feat: add MiniMax provider with M3.1 Flash default

README.md

@@ -43,7 +43,7 @@ App-owned runtimes, provider settings, chat history, attachments, caches, and lo
 | Feature | Details |
 |---|---|
 | **Official Claude Code** | Uses Anthropic's official CLI and Agent SDK — no replacement agent engine |
-| **9 AI Providers** | NVIDIA NIM · DeepSeek · OpenRouter · Google Gemini · Anthropic · OpenAI · Ollama · LM Studio · Custom API |
+| **10 AI Providers** | NVIDIA NIM · DeepSeek · MiniMax · OpenRouter · Google Gemini · Anthropic · OpenAI · Ollama · LM Studio · Custom API |
 | **Portable Runtime** | Node.js, Claude Code, caches, settings, and app data stay in the project folder |
 | **Studio Dashboard** | Streaming chat, Markdown, code highlighting, compact tool steps, approvals, attachments, and diagnostics |
 | **Shared Claude History** | GUI and portable terminal sessions use real Claude Code conversation IDs and can resume the same history |
@@ -123,6 +123,7 @@ Pressing Enter selects the dashboard.
 |---|---|---|
 | **NVIDIA NIM** | Chat Completions adapter | [build.nvidia.com](https://build.nvidia.com) |
 | **DeepSeek** | Anthropic-compatible API | [platform.deepseek.com](https://platform.deepseek.com) |
+| **MiniMax** | Chat Completions adapter | [platform.minimax.io](https://platform.minimax.io) |
 | **OpenRouter** | Anthropic-compatible API | [openrouter.ai](https://openrouter.ai) |
 | **Google Gemini** | Chat Completions adapter | [aistudio.google.com](https://aistudio.google.com) |
 | **Anthropic Claude** | Native Messages API / terminal login | [console.anthropic.com](https://console.anthropic.com) |
@@ -133,6 +134,14 @@ Pressing Enter selects the dashboard.
 
 > **Non-Claude models are experimental and are not supported by Anthropic.** Tool calling, images, context limits, resume behavior, and other Claude Code features depend on the selected model and provider.
 
+### MiniMax
+
+The MiniMax preset defaults to **MiniMax-M3.1-Flash** and offers **MiniMax-M3** as the second model. Both have a 1,000,000-token context window. Existing saved models and manually entered IDs remain unchanged. Use `https://api.minimax.io/v1` globally or `https://api.minimaxi.com/v1` in China with the matching API key.
+
+The built-in Chat Completions adapter uses adaptive thinking by default for these models. Flash always thinks and defaults to `max` reasoning effort; its supported effort values are `low`, `medium`, `high`, `xhigh`, and `max`. M3 also accepts disabled thinking and ignores effort. Flash does not accept `service_tier` or `reasoning_split=false`. The dashboard preserves its existing image attachment support; video input is not implemented by this adapter.
+
+See MiniMax's [model documentation](https://platform.minimax.io/docs/guides/models-intro.md) and [Chat Completions API](https://platform.minimax.io/docs/api-reference/text-chat-openai.md).
+
 ## Custom OpenAI-Compatible Provider
 
 OpenAI-compatible providers can use:

