guardlink 1.1.0 → 1.2.0

package/dist/analyze/llm.js

@@ -2,8 +2,11 @@
  * GuardLink Threat Reports — Lightweight LLM client using raw fetch.
  *
  * Supports:
- *   - Anthropic Messages API (claude-sonnet-4-5-20250929, etc.)
- *   - OpenAI-compatible Chat Completions (GPT-4o, DeepSeek, OpenRouter)
+ *   - Anthropic Messages API (claude-sonnet-4-6, claude-opus-4-6, etc.) with extended thinking + tool use
+ *   - OpenAI Responses API (gpt-5.2, o3, etc.) with web search, tools, structured output
+ *   - Google Gemini API (gemini-2.5-flash, gemini-3-pro, etc.) via OpenAI-compatible endpoint
+ *   - OpenAI-compatible Chat Completions (DeepSeek, OpenRouter, Ollama)
+ *   - DeepSeek reasoning mode (deepseek-reasoner)
  *
  * Zero dependencies — uses Node 20+ built-in fetch.
  *
@@ -18,51 +21,43 @@
  * @flows #llm-client -> External_LLM_APIs via fetch -- "HTTP POST with auth headers and prompt payload"
  * @flows External_LLM_APIs -> #llm-client via response -- "Streaming or complete response from LLM provider"
  */
+// ─── Defaults ────────────────────────────────────────────────────────
 const DEFAULT_MODELS = {
-    anthropic: 'claude-sonnet-4-5-20250929',
-    openai: 'gpt-4o',
-    openrouter: 'anthropic/claude-sonnet-4-5-20250929',
+    anthropic: 'claude-sonnet-4-6',
+    openai: 'gpt-5.2',
+    google: 'gemini-2.5-flash',
+    openrouter: 'anthropic/claude-sonnet-4-6',
     deepseek: 'deepseek-chat',
+    ollama: 'llama3.2',
 };
 const BASE_URLS = {
     anthropic: 'https://api.anthropic.com',
     openai: 'https://api.openai.com',
+    google: 'https://generativelanguage.googleapis.com/v1beta/openai',
     openrouter: 'https://openrouter.ai/api',
     deepseek: 'https://api.deepseek.com',
+    ollama: 'http://localhost:11434',
 };
+// ─── Auto-detect ─────────────────────────────────────────────────────
 /**
  * Auto-detect provider from environment variables.
  * Returns null if no API key found.
  */
 export function autoDetectConfig() {
-    // Priority: Anthropic > OpenAI > OpenRouter > DeepSeek
     if (process.env.ANTHROPIC_API_KEY) {
-        return {
-            provider: 'anthropic',
-            model: DEFAULT_MODELS.anthropic,
-            apiKey: process.env.ANTHROPIC_API_KEY,
-        };
+        return { provider: 'anthropic', model: DEFAULT_MODELS.anthropic, apiKey: process.env.ANTHROPIC_API_KEY };
     }
     if (process.env.OPENAI_API_KEY) {
-        return {
-            provider: 'openai',
-            model: DEFAULT_MODELS.openai,
-            apiKey: process.env.OPENAI_API_KEY,
-        };
+        return { provider: 'openai', model: DEFAULT_MODELS.openai, apiKey: process.env.OPENAI_API_KEY };
     }
     if (process.env.OPENROUTER_API_KEY) {
-        return {
-            provider: 'openrouter',
-            model: DEFAULT_MODELS.openrouter,
-            apiKey: process.env.OPENROUTER_API_KEY,
-        };
+        return { provider: 'openrouter', model: DEFAULT_MODELS.openrouter, apiKey: process.env.OPENROUTER_API_KEY };
+    }
+    if (process.env.GOOGLE_API_KEY || process.env.GEMINI_API_KEY) {
+        return { provider: 'google', model: DEFAULT_MODELS.google, apiKey: (process.env.GOOGLE_API_KEY || process.env.GEMINI_API_KEY) };
     }
     if (process.env.DEEPSEEK_API_KEY) {
-        return {
-            provider: 'deepseek',
-            model: DEFAULT_MODELS.deepseek,
-            apiKey: process.env.DEEPSEEK_API_KEY,
-        };
+        return { provider: 'deepseek', model: DEFAULT_MODELS.deepseek, apiKey: process.env.DEEPSEEK_API_KEY };
     }
     return null;
 }
@@ -70,13 +65,13 @@ export function autoDetectConfig() {
  * Build config from explicit flags + env vars.
