feat(providers): add Atomic Chat local OpenAI-compatible provider

src/index.ts

@@ -149,6 +149,7 @@ const PROVIDER_OPTIONS: Array<{ key: ProviderName; label: string }> = [
   { key: 'grok', label: 'Grok (xAI)' },
   { key: 'ollamaCloud', label: 'Ollama Cloud' },
   { key: 'ollamaLocal', label: 'Ollama Local' },
+  { key: 'atomicChat', label: 'Atomic Chat' },
   { key: 'openaiCompat', label: 'OpenAI Compilations' },
   { key: 'mimo', label: 'MiMo (Xiaomi)' },
   { key: 'mimoTokenPlan', label: 'MiMo Token Plan (Xiaomi)' },
@@ -500,6 +501,65 @@ async function promptOllamaLocalModelSelection(config: MercuryConfig, isReconfig
   }
 }
 
+async function promptAtomicChatModelSelection(config: MercuryConfig, isReconfig: boolean): Promise<{ baseUrl?: string; model?: string; skipped: boolean }> {
+  const existingConfig = config.providers.atomicChat;
+  const defaultBaseUrl = existingConfig.baseUrl || 'http://127.0.0.1:1337/v1';
+
+  const baseUrlPrompt = isReconfig && existingConfig.baseUrl
+    ? chalk.white(`  Atomic Chat base URL [${existingConfig.baseUrl}]: `)
+    : chalk.white(`  Atomic Chat base URL [${defaultBaseUrl}]: `);
+  const baseUrlInput = await ask(baseUrlPrompt);
+  if (baseUrlInput.toLowerCase() === 'none') {
+    if (isReconfig && existingConfig.baseUrl) {
+      return { baseUrl: existingConfig.baseUrl, model: existingConfig.model, skipped: true };
+    }
+    return { skipped: true };
+  }
+  const baseUrl = baseUrlInput || defaultBaseUrl;
+  const baseUrlError = validateBaseUrl(baseUrl);
+  if (baseUrlError) {
+    console.log(chalk.red(`  ${baseUrlError}`));
+    if (isReconfig && existingConfig.baseUrl) {
+      return { baseUrl: existingConfig.baseUrl, model: existingConfig.model, skipped: true };
+    }
+    return { skipped: true };
+  }
+
+  console.log(chalk.dim('  Fetching Atomic Chat models...'));
+  try {
+    const catalog = await fetchProviderModelCatalog('atomicChat', {
+      ...existingConfig,
+      baseUrl,
+    });
+    const model = await chooseProviderModel(
+      'Atomic Chat',
+      catalog.recommendedModel,
+      catalog.models,
+    );
+    return { baseUrl, model, skipped: false };
+  } catch (error) {
+    const message = error instanceof ProviderModelFetchError
+      ? error.message
+      : 'Mercury could not fetch Atomic Chat models.';
+    console.log(chalk.yellow(`  ${message}`));
+    console.log(chalk.dim('  Make sure Atomic Chat is running and a model is loaded, or enter the model id manually.'));
+    console.log(chalk.dim('  You can run `mercury doctor` later after starting Atomic Chat.'));
+
+    const manualModel = await ask(chalk.white('  Atomic Chat model id (Enter to skip Atomic Chat for now): '));
+    if (!manualModel) {
+      return { skipped: true };
+    }
+
+    const modelError = validateModelName(manualModel);
+    if (modelError) {
+      console.log(chalk.red(`  ${modelError}`));
+      return { skipped: true };
+    }
+
+    return { baseUrl, model: manualModel, skipped: false };
+  }
+}
+
 async function promptOpenAICompatSetup(config: MercuryConfig, isReconfig: boolean): Promise<{ baseUrl?: string; apiKey?: string; model?: string; skipped: boolean }> {
   const existingConfig = config.providers.openaiCompat;
 
@@ -1281,6 +1341,16 @@ async function configure(existingConfig?: MercuryConfig): Promise<void> {
         continue;
       }
 
+      if (provider === 'atomicChat') {
+        const result = await promptAtomicChatModelSelection(config, isReconfig);
+        if (!result.skipped && result.baseUrl && result.model) {
+          config.providers.atomicChat.baseUrl = result.baseUrl;
+          config.providers.atomicChat.model = result.model;
+          config.providers.atomicChat.enabled = true;
+        }
+        continue;
+      }
+
       if (provider === 'openaiCompat') {
         const result = await promptOpenAICompatSetup(config, isReconfig);
         if (!result.skipped && result.baseUrl && result.model) {

src/providers/registry.ts

@@ -10,7 +10,7 @@ async function createProvider(pc: ProviderConfig): Promise<BaseProvider> {
   } else if (pc.name === 'deepseek') {
     const { DeepSeekProvider } = await import('./deepseek.js');
     return new DeepSeekProvider(pc);
-  } else if (pc.name === 'ollamaLocal') {
+  } else if (pc.name === 'ollamaLocal' || pc.name === 'atomicChat') {
     // Route through OpenAI-compatible provider — local Ollama exposes
     // /v1/chat/completions since v0.1.14. The ollama-ai-provider package
     // declares specificationVersion = "v1" which is incompatible with
@@ -55,6 +55,7 @@ export class ProviderRegistry {
       config.providers.grok,
       config.providers.ollamaCloud,
       config.providers.ollamaLocal,
+      config.providers.atomicChat,
       config.providers.openaiCompat,
       config.providers.mimo,
       config.providers.mimoTokenPlan,