lib/adapter.mjs

@@ -19,9 +19,17 @@ function toolResultContent(block) {
   for(const part of block.content){if(part.type==='text')text.push(part.text);else if(part.type==='image')images.push(imageContent(part));else throw new Error(`Unsupported content: ${part.type}`);}
   return {text:text.join('\n'),images};
 }
-export function translateRequest(body, model, {toolMetadata}={}) {
+const MINIMAX_FLASH='MiniMax-M3.1-Flash';
+const MINIMAX_M3='MiniMax-M3';
+
+export function translateRequest(body, model, {toolMetadata, provider}={}) {
   if (!Array.isArray(body.messages)) throw new Error('messages must be an array');
-  if (body.thinking && body.thinking.type !== 'disabled') throw new Error('Extended thinking is unavailable through the Chat Completions adapter. Disable thinking for this provider.');
+  const minimax = provider === 'minimax' && [MINIMAX_FLASH, MINIMAX_M3].includes(model);
+  if (minimax) {
+    if (body.thinking && !['adaptive','disabled'].includes(body.thinking.type)) throw new Error('MiniMax thinking must be adaptive or disabled');
+    if (model === MINIMAX_FLASH && body.thinking?.type === 'disabled') throw new Error('MiniMax-M3.1-Flash thinking cannot be disabled');
+    if (model === MINIMAX_FLASH && (body.service_tier !== undefined || body.reasoning_split === false)) throw new Error('MiniMax-M3.1-Flash does not support service_tier or reasoning_split=false');
+  } else if (body.thinking && body.thinking.type !== 'disabled') throw new Error('Extended thinking is unavailable through the Chat Completions adapter. Disable thinking for this provider.');
   const messages = [];
   if (body.system) messages.push({ role: 'system', content: textContent(body.system) });
   for (const message of body.messages) {
@@ -47,6 +55,14 @@ export function translateRequest(body, model, {toolMetadata}={}) {
     if (content.length || calls.length) messages.push({ role: message.role, content: content.length ? content.every(b=>b.type==='text') ? content.map(b=>b.text).join('\n') : content : null, ...(calls.length ? { tool_calls: calls } : {}) });
   }
   const request = { model, messages, stream: !!body.stream };
+  if (minimax) {
+    request.thinking = { type: body.thinking?.type || 'adaptive' };
+    if (model === MINIMAX_FLASH) {
+      const effort = body.reasoning_effort ?? body.output_config?.effort ?? 'max';
+      if (!['low','medium','high','xhigh','max'].includes(effort)) throw new Error('Unsupported MiniMax-M3.1-Flash reasoning effort');
+      request.reasoning_effort = effort;
+    }
+  }
   if (body.max_tokens) request.max_tokens = body.max_tokens;
   if (body.temperature !== undefined) request.temperature = body.temperature;
   if (body.top_p !== undefined) request.top_p = body.top_p;
@@ -139,7 +155,7 @@ export async function startAdapter(profile, { fetchImpl = fetch, onError = () =>
         return json(res, 200, { input_tokens: count, estimated: true });
       }
       const remember=(id,metadata)=>{if(id&&metadata)toolMetadata.set(id,metadata);};
-      const request = translateRequest(body, profile.model,{toolMetadata});
+      const request = translateRequest(body, profile.model,{toolMetadata,provider:profile.provider});
       const upstream = await fetchImpl(`${profile.baseUrl.replace(/\/+$/,'')}/chat/completions`, {
         method: 'POST', headers: { 'Content-Type': 'application/json', ...(profile.key ? { Authorization: `Bearer ${profile.key}` } : {}) }, body: JSON.stringify(request), redirect: 'error', signal: AbortSignal.any([controller.signal, AbortSignal.timeout(180000)])
       });

lib/agent.mjs

@@ -65,7 +65,9 @@ export class AgentManager {
         includePartialMessages:true,settingSources:[],systemPrompt:{type:'preset',preset:'claude_code'},
         maxTurns:40,abortController:run.controller,
         ...(session.sdkSessionId?{resume:session.sdkSessionId}:{}),
-        ...(adapter?{thinking:{type:'disabled'},effort:'low'}:{}),
+        ...(adapter?(profile.provider==='minimax'&&(!profile.adapter||profile.adapter==='builtin')&&['MiniMax-M3.1-Flash','MiniMax-M3'].includes(profile.model)
+          ?{thinking:{type:'adaptive'},...(profile.model==='MiniMax-M3.1-Flash'?{effort:'max'}:{})}
+          :{thinking:{type:'disabled'},effort:'low'}):{}),
         canUseTool:async(name,input,context)=>{
           clearResponseTimer();
           if(run.cancelled)return {behavior:'deny',message:'Session cancelled'};