  */
 export function buildConfig(opts) {
-    // If provider specified, use it
     if (opts.provider) {
         const provider = opts.provider;
         const envKeyMap = {
             anthropic: 'ANTHROPIC_API_KEY',
             openai: 'OPENAI_API_KEY',
             openrouter: 'OPENROUTER_API_KEY',
+            google: 'GOOGLE_API_KEY',
             deepseek: 'DEEPSEEK_API_KEY',
         };
         const apiKey = opts.apiKey || process.env[envKeyMap[provider] || ''];
@@ -84,58 +79,120 @@ export function buildConfig(opts) {
             return null;
         return {
             provider,
-            model: opts.model || DEFAULT_MODELS[provider] || 'gpt-4o',
+            model: opts.model || DEFAULT_MODELS[provider] || 'gpt-5.2',
             apiKey,
         };
     }
-    // Auto-detect
     const config = autoDetectConfig();
     if (!config)
         return null;
-    // Override model if specified
     if (opts.model)
         config.model = opts.model;
     return config;
 }
+// ─── Main entry point ────────────────────────────────────────────────
 /**
  * Send a message to the LLM and return the response.
+ * Supports streaming, tool use (agentic loop), extended thinking,
+ * web search, and structured output.
  */
 export async function chatCompletion(config, systemPrompt, userMessage, onChunk) {
     if (config.provider === 'anthropic') {
-        return callAnthropic(config, systemPrompt, userMessage, onChunk);
+        return callAnthropicWithTools(config, systemPrompt, userMessage, onChunk);
+    }
+    else if (config.provider === 'openai') {
+        return callOpenAIResponses(config, systemPrompt, userMessage, onChunk);
     }
     else {
+        // Google Gemini, DeepSeek, OpenRouter, Ollama all use OpenAI-compatible Chat Completions
         return callOpenAICompatible(config, systemPrompt, userMessage, onChunk);
     }
 }
-// ─── Anthropic Messages API ──────────────────────────────────────────
-async function callAnthropic(config, systemPrompt, userMessage, onChunk) {
+// ─── Anthropic Messages API (2025) ──────────────────────────────────
+const ANTHROPIC_API_VERSION = '2025-04-14';
+/** Wrapper with agentic tool-call loop */
+async function callAnthropicWithTools(config, systemPrompt, userMessage, onChunk) {
+    const maxRounds = config.maxToolRounds ?? 5;
+    let messages = [{ role: 'user', content: userMessage }];
+    const allToolCalls = [];
+    let finalResponse = null;
+    for (let round = 0; round <= maxRounds; round++) {
+        const response = await callAnthropic(config, systemPrompt, messages, round === 0 ? onChunk : undefined);
+        if (response.toolCalls?.length)
+            allToolCalls.push(...response.toolCalls);
+        if (!response.toolCalls?.length || !config.toolExecutor) {
+            finalResponse = response;
+            break;
+        }
+        // Add assistant response and tool results for next round
+        messages.push({ role: 'assistant', content: response._rawContent });
+        for (const tc of response.toolCalls) {
+            let resultText;
+            try {
+                resultText = await config.toolExecutor(tc.name, tc.arguments);
+            }
+            catch (err) {
+                resultText = `Error: ${err.message}`;
+            }
+            messages.push({
+                role: 'user',
+                content: [{ type: 'tool_result', tool_use_id: tc.id, content: resultText }],
+            });
+        }
+    }
+    if (!finalResponse)
+        throw new Error('Max tool call rounds exceeded');
+    finalResponse.toolCalls = allToolCalls.length ? allToolCalls : undefined;
+    return finalResponse;
+}
+async function callAnthropic(config, systemPrompt, messages, onChunk) {
     const baseUrl = config.baseUrl || BASE_URLS.anthropic;
     const maxTokens = config.maxTokens || 8192;
+    const headers = {
+        'Content-Type': 'application/json',
+        'x-api-key': config.apiKey,
+        'anthropic-version': ANTHROPIC_API_VERSION,
+    };
+    if (config.extendedThinking) {
+        headers['anthropic-beta'] = 'interleaved-thinking-2025-05-14';
+    }
+    const body = {
+        model: config.model,
+        max_tokens: maxTokens,
+        system: systemPrompt,
+        messages,
+    };
+    if (config.extendedThinking) {
+        body.thinking = { type: 'enabled', budget_tokens: config.thinkingBudget || 10000 };
+    }
+    if (config.tools?.length) {
+        body.tools = config.tools.map(t => ({
+            name: t.name,
+            description: t.description,
+            input_schema: {
+                type: 'object',
+                properties: t.parameters.properties,
+                required: t.parameters.required,
+            },
+        }));
+    }
     if (onChunk) {
-        // Streaming
+        body.stream = true;
         const res = await fetch(`${baseUrl}/v1/messages`, {
-            method: 'POST',
-            headers: {
-                'Content-Type': 'application/json',
