.equal(env.OPENAI_BASE_URL, 'http://127.0.0.1:1337/v1')
+  assert.equal(env.OPENAI_MODEL, 'local-model')
+  assert.equal(env.OPENAI_API_KEY, undefined)
+  assert.equal(env.CODEX_API_KEY, undefined)
+  assert.equal(env.CHATGPT_ACCOUNT_ID, undefined)
+})
--- a/src/utils/providerProfile.ts
+++ b/src/utils/providerProfile.ts
@@ -44,7 +44,7 @@ const SECRET_ENV_KEYS = [
  'GOOGLE_API_KEY',
 ] as const

-export type ProviderProfile = 'openai' | 'ollama' | 'codex' | 'gemini'
+export type ProviderProfile = 'openai' | 'ollama' | 'codex' | 'gemini' | 'atomic-chat'

 export type ProfileEnv = {
  OPENAI_BASE_URL?: string
@@ -89,7 +89,8 @@ export function isProviderProfile(value: unknown): value is ProviderProfile {
    value === 'openai' ||
    value === 'ollama' ||
    value === 'codex' ||
-    value === 'gemini'
+    value === 'gemini' ||
+    value === 'atomic-chat'
  )
 }

@@ -202,6 +203,19 @@ export function buildOllamaProfileEnv(
  }
 }

+export function buildAtomicChatProfileEnv(
+  model: string,
+  options: {
+    baseUrl?: string | null
+    getAtomicChatChatBaseUrl: (baseUrl?: string) => string
+  },
+): ProfileEnv {
+  return {
+    OPENAI_BASE_URL: options.getAtomicChatChatBaseUrl(options.baseUrl ?? undefined),
+    OPENAI_MODEL: model,
+  }
+}
+
 export function buildGeminiProfileEnv(options: {
  model?: string | null
  baseUrl?: string | null
@@ -385,6 +399,7 @@ export function hasExplicitProviderSelection(
 ): boolean {
  return (
    processEnv.CLAUDE_CODE_USE_OPENAI !== undefined ||
+    processEnv.CLAUDE_CODE_USE_GITHUB !== undefined ||
    processEnv.CLAUDE_CODE_USE_GEMINI !== undefined ||
    processEnv.CLAUDE_CODE_USE_BEDROCK !== undefined ||
    processEnv.CLAUDE_CODE_USE_VERTEX !== undefined ||
@@ -405,6 +420,8 @@ export async function buildLaunchEnv(options: {
  processEnv?: NodeJS.ProcessEnv
  getOllamaChatBaseUrl?: (baseUrl?: string) => string
  resolveOllamaDefaultModel?: (goal: RecommendationGoal) => Promise<string>
+  getAtomicChatChatBaseUrl?: (baseUrl?: string) => string
+  resolveAtomicChatDefaultModel?: () => Promise<string | null>
 }): Promise<NodeJS.ProcessEnv> {
  const processEnv = options.processEnv ?? process.env
  const persistedEnv =
@@ -456,6 +473,7 @@ export async function buildLaunchEnv(options: {
    }

    delete env.CLAUDE_CODE_USE_OPENAI
+    delete env.CLAUDE_CODE_USE_GITHUB

    env.GEMINI_MODEL =
      shellGeminiModel ||
@@ -490,6 +508,7 @@ export async function buildLaunchEnv(options: {
  }

  delete env.CLAUDE_CODE_USE_GEMINI
+  delete env.CLAUDE_CODE_USE_GITHUB
  delete env.GEMINI_API_KEY
  delete env.GEMINI_MODEL
  delete env.GEMINI_BASE_URL
@@ -514,6 +533,26 @@ export async function buildLaunchEnv(options: {
    return env
  }

