feat: add ENABLE_THINKING setting

.gitignore

@@ -5,3 +5,11 @@
 *.stTheme.cache
 *.sublime-workspace
 *.sublime-project
+logs
+*.log
+npm-debug.log*
+yarn-debug.log*
+yarn-error.log*
+lerna-debug.log*
+node_modules/
+dist

README.md

@@ -127,6 +127,16 @@ or save the option as a preference using this command:
 ai config set SILENT_MODE=true
 ```
 
+### Thinking mode
+
+For OpenAI-compatible endpoints that support `chat_template_kwargs.enable_thinking`, you can explicitly enable or disable thinking:
+
+```bash
+ai config set ENABLE_THINKING=false
+```
+
+When unset, AI Shell does not send `chat_template_kwargs`.
+
 ### Custom API endpoint
 
 You can custom OpenAI API endpoint to set OPENAI_API_ENDPOINT（default: `https://api.openai.com/v1`）
@@ -181,6 +191,7 @@ To get an interactive UI like below:
 │  ○ OpenAI Key
 │  ○ OpenAI API Endpoint
 │  ○ Silent Mode
+│  ○ Enable Thinking
 │  ● Model (gpt-4o-mini)
 │  ○ Language
 │  ○ Cancel

src/commands/chat.ts

@@ -20,6 +20,7 @@ export default command(
       OPENAI_KEY: key,
       OPENAI_API_ENDPOINT: apiEndpoint,
       MODEL: model,
+      ENABLE_THINKING: enableThinking,
     } = await getConfig();
     const chatHistory: ChatCompletionRequestMessage[] = [];
 
@@ -51,6 +52,7 @@ export default command(
         key,
         model,
         apiEndpoint,
+        enableThinking,
       });
 
       infoSpin.stop(`${green('AI Shell:')}`);
@@ -77,19 +79,22 @@ async function getResponse({
   key,
   model,
   apiEndpoint,
+  enableThinking,
 }: {
   prompt: string | ChatCompletionRequestMessage[];
   number?: number;
   model?: string;
   key: string;
   apiEndpoint: string;
+  enableThinking?: boolean;
 }) {
   const stream = await generateCompletion({
     prompt,
     key,
     model,
     number,
     apiEndpoint,
+    enableThinking,
   });
 
   const iterableStream = streamToIterable(stream);

src/helpers/completion.ts

@@ -34,11 +34,13 @@ export async function getScriptAndInfo({
   key,
   model,
   apiEndpoint,
+  enableThinking,
 }: {
   prompt: string;
   key: string;
   model?: string;
   apiEndpoint: string;
+  enableThinking?: boolean;
 }) {
   const fullPrompt = getFullPrompt(prompt);
   const stream = await generateCompletion({
@@ -47,6 +49,7 @@ export async function getScriptAndInfo({
     key,
     model,
     apiEndpoint,
+    enableThinking,
   });
   const iterableStream = streamToIterable(stream);
   return {
@@ -61,26 +64,37 @@ export async function generateCompletion({
   key,
   model,
   apiEndpoint,
+  enableThinking,
 }: {
   prompt: string | ChatCompletionRequestMessage[];
   number?: number;
   model?: string;
   key: string;
   apiEndpoint: string;
+  enableThinking?: boolean;
 }) {
   const openAi = getOpenAi(key, apiEndpoint);
   try {
-    const completion = await openAi.createChatCompletion(
-      {
-        model: model || 'gpt-4o-mini',
-        messages: Array.isArray(prompt)
-          ? prompt
-          : [{ role: 'user', content: prompt }],
-        n: Math.min(number, 10),
-        stream: true,
-      },
-      { responseType: 'stream' }
-    );
+    const request = {
+      model: model || 'gpt-4o-mini',
+      messages: Array.isArray(prompt)
+        ? prompt
+        : [{ role: 'user', content: prompt }],
+      n: Math.min(number, 10),
+      stream: true,
+    };
+
+    if (enableThinking !== undefined) {
+      Object.assign(request, {
+        chat_template_kwargs: {
+          enable_thinking: enableThinking,
+        },
+      });
+    }
+
+    const completion = await openAi.createChatCompletion(request as any, {
+      responseType: 'stream',
+    });
 
     return completion.data as unknown as IncomingMessage;
   } catch (err) {
@@ -141,11 +155,13 @@ export async function getExplanation({
   key,
   model,
   apiEndpoint,
+  enableThinking,
 }: {
   script: string;
   key: string;
   model?: string;
   apiEndpoint: string;
+  enableThinking?: boolean;
 }) {
   const prompt = getExplanationPrompt(script);
   const stream = await generateCompletion({
@@ -154,6 +170,7 @@ export async function getExplanation({
     number: 1,
     model,
     apiEndpoint,
+    enableThinking,
   });
   const iterableStream = streamToIterable(stream);
   return { readExplanation: readData(iterableStream) };
@@ -165,12 +182,14 @@ export async function getRevision({
   key,
   model,
   apiEndpoint,
+  enableThinking,
 }: {
   prompt: string;
   code: string;
   key: string;
   model?: string;
   apiEndpoint: string;
+  enableThinking?: boolean;