-                'x-api-key': config.apiKey,
-                'anthropic-version': '2023-06-01',
-            },
-            body: JSON.stringify({
-                model: config.model,
-                max_tokens: maxTokens,
-                system: systemPrompt,
-                stream: true,
-                messages: [{ role: 'user', content: userMessage }],
-            }),
+            method: 'POST', headers, body: JSON.stringify(body),
         });
         if (!res.ok) {
             const err = await res.text();
             throw new Error(`Anthropic API error ${res.status}: ${err}`);
         }
         let content = '';
+        let thinking = '';
         let inputTokens = 0;
         let outputTokens = 0;
+        const toolCalls = [];
+        let curToolId = '';
+        let curToolName = '';
+        let curToolArgs = '';
         const reader = res.body?.getReader();
         if (!reader)
             throw new Error('No response body');
@@ -155,52 +212,313 @@ async function callAnthropic(config, systemPrompt, userMessage, onChunk) {
                 if (data === '[DONE]')
                     continue;
                 try {
-                    const event = JSON.parse(data);
-                    if (event.type === 'content_block_delta' && event.delta?.text) {
-                        content += event.delta.text;
-                        onChunk(event.delta.text);
+                    const ev = JSON.parse(data);
+                    if (ev.type === 'content_block_start' && ev.content_block?.type === 'tool_use') {
+                        curToolId = ev.content_block.id || '';
+                        curToolName = ev.content_block.name || '';
+                        curToolArgs = '';
                     }
-                    if (event.type === 'message_delta' && event.usage) {
-                        outputTokens = event.usage.output_tokens || 0;
+                    if (ev.type === 'content_block_delta') {
+                        if (ev.delta?.type === 'text_delta' && ev.delta?.text) {
+                            content += ev.delta.text;
+                            onChunk(ev.delta.text);
+                        }
+                        if (ev.delta?.type === 'thinking_delta' && ev.delta?.thinking) {
+                            thinking += ev.delta.thinking;
+                        }
+                        if (ev.delta?.type === 'input_json_delta' && ev.delta?.partial_json) {
+                            curToolArgs += ev.delta.partial_json;
+                        }
                     }
-                    if (event.type === 'message_start' && event.message?.usage) {
-                        inputTokens = event.message.usage.input_tokens || 0;
+                    if (ev.type === 'content_block_stop' && curToolId) {
+                        try {
+                            toolCalls.push({ id: curToolId, name: curToolName, arguments: JSON.parse(curToolArgs || '{}') });
+                        }
+                        catch { /* skip */ }
+                        curToolId = '';
+                        curToolName = '';
+                        curToolArgs = '';
                     }
+                    if (ev.type === 'message_delta' && ev.usage)
+                        outputTokens = ev.usage.output_tokens || 0;
+                    if (ev.type === 'message_start' && ev.message?.usage)
+                        inputTokens = ev.message.usage.input_tokens || 0;
                 }
-                catch { /* skip non-JSON lines */ }
+                catch { /* skip */ }
             }
         }
-        return { content, model: config.model, inputTokens, outputTokens };
+        return {
+            content, model: config.model, inputTokens, outputTokens,
+            thinking: thinking || undefined, thinkingTokens: undefined,
+            toolCalls: toolCalls.length ? toolCalls : undefined,
+            _rawContent: buildRawContent(content, thinking, toolCalls),
+        };
     }
     else {
         // Non-streaming
         const res = await fetch(`${baseUrl}/v1/messages`, {
-            method: 'POST',
-            headers: {
-                'Content-Type': 'application/json',
-                'x-api-key': config.apiKey,
-                'anthropic-version': '2023-06-01',
-            },
-            body: JSON.stringify({
-                model: config.model,
-                max_tokens: maxTokens,
-                system: systemPrompt,
-                messages: [{ role: 'user', content: userMessage }],
-            }),
+            method: 'POST', headers, body: JSON.stringify(body),
         });
         if (!res.ok) {
             const err = await res.text();
             throw new Error(`Anthropic API error ${res.status}: ${err}`);
         }
         const data = await res.json();
+        let content = '';
+        let thinking = '';
+        const toolCalls = [];
+        for (const block of (data.content || [])) {
+            if (block.type === 'text')
+                content += block.text;