src/utils/config.ts

@@ -54,6 +54,7 @@ export type ProviderName =
   | 'grok'
   | 'ollamaCloud'
   | 'ollamaLocal'
+  | 'atomicChat'
   | 'openaiCompat'
   | 'mimo'
   | 'mimoTokenPlan'
@@ -74,6 +75,7 @@ export interface MercuryConfig {
     grok: ProviderConfig;
     ollamaCloud: ProviderConfig;
     ollamaLocal: ProviderConfig;
+    atomicChat: ProviderConfig;
     openaiCompat: ProviderConfig;
     mimo: ProviderConfig;
     mimoTokenPlan: ProviderConfig;
@@ -246,6 +248,13 @@ export function getDefaultConfig(): MercuryConfig {
         model: getEnv('OLLAMA_LOCAL_MODEL', ''),
         enabled: getEnvBool('OLLAMA_LOCAL_ENABLED', false),
       },
+      atomicChat: {
+        name: 'atomicChat',
+        apiKey: '',
+        baseUrl: getEnv('ATOMIC_CHAT_BASE_URL', 'http://127.0.0.1:1337/v1'),
+        model: getEnv('ATOMIC_CHAT_MODEL', ''),
+        enabled: getEnvBool('ATOMIC_CHAT_ENABLED', false),
+      },
       openaiCompat: {
         name: 'openaiCompat',
         apiKey: getEnv('OPENAI_COMPAT_API_KEY', ''),
@@ -446,7 +455,7 @@ export function getActiveProviders(config: MercuryConfig): ProviderConfig[] {
 
 export function isProviderConfigured(provider: ProviderConfig): boolean {
   if (!provider.enabled) return false;
-  if (provider.name === 'ollamaLocal') {
+  if (provider.name === 'ollamaLocal' || provider.name === 'atomicChat') {
     return provider.baseUrl.length > 0 && provider.model.length > 0;
   }
   if (provider.name === 'ollamaCloud') {

src/utils/provider-models.test.ts

@@ -116,4 +116,14 @@ describe('buildModelCatalog', () => {
 
     expect(catalog.recommendedModel).toBe('model-a');
   });
+
+  it('uses the first model as recommended for Atomic Chat when no preferred list exists', () => {
+    const catalog = buildModelCatalog('atomicChat', [
+      'ornith-9b',
+      'qwen3-coder',
+    ]);
+
+    expect(catalog.recommendedModel).toBe('ornith-9b');
+    expect(catalog.models).toContain('qwen3-coder');
+  });
 });

src/utils/provider-models.ts

@@ -72,6 +72,7 @@ const MIMO_PREFERRED_MODELS = [
 const MIMO_TOKEN_PLAN_PREFERRED_MODELS = MIMO_PREFERRED_MODELS;
 
 const OPENAI_COMPAT_PREFERRED_MODELS = [] as const;
+const ATOMIC_CHAT_PREFERRED_MODELS = [] as const;
 
 const CHATGPT_WEB_PREFERRED_MODELS = [
   'gpt-5.5',
@@ -195,6 +196,7 @@ function chooseRecommendedModel(
     grok: GROK_PREFERRED_MODELS,
     ollamaCloud: OLLAMA_CLOUD_PREFERRED_MODELS,
     ollamaLocal: OLLAMA_LOCAL_PREFERRED_MODELS,
+    atomicChat: ATOMIC_CHAT_PREFERRED_MODELS,
     openaiCompat: OPENAI_COMPAT_PREFERRED_MODELS,
     mimo: MIMO_PREFERRED_MODELS,
     mimoTokenPlan: MIMO_TOKEN_PLAN_PREFERRED_MODELS,
@@ -233,6 +235,7 @@ export function buildModelCatalog(
     grok: GROK_PREFERRED_MODELS,
     ollamaCloud: OLLAMA_CLOUD_PREFERRED_MODELS,
     ollamaLocal: OLLAMA_LOCAL_PREFERRED_MODELS,
+    atomicChat: ATOMIC_CHAT_PREFERRED_MODELS,
     openaiCompat: OPENAI_COMPAT_PREFERRED_MODELS,
     mimo: MIMO_PREFERRED_MODELS,
     mimoTokenPlan: MIMO_TOKEN_PLAN_PREFERRED_MODELS,
@@ -260,7 +263,7 @@ async function fetchOpenAICompatModels(provider: ProviderName, config: ProviderC
     errorMessage = 'Mercury could not fetch models for this Grok key. Please re-enter it.';
   } else if (provider === 'deepseek') {
     errorMessage = 'Mercury could not fetch models for this DeepSeek key. Please re-enter it.';
-  } else if (provider === 'openaiCompat') {
+  } else if (provider === 'openaiCompat' || provider === 'atomicChat') {
     errorMessage = 'Mercury could not fetch models from this server. Please check the base URL and try again.';
   } else {
     errorMessage = 'Mercury could not fetch models for this OpenAI key. Please re-enter it.';
@@ -278,7 +281,7 @@ async function fetchOpenAICompatModels(provider: ProviderName, config: ProviderC
       if (provider === 'deepseek') {
         return id.startsWith('deepseek-');
       }
-      if (provider === 'openaiCompat') {
+      if (provider === 'openaiCompat' || provider === 'atomicChat') {
         return id.length > 0;
       }
       return isOpenAIChatModel(id);
@@ -419,7 +422,7 @@ export async function fetchProviderModelCatalog(
     return fetchOllamaLocalModels(config);
   }
 