+  if (options.profile === 'atomic-chat') {
+    const getAtomicChatBaseUrl =
+      options.getAtomicChatChatBaseUrl ?? (() => 'http://127.0.0.1:1337/v1')
+    const resolveModel =
+      options.resolveAtomicChatDefaultModel ?? (async () => null as string | null)
+
+    env.OPENAI_BASE_URL = persistedEnv.OPENAI_BASE_URL || getAtomicChatBaseUrl()
+    env.OPENAI_MODEL =
+      persistedEnv.OPENAI_MODEL ||
+      (await resolveModel()) ||
+      ''
+
+    delete env.OPENAI_API_KEY
+    delete env.CODEX_API_KEY
+    delete env.CHATGPT_ACCOUNT_ID
+    delete env.CODEX_ACCOUNT_ID
+
+    return env
+  }
+
  if (options.profile === 'codex') {
    env.OPENAI_BASE_URL =
      persistedOpenAIBaseUrl && isCodexBaseUrl(persistedOpenAIBaseUrl)
--- a/src/utils/swarm/spawnUtils.ts
+++ b/src/utils/swarm/spawnUtils.ts
@@ -99,6 +99,18 @@ const TEAMMATE_ENV_VARS = [
  'CLAUDE_CODE_USE_BEDROCK',
  'CLAUDE_CODE_USE_VERTEX',
  'CLAUDE_CODE_USE_FOUNDRY',
+  'CLAUDE_CODE_USE_GITHUB',
+  'CLAUDE_CODE_USE_GEMINI',
+  'CLAUDE_CODE_USE_OPENAI',
+  'GITHUB_TOKEN',
+  'GH_TOKEN',
+  'OPENAI_API_KEY',
+  'OPENAI_BASE_URL',
+  'OPENAI_MODEL',
+  'GEMINI_API_KEY',
+  'GEMINI_BASE_URL',
+  'GEMINI_MODEL',
+  'GOOGLE_API_KEY',
  // Custom API endpoint
  'ANTHROPIC_BASE_URL',
  // Config directory override
--- a/test_atomic_chat_provider.py
+++ b/test_atomic_chat_provider.py
@@ -0,0 +1,130 @@
+"""
+test_atomic_chat_provider.py
+Run: pytest test_atomic_chat_provider.py -v
+"""
+
+import pytest
+from unittest.mock import AsyncMock, MagicMock, patch
+from atomic_chat_provider import (
+    atomic_chat,
+    list_atomic_chat_models,
+    check_atomic_chat_running,
+)
+
+
+@pytest.mark.asyncio
+async def test_atomic_chat_running_true():
+    mock_response = MagicMock()
+    mock_response.status_code = 200
+    with patch("atomic_chat_provider.httpx.AsyncClient") as MockClient:
+        MockClient.return_value.__aenter__.return_value.get = AsyncMock(return_value=mock_response)
+        result = await check_atomic_chat_running()
+    assert result is True
+
+
+@pytest.mark.asyncio
+async def test_atomic_chat_running_false_on_exception():
+    with patch("atomic_chat_provider.httpx.AsyncClient") as MockClient:
+        MockClient.return_value.__aenter__.return_value.get = AsyncMock(side_effect=Exception("refused"))
+        result = await check_atomic_chat_running()
+    assert result is False
+
+
+@pytest.mark.asyncio
+async def test_list_models_returns_ids():
+    mock_response = MagicMock()
+    mock_response.status_code = 200
+    mock_response.json.return_value = {
+        "data": [{"id": "llama-3.1-8b"}, {"id": "mistral-7b"}],
+    }
+    mock_response.raise_for_status = MagicMock()
+    with patch("atomic_chat_provider.httpx.AsyncClient") as MockClient:
+        MockClient.return_value.__aenter__.return_value.get = AsyncMock(return_value=mock_response)
+        models = await list_atomic_chat_models()
+    assert "llama-3.1-8b" in models
+    assert "mistral-7b" in models
+
+
+@pytest.mark.asyncio
+async def test_list_models_empty_on_failure():
+    with patch("atomic_chat_provider.httpx.AsyncClient") as MockClient:
+        MockClient.return_value.__aenter__.return_value.get = AsyncMock(side_effect=Exception("down"))
+        models = await list_atomic_chat_models()
+    assert models == []
+
+
+@pytest.mark.asyncio
+async def test_atomic_chat_returns_anthropic_format():
+    mock_response = MagicMock()
+    mock_response.raise_for_status = MagicMock()
+    mock_response.json.return_value = {
+        "id": "chatcmpl-abc123",
+        "choices": [{"message": {"content": "42 is the answer."}}],
+        "usage": {"prompt_tokens": 10, "completion_tokens": 8},
+    }
+    with patch("atomic_chat_provider.httpx.AsyncClient") as MockClient:
+        MockClient.return_value.__aenter__.return_value.post = AsyncMock(return_value=mock_response)
+        result = await atomic_chat(
+            model="llama-3.1-8b",
+            messages=[{"role": "user", "content": "What is 6*7?"}],
+        )
+    assert result["type"] == "message"
+    assert result["role"] == "assistant"
+    assert "42" in result["content"][0]["text"]
+    assert result["usage"]["input_tokens"] == 10
+    assert result["usage"]["output_tokens"] == 8
+
+
+@pytest.mark.asyncio
+async def test_atomic_chat_prepends_system():
+    captured = {}
+
+    async def mock_post(url, json=None, **kwargs):
+        captured.update(json or {})
+        m = MagicMock()
+        m.raise_for_status = MagicMock()
+        m.json.return_value = {
+            "id": "chatcmpl-xyz",
+            "choices": [{"message": {"content": "ok"}}],
+            "usage": {"prompt_tokens": 1, "completion_tokens": 1},
+        }
+        return m
+
+    with patch("atomic_chat_provider.httpx.AsyncClient") as MockClient:
+        MockClient.return_value.__aenter__.return_value.post = mock_post
+        await atomic_chat(
+            model="llama-3.1-8b",
+            messages=[{"role": "user", "content": "Hi"}],
+            system="Be helpful.",
+        )
+    assert captured["messages"][0]["role"] == "system"
+    assert "helpful" in captured["messages"][0]["content"]
+
+
+@pytest.mark.asyncio
+async def test_atomic_chat_sends_correct_payload():
+    captured = {}
+
+    async def mock_post(url, json=None, **kwargs):
+        captured.update(json or {})
+        m = MagicMock()
+        m.raise_for_status = MagicMock()
+        m.json.return_value = {
+            "id": "chatcmpl-xyz",
+            "choices": [{"message": {"content": "ok"}}],
+            "usage": {"prompt_tokens": 1, "completion_tokens": 1},
+        }
+        return m
+
+    with patch("atomic_chat_provider.httpx.AsyncClient") as MockClient:
+        MockClient.return_value.__aenter__.return_value.post = mock_post
+        await atomic_chat(
+            model="test-model",
+            messages=[{"role": "user", "content": "Test"}],
+            max_tokens=2048,
+            temperature=0.5,
+        )
+    assert captured["model"] == "test-model"
+    assert captured["max_tokens"] == 2048
+    assert captured["temperature"] == 0.5
+    assert captured["stream"] is False