+            if (block.type === 'thinking')
+                thinking += block.thinking;
+            if (block.type === 'tool_use') {
+                toolCalls.push({ id: block.id, name: block.name, arguments: block.input || {} });
+            }
+        }
         return {
-            content: data.content?.[0]?.text || '',
-            model: data.model || config.model,
+            content, model: data.model || config.model,
             inputTokens: data.usage?.input_tokens,
             outputTokens: data.usage?.output_tokens,
+            thinking: thinking || undefined,
+            toolCalls: toolCalls.length ? toolCalls : undefined,
+            _rawContent: data.content,
         };
     }
 }
+function buildRawContent(content, thinking, toolCalls) {
+    const blocks = [];
+    if (thinking)
+        blocks.push({ type: 'thinking', thinking });
+    if (content)
+        blocks.push({ type: 'text', text: content });
+    for (const tc of toolCalls)
+        blocks.push({ type: 'tool_use', id: tc.id, name: tc.name, input: tc.arguments });
+    return blocks;
+}
+// ─── OpenAI Responses API ────────────────────────────────────────────
+async function callOpenAIResponses(config, systemPrompt, userMessage, onChunk) {
+    const baseUrl = config.baseUrl || BASE_URLS.openai;
+    const maxTokens = config.maxTokens || 8192;
+    const headers = {
+        'Content-Type': 'application/json',
+        'Authorization': `Bearer ${config.apiKey}`,
+    };
+    const input = [
+        { role: 'developer', content: systemPrompt },
+        { role: 'user', content: userMessage },
+    ];
+    const tools = [];
+    if (config.webSearch)
+        tools.push({ type: 'web_search' });
+    if (config.tools?.length) {
+        for (const t of config.tools) {
+            tools.push({
+                type: 'function', name: t.name, description: t.description,
+                parameters: t.parameters, strict: true,
+            });
+        }
+    }
+    const body = { model: config.model, input, max_output_tokens: maxTokens };
+    if (tools.length)
+        body.tools = tools;
+    if (config.responseFormat === 'json')
+        body.text = { format: { type: 'json_object' } };
+    if (onChunk) {
+        body.stream = true;
+        const res = await fetch(`${baseUrl}/v1/responses`, {
+            method: 'POST', headers, body: JSON.stringify(body),
+        });
+        if (!res.ok) {
+            const err = await res.text();
+            // Fallback to Chat Completions if Responses API not available
+            if (res.status === 404)
+                return callOpenAICompatible(config, systemPrompt, userMessage, onChunk);
+            throw new Error(`OpenAI API error ${res.status}: ${err}`);
+        }
+        let content = '';
+        let inputTokens = 0;
+        let outputTokens = 0;
+        const toolCalls = [];
+        const reader = res.body?.getReader();
+        if (!reader)
+            throw new Error('No response body');
+        const decoder = new TextDecoder();
+        let buffer = '';
+        while (true) {
+            const { done, value } = await reader.read();
+            if (done)
+                break;
+            buffer += decoder.decode(value, { stream: true });
+            const lines = buffer.split('\n');
+            buffer = lines.pop() || '';
+            for (const line of lines) {
+                if (!line.startsWith('data: '))
+                    continue;
+                const d = line.slice(6).trim();
+                if (d === '[DONE]')
+                    continue;
+                try {
+                    const ev = JSON.parse(d);
+                    if (ev.type === 'response.output_text.delta' && ev.delta) {
+                        content += ev.delta;
+                        onChunk(ev.delta);
+                    }
+                    if (ev.type === 'response.function_call_arguments.done') {
+                        try {
+                            toolCalls.push({ id: ev.call_id || '', name: ev.name || '', arguments: JSON.parse(ev.arguments || '{}') });
+                        }
+                        catch { /* skip */ }
+                    }
+                    if (ev.type === 'response.completed' && ev.response?.usage) {
+                        inputTokens = ev.response.usage.input_tokens || 0;
+                        outputTokens = ev.response.usage.output_tokens || 0;
+                    }
+                }
+                catch { /* skip */ }
+            }
+        }
+        if (toolCalls.length && config.toolExecutor) {
+            return handleOpenAIToolLoop(config, baseUrl, headers, body, content, toolCalls, inputTokens, outputTokens, onChunk);
+        }
+        return { content, model: config.model, inputTokens, outputTokens, toolCalls: toolCalls.length ? toolCalls : undefined };
+    }
+    else {
+        // Non-streaming
+        const res = await fetch(`${baseUrl}/v1/responses`, {
+            method: 'POST', headers, body: JSON.stringify(body),