-  if (provider === 'openaiCompat') {
+  if (provider === 'openaiCompat' || provider === 'atomicChat') {
     return fetchOpenAICompatModels(provider, config);
   }
 

src/web/api/providers.ts

@@ -26,7 +26,7 @@ providers.post('/api/providers/:name', async (c) => {
   const body = await c.req.json();
   const config = loadConfig();
 
-  const validNames: ProviderName[] = ['openai', 'anthropic', 'deepseek', 'grok', 'ollamaCloud', 'ollamaLocal'];
+  const validNames: ProviderName[] = ['openai', 'anthropic', 'deepseek', 'grok', 'ollamaCloud', 'ollamaLocal', 'atomicChat'];
   if (!validNames.includes(providerName)) {
     return c.json({ error: 'Unknown provider' }, 400);
   }

website/docs/getting-started/setup.mdx

@@ -7,7 +7,7 @@ Run `mercury` for the first time. The onboarding wizard asks for:
 
 - **Your name** — used by the agent to address you
 - **Agent name** — defaults to "Mercury", customize it
-- **LLM providers** — choose one or more: DeepSeek, OpenAI, Anthropic, Grok, Ollama Cloud, Ollama Local, or OpenAI Compatible
+- **LLM providers** — choose one or more: DeepSeek, OpenAI, Anthropic, Grok, Ollama Cloud, Ollama Local, Atomic Chat, or OpenAI Compatible
 - **Model selection** — after each valid key, Mercury fetches recommended chat models and lets you choose a default
 - **Telegram bot token** — optional, for remote access in private 1:1 chats
 

website/docs/reference/provider-fallback.mdx

@@ -27,6 +27,7 @@ This is a session-level switch for the running process. Use `mercury doctor` to
 | **Grok (xAI)** | `grok-4` | `GROK_API_KEY` | OpenAI-compatible endpoint |
 | **Ollama Cloud** | `gpt-oss:120b` | `OLLAMA_CLOUD_API_KEY` | Remote Ollama models via API |
 | **Ollama Local** | `gpt-oss:20b` | No key needed | Local Ollama instance (127.0.0.1:11434) |
+| **Atomic Chat** | Configurable | No key needed | Local Atomic Chat OpenAI API (127.0.0.1:1337/v1) |
 | **OpenAI Compatible** | Configurable | `OPENAI_COMPAT_API_KEY` | Any OpenAI-compatible server |
 
 ### OAuth Providers (No API Key Needed)
@@ -128,6 +129,6 @@ You can also set environment variables directly or edit `~/.mercury/mercury.yaml
 
 :::tip[Environment variables]
 
-Each provider has env vars for its API key, base URL, and model. For example, DeepSeek uses `DEEPSEEK_API_KEY`, `DEEPSEEK_BASE_URL`, and `DEEPSEEK_MODEL`. Ollama Local requires no API key — just set `OLLAMA_LOCAL_ENABLED=true` and `OLLAMA_LOCAL_MODEL` to your preferred local model.
+Each provider has env vars for its API key, base URL, and model. For example, DeepSeek uses `DEEPSEEK_API_KEY`, `DEEPSEEK_BASE_URL`, and `DEEPSEEK_MODEL`. Ollama Local requires no API key — just set `OLLAMA_LOCAL_ENABLED=true` and `OLLAMA_LOCAL_MODEL` to your preferred local model. Atomic Chat uses `ATOMIC_CHAT_ENABLED=true`, `ATOMIC_CHAT_BASE_URL` (default `http://127.0.0.1:1337/v1`), and `ATOMIC_CHAT_MODEL`.
 
 :::

website/src/pages/index.tsx

@@ -877,6 +877,7 @@ export default function LandingPage(): React.ReactElement {
                 { name: 'Grok (xAI)', desc: "xAI's models via OpenAI-compatible endpoint." },
                 { name: 'Ollama Cloud', desc: 'Remote Ollama models via API. No local setup.' },
                 { name: 'Ollama Local', desc: 'On your machine. Zero cost, fully private.' },
+                { name: 'Atomic Chat', desc: 'Local models via Atomic Chat OpenAI-compatible API.' },
               ].map((p, i) => (
                 <div key={i} className="lp-provider-card">
                   <h4>{p.name} {(p as any).badge && <span className="lp-provider-badge">{(p as any).badge}</span>}</h4>