 }) {
   const fullPrompt = getRevisionPrompt(prompt, code);
   const stream = await generateCompletion({
@@ -179,6 +198,7 @@ export async function getRevision({
     number: 1,
     model,
     apiEndpoint,
+    enableThinking,
   });
   const iterableStream = streamToIterable(stream);
   return {

src/helpers/config.ts

@@ -48,6 +48,13 @@ const configParsers = {
   SILENT_MODE(mode?: string) {
     return String(mode).toLowerCase() === 'true';
   },
+  ENABLE_THINKING(mode?: string) {
+    if (mode === undefined || mode === '') {
+      return undefined;
+    }
+
+    return String(mode).toLowerCase() === 'true';
+  },
   OPENAI_API_ENDPOINT(apiEndpoint?: string) {
     return apiEndpoint || 'https://api.openai.com/v1';
   },
@@ -142,6 +149,14 @@ export const showConfigUI = async () => {
             ? config.SILENT_MODE.toString()
             : i18n.t('(not set)'),
         },
+        {
+          label: i18n.t('Enable Thinking'),
+          value: 'ENABLE_THINKING',
+          hint:
+            config.ENABLE_THINKING === undefined
+              ? i18n.t('(not set)')
+              : config.ENABLE_THINKING.toString(),
+        },
         {
           label: i18n.t('Model'),
           value: 'MODEL',
@@ -187,6 +202,14 @@ export const showConfigUI = async () => {
       });
       if (p.isCancel(silentMode)) return;
       await setConfigs([['SILENT_MODE', silentMode ? 'true' : 'false']]);
+    } else if (choice === 'ENABLE_THINKING') {
+      const enableThinking = await p.confirm({
+        message: i18n.t('Enable thinking?'),
+      });
+      if (p.isCancel(enableThinking)) return;
+      await setConfigs([
+        ['ENABLE_THINKING', enableThinking ? 'true' : 'false'],
+      ]);
     } else if (choice === 'MODEL') {
       const { OPENAI_KEY: key, OPENAI_API_ENDPOINT: apiEndpoint } =
         await getConfig();

src/prompt.ts

@@ -106,6 +106,7 @@ export async function prompt({
     SILENT_MODE,
     OPENAI_API_ENDPOINT: apiEndpoint,
     MODEL: model,
+    ENABLE_THINKING: enableThinking,
   } = await getConfig();
   const skipCommandExplanation = silentMode || SILENT_MODE;
 
@@ -120,6 +121,7 @@ export async function prompt({
     key,
     model,
     apiEndpoint,
+    enableThinking,
   });
   spin.stop(`${i18n.t('Your script')}:`);
   console.log('');
@@ -136,6 +138,7 @@ export async function prompt({
         key,
         model,
         apiEndpoint,
+        enableThinking,
       });
       spin.stop(`${i18n.t('Explanation')}:`);
       console.log('');
@@ -146,15 +149,23 @@ export async function prompt({
     }
   }
 
-  await runOrReviseFlow(script, key, model, apiEndpoint, silentMode);
+  await runOrReviseFlow(
+    script,
+    key,
+    model,
+    apiEndpoint,
+    silentMode,
+    enableThinking
+  );
 }
 
 async function runOrReviseFlow(
   script: string,
   key: string,
   model: string,
   apiEndpoint: string,
-  silentMode?: boolean
+  silentMode?: boolean,
+  enableThinking?: boolean
 ) {
   const emptyScript = script.trim() === '';
 
@@ -191,7 +202,14 @@ async function runOrReviseFlow(
         label: '🔁 ' + i18n.t('Revise'),
         hint: i18n.t('Give feedback via prompt and get a new result'),
         value: async () => {
-          await revisionFlow(script, key, model, apiEndpoint, silentMode);
+          await revisionFlow(
+            script,
+            key,
+            model,
+            apiEndpoint,
+            silentMode,
+            enableThinking
+          );
         },
       },
       {
@@ -223,7 +241,8 @@ async function revisionFlow(
   key: string,
   model: string,
   apiEndpoint: string,
-  silentMode?: boolean
+  silentMode?: boolean,
+  enableThinking?: boolean
 ) {
   const revision = await promptForRevision();
   const spin = p.spinner();
@@ -234,6 +253,7 @@ async function revisionFlow(
     key,
     model,
     apiEndpoint,
+    enableThinking,
   });
   spin.stop(`${i18n.t(`Your new script`)}:`);
 
@@ -251,6 +271,7 @@ async function revisionFlow(
       key,
       model,
       apiEndpoint,
+      enableThinking,
     });
 
     infoSpin.stop(`${i18n.t('Explanation')}:`);
@@ -261,7 +282,14 @@ async function revisionFlow(
     console.log(dim('•'));
   }
 
-  await runOrReviseFlow(script, key, model, apiEndpoint, silentMode);
+  await runOrReviseFlow(
+    script,
+    key,
+    model,
+    apiEndpoint,
+    silentMode,
+    enableThinking
+  );
 }
 
 export const parseAssert = (name: string, condition: any, message: string) => {