+        });
+        if (!res.ok) {
+            const err = await res.text();
+            if (res.status === 404)
+                return callOpenAICompatible(config, systemPrompt, userMessage, undefined);
+            throw new Error(`OpenAI API error ${res.status}: ${err}`);
+        }
+        const data = await res.json();
+        let content = '';
+        const toolCalls = [];
+        for (const item of (data.output || [])) {
+            if (item.type === 'message') {
+                for (const part of (item.content || [])) {
+                    if (part.type === 'output_text')
+                        content += part.text;
+                }
+            }
+            if (item.type === 'function_call') {
+                try {
+                    toolCalls.push({ id: item.call_id || item.id || '', name: item.name || '', arguments: JSON.parse(item.arguments || '{}') });
+                }
+                catch { /* skip */ }
+            }
+        }
+        if (!content && data.output_text)
+            content = data.output_text;
+        if (toolCalls.length && config.toolExecutor) {
+            return handleOpenAIToolLoop(config, baseUrl, headers, body, content, toolCalls, data.usage?.input_tokens, data.usage?.output_tokens, undefined);
+        }
+        return {
+            content, model: data.model || config.model,
+            inputTokens: data.usage?.input_tokens, outputTokens: data.usage?.output_tokens,
+            toolCalls: toolCalls.length ? toolCalls : undefined,
+        };
+    }
+}
+/** Agentic tool-call loop for OpenAI Responses API */
+async function handleOpenAIToolLoop(config, baseUrl, headers, origBody, partialContent, pending, inTok, outTok, onChunk) {
+    const maxRounds = config.maxToolRounds ?? 5;
+    const all = [...pending];
+    let content = partialContent;
+    let inputTokens = inTok;
+    let outputTokens = outTok;
+    for (let round = 0; round < maxRounds && pending.length; round++) {
+        const results = [];
+        for (const tc of pending) {
+            let r;
+            try {
+                r = await config.toolExecutor(tc.name, tc.arguments);
+            }
+            catch (e) {
+                r = `Error: ${e.message}`;
+            }
+            results.push({ type: 'function_call_output', call_id: tc.id, output: r });
+        }
+        const followUp = { ...origBody, input: results, stream: !!onChunk };
+        const res = await fetch(`${baseUrl}/v1/responses`, { method: 'POST', headers, body: JSON.stringify(followUp) });
+        if (!res.ok) {
+            const err = await res.text();
+            throw new Error(`OpenAI tool follow-up error ${res.status}: ${err}`);
+        }
+        pending = [];
+        if (onChunk) {
+            const reader = res.body?.getReader();
+            if (!reader)
+                throw new Error('No response body');
+            const dec = new TextDecoder();
+            let buf = '';
+            while (true) {
+                const { done, value } = await reader.read();
+                if (done)
+                    break;
+                buf += dec.decode(value, { stream: true });
+                const lines = buf.split('\n');
+                buf = lines.pop() || '';
+                for (const ln of lines) {
+                    if (!ln.startsWith('data: '))
+                        continue;
+                    const d = ln.slice(6).trim();
+                    if (d === '[DONE]')
+                        continue;
+                    try {
+                        const ev = JSON.parse(d);
+                        if (ev.type === 'response.output_text.delta' && ev.delta) {
+                            content += ev.delta;
+                            onChunk(ev.delta);
+                        }
+                        if (ev.type === 'response.function_call_arguments.done') {
+                            try {
+                                const tc = { id: ev.call_id || '', name: ev.name || '', arguments: JSON.parse(ev.arguments || '{}') };
+                                pending.push(tc);
+                                all.push(tc);
+                            }
+                            catch { /* skip */ }
+                        }
+                        if (ev.type === 'response.completed' && ev.response?.usage) {
+                            inputTokens = (inputTokens || 0) + (ev.response.usage.input_tokens || 0);
+                            outputTokens = (outputTokens || 0) + (ev.response.usage.output_tokens || 0);
+                        }
+                    }
+                    catch { /* skip */ }
+                }
+            }
+        }
+        else {
+            const data = await res.json();
+            for (const item of (data.output || [])) {
+                if (item.type === 'message') {
+                    for (const p of (item.content || [])) {
+                        if (p.type === 'output_text')