lib/providers.mjs

@@ -9,6 +9,7 @@ export const PROVIDERS = {
   deepseek: { name: 'DeepSeek', short: 'DS', description: 'Direct Anthropic-compatible API', transport: 'anthropic', baseUrl: 'https://api.deepseek.com/anthropic', modelsUrl: 'https://api.deepseek.com/models' },
   nvidia: { name: 'NVIDIA NIM', short: 'NV', description: 'Cloud inference · local adapter', transport: 'openai', baseUrl: 'https://integrate.api.nvidia.com/v1' },
   gemini: { name: 'Google Gemini', short: 'GE', description: 'Google AI · local adapter', transport: 'openai', baseUrl: 'https://generativelanguage.googleapis.com/v1beta/openai' },
+  minimax: { name: 'MiniMax', short: 'MM', description: 'MiniMax · OpenAI-compatible API', transport: 'openai', baseUrl: 'https://api.minimax.io/v1', modelsUrl: 'https://api.minimax.io/v1/models', defaultModel: 'MiniMax-M3.1-Flash', models: [{ id: 'MiniMax-M3.1-Flash', name: 'MiniMax-M3.1-Flash', contextWindow: 1000000 }, { id: 'MiniMax-M3', name: 'MiniMax-M3', contextWindow: 1000000 }], endpoints: [{ region: 'global_en', baseUrl: 'https://api.minimax.io/v1', anthropicBaseUrl: 'https://api.minimax.io/anthropic', docsRoot: 'https://platform.minimax.io/docs' }, { region: 'cn_zh', baseUrl: 'https://api.minimaxi.com/v1', anthropicBaseUrl: 'https://api.minimaxi.com/anthropic', docsRoot: 'https://platform.minimaxi.com/docs' }] },
   openai: { name: 'OpenAI', short: 'OA', description: 'Chat Completions · local adapter', transport: 'openai', baseUrl: 'https://api.openai.com/v1' },
   ollama: { name: 'Ollama', short: 'OL', description: 'Local models · offline inference', transport: 'anthropic', baseUrl: 'http://127.0.0.1:11434', local: true },
   lmstudio: { name: 'LM Studio', short: 'LM', description: 'Your local model server', transport: 'anthropic', baseUrl: 'http://127.0.0.1:1234', local: true },
@@ -63,6 +64,7 @@ export async function discoverModels(profile, signal) {
   const res = await fetch(providerURL(profile), { headers:providerHeaders(profile), signal:signal||AbortSignal.timeout(15000), redirect:'error' });
   if (!res.ok) throw new Error(`Model discovery returned HTTP ${res.status}. Check the endpoint/credential or enter a model manually.`);
   const body = await res.json();
+  if (profile.provider === 'minimax') return p.models.map(model => ({...model}));
   return (body.data || body.models || []).map(m => {
     const raw=m.context_length??m.context_window??m.max_context_length??m.max_input_tokens??m.inputTokenLimit;
     const contextWindow=Number(raw);

tests/backend.test.mjs

@@ -16,8 +16,8 @@ const { pickWorkspace }=await import('../lib/workspace-picker.mjs');
 const { nativeSessionId }=await import('../lib/native-sessions.mjs');
 test.after(()=>rmSync(temp,{recursive:true,force:true}));
 const profile={provider:'custom',model:'test-model',auth:'api',baseUrl:'http://127.0.0.1:9099/v1',key:'test-secret-not-real'};
-test('all nine providers and safe configuration round trips',()=>{
-  assert.equal(Object.keys(PROVIDERS).length,9);saveProfile({...profile,contextWindow:128000});
+test('all ten providers and safe configuration round trips',()=>{
+  assert.equal(Object.keys(PROVIDERS).length,10);assert.equal(PROVIDERS.minimax.defaultModel,'MiniMax-M3.1-Flash');assert.equal(PROVIDERS.minimax.baseUrl,'https://api.minimax.io/v1');assert.equal(PROVIDERS.minimax.endpoints[1].baseUrl,'https://api.minimaxi.com/v1');saveProfile({...profile,contextWindow:128000});
   assert.equal(readConfig().profiles.custom.contextWindow,128000);assert.throws(()=>saveProfile({...profile,contextWindow:100}),/Context window/);
   assert.equal(readConfig().profiles.custom.key,profile.key);assert.equal(publicConfig().profiles.custom.hasKey,true);assert.equal(publicConfig().profiles.custom.key,undefined);
   saveProfile({...profile,key:undefined,model:'next'});assert.equal(readConfig().profiles.custom.key,profile.key);saveProfile(profile);