+                            content += p.text;
+                    }
+                }
+                if (item.type === 'function_call') {
+                    try {
+                        const tc = { id: item.call_id || item.id || '', name: item.name || '', arguments: JSON.parse(item.arguments || '{}') };
+                        pending.push(tc);
+                        all.push(tc);
+                    }
+                    catch { /* skip */ }
+                }
+            }
+            if (data.output_text && !content)
+                content = data.output_text;
+            if (data.usage) {
+                inputTokens = (inputTokens || 0) + (data.usage.input_tokens || 0);
+                outputTokens = (outputTokens || 0) + (data.usage.output_tokens || 0);
+            }
+        }
+    }
+    return { content, model: config.model, inputTokens, outputTokens, toolCalls: all.length ? all : undefined };
+}
 // ─── OpenAI-compatible Chat Completions ──────────────────────────────
 async function callOpenAICompatible(config, systemPrompt, userMessage, onChunk) {
     const baseUrl = config.baseUrl || BASE_URLS[config.provider] || BASE_URLS.openai;
@@ -209,31 +527,39 @@ async function callOpenAICompatible(config, systemPrompt, userMessage, onChunk)
         'Content-Type': 'application/json',
         'Authorization': `Bearer ${config.apiKey}`,
     };
-    // OpenRouter requires extra headers
     if (config.provider === 'openrouter') {
         headers['HTTP-Referer'] = 'https://guardlink.bugb.io';
         headers['X-Title'] = 'GuardLink CLI';
     }
+    const isDeepSeekReasoner = config.provider === 'deepseek' && config.model.includes('reasoner');
+    const body = {
+        model: config.model,
+        max_tokens: maxTokens,
+        messages: [
+            { role: 'system', content: systemPrompt },
+            { role: 'user', content: userMessage },
+        ],
+    };
+    if (config.responseFormat === 'json') {
+        body.response_format = { type: 'json_object' };
+    }
+    if (config.tools?.length) {
+        body.tools = config.tools.map(t => ({
+            type: 'function',
+            function: { name: t.name, description: t.description, parameters: t.parameters },
+        }));
+    }
     if (onChunk) {
-        // Streaming
+        body.stream = true;
         const res = await fetch(`${baseUrl}/v1/chat/completions`, {
-            method: 'POST',
-            headers,
-            body: JSON.stringify({
-                model: config.model,
-                max_tokens: maxTokens,
-                stream: true,
-                messages: [
-                    { role: 'system', content: systemPrompt },
-                    { role: 'user', content: userMessage },
-                ],
-            }),
+            method: 'POST', headers, body: JSON.stringify(body),
         });
         if (!res.ok) {
             const err = await res.text();
             throw new Error(`${config.provider} API error ${res.status}: ${err}`);
         }
         let content = '';
+        let reasoning = '';
         const reader = res.body?.getReader();
         if (!reader)
             throw new Error('No response body');
@@ -259,36 +585,31 @@ async function callOpenAICompatible(config, systemPrompt, userMessage, onChunk)
                         content += delta;
                         onChunk(delta);
                     }
+                    const reasoningDelta = event.choices?.[0]?.delta?.reasoning_content;
+                    if (reasoningDelta)
+                        reasoning += reasoningDelta;
                 }
                 catch { /* skip */ }
             }
         }
-        return { content, model: config.model };
+        return { content, model: config.model, thinking: reasoning || undefined };
     }
     else {
-        // Non-streaming
         const res = await fetch(`${baseUrl}/v1/chat/completions`, {
-            method: 'POST',
-            headers,
-            body: JSON.stringify({
-                model: config.model,
-                max_tokens: maxTokens,
-                messages: [
-                    { role: 'system', content: systemPrompt },
-                    { role: 'user', content: userMessage },
-                ],
-            }),
+            method: 'POST', headers, body: JSON.stringify(body),
         });
         if (!res.ok) {
             const err = await res.text();
             throw new Error(`${config.provider} API error ${res.status}: ${err}`);
         }
         const data = await res.json();
+        const choice = data.choices?.[0];
         return {
-            content: data.choices?.[0]?.message?.content || '',
+            content: choice?.message?.content || '',
             model: data.model || config.model,
             inputTokens: data.usage?.prompt_tokens,
             outputTokens: data.usage?.completion_tokens,
+            thinking: isDeepSeekReasoner ? (choice?.message?.reasoning_content || undefined) : undefined,
         };
     }
 }