trickle-observe 0.2.119 → 0.2.120

package/dist/llm-observer.d.ts

@@ -11,6 +11,7 @@
  */
 export declare function patchOpenAI(openaiModule: any, debug: boolean): void;
 export declare function patchAnthropic(anthropicModule: any, debug: boolean): void;
+export declare function patchGemini(geminiModule: any, debug: boolean): void;
 /**
  * Initialize the LLM observer — clears previous data file.
  */

package/dist/llm-observer.js

@@ -46,6 +46,7 @@ var __importStar = (this && this.__importStar) || (function () {
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.patchOpenAI = patchOpenAI;
 exports.patchAnthropic = patchAnthropic;
+exports.patchGemini = patchGemini;
 exports.initLlmObserver = initLlmObserver;
 const fs = __importStar(require("fs"));
 const path = __importStar(require("path"));
@@ -65,6 +66,12 @@ const PRICING = {
     'claude-3-5-sonnet-20241022': { input: 3, output: 15 },
     'claude-3-5-haiku-20241022': { input: 0.8, output: 4 },
     'claude-3-haiku-20240307': { input: 0.25, output: 1.25 },
+    'gemini-2.5-flash-lite': { input: 0.1, output: 0.4 },
+    'gemini-2.5-flash': { input: 0.3, output: 2.5 },
+    'gemini-2.5-pro': { input: 1.25, output: 10 },
+    'gemini-2.0-flash': { input: 0.1, output: 0.4 },
+    'gemini-1.5-flash': { input: 0.075, output: 0.3 },
+    'gemini-1.5-pro': { input: 1.25, output: 5 },
 };
 function getLlmFile() {
     if (llmFile)
@@ -489,6 +496,223 @@ function captureAnthropicError(params, err, startTime, debug) {
     });
 }
 // ────────────────────────────────────────────────────
+// Google Gemini SDK instrumentation (@google/genai)
+// ────────────────────────────────────────────────────
+function patchGemini(geminiModule, debug) {
+    if (!geminiModule || getattr(geminiModule, '_trickle_llm_patched'))
+        return;
+    setattr(geminiModule, '_trickle_llm_patched', true);
+    // @google/genai exports GoogleGenAI class
+    // Usage: const ai = new GoogleGenAI({ apiKey }); ai.models.generateContent({...})
+    const GoogleGenAI = geminiModule.GoogleGenAI || geminiModule.default?.GoogleGenAI;
+    if (typeof GoogleGenAI !== 'function') {
+        if (debug)
+            console.log('[trickle/llm] Gemini: GoogleGenAI class not found');
+        return;
+    }
+    try {
+        // GoogleGenAI creates models as own property in the constructor.
+        // Patch the GoogleGenAI constructor to wrap generateContent after creation.
+        const origGoogleGenAIInit = GoogleGenAI.prototype.constructor;
+        // Use a post-construction hook: after new GoogleGenAI() creates the instance
+        // with models.generateContent as an own property, wrap that method.
+        const tmpClient = new GoogleGenAI({ apiKey: 'trickle-probe' });
+        const ModelsClass = Object.getPrototypeOf(tmpClient.models)?.constructor;
+        if (ModelsClass) {
+            const origModelsInit = ModelsClass;
+            // Patch the Models constructor to wrap generateContent after instance creation
+            const origConstruct = ModelsClass.prototype.constructor;
+            // We can't replace the ES6 class constructor, so instead we use a
+            // post-construction approach: hook into GoogleGenAI's prototype to
+            // patch models on each new client instance.
+            const origGAIProto = GoogleGenAI.prototype;
+            const origInitDescriptors = Object.getOwnPropertyDescriptors(origGAIProto);
+            // Define a lazy wrapper: first time models.generateContent is called,
+            // install the instrumentation wrapper
+            function wrapModelsInstance(models) {
+                if (!models || models.__trickle_patched)
+                    return;
+                models.__trickle_patched = true;
+                if (typeof models.generateContent === 'function') {
+                    const origGenerate = models.generateContent.bind(models);
+                    models.generateContent = function patchedGenerateContent(...args) {
+                        const params = args[0] || {};
+                        const startTime = performance.now();
+                        const result = origGenerate(...args);
+                        if (result && typeof result.then === 'function') {
+                            return result.then((response) => {
+                                captureGeminiResponse(params, response, startTime, false, debug);
+                                return response;
+                            }).catch((err) => {
+                                captureGeminiError(params, err, startTime, debug);
+                                throw err;
+                            });
+                        }
+                        return result;
+                    };
+                }
+                if (typeof models.generateContentStream === 'function') {
+                    const origStream = models.generateContentStream.bind(models);
+                    models.generateContentStream = function patchedStream(...args) {
+                        const params = args[0] || {};
+                        const startTime = performance.now();
+                        const result = origStream(...args);
+                        if (result && typeof result.then === 'function') {
+                            return result.then((stream) => handleGeminiStream(stream, params, startTime, debug));
+                        }
+                        return result;
+                    };
+                }
+            }
+            // Intercept the GoogleGenAI constructor to patch each instance's models
+            // Since we can't replace ES6 class constructors, we use a Proxy
+            const proxyHandler = {
+                construct(target, args, newTarget) {
+                    const instance = Reflect.construct(target, args, newTarget);
+                    if (instance.models)
+                        wrapModelsInstance(instance.models);
+                    return instance;
+                }
+            };
+            const ProxiedGoogleGenAI = new Proxy(GoogleGenAI, proxyHandler);
+            // Replace on the module (try both export styles)
+            try {
+                geminiModule.GoogleGenAI = ProxiedGoogleGenAI;
+            }
+            catch { }
+            try {
+                if (geminiModule.default?.GoogleGenAI)
+                    geminiModule.default.GoogleGenAI = ProxiedGoogleGenAI;
+            }
+            catch { }
+            // Also patch the already-created probe instance (in case someone imported before us)
+            // This is a no-op since the probe is discarded.
+            if (debug)
+                console.log('[trickle/llm] Patched Gemini SDK');
+        }
+    }
+    catch (e) {
+        if (debug)
+            console.log('[trickle/llm] Gemini patch probe failed:', e.message);
+    }
+}
+function captureGeminiResponse(params, response, startTime, isStream, debug) {
+    const usage = response.usageMetadata || {};
+    const model = params.model || 'gemini-unknown';
+    const inputTokens = usage.promptTokenCount || 0;
+    const outputTokens = usage.candidatesTokenCount || 0;
+    const totalTokens = usage.totalTokenCount || inputTokens + outputTokens;
+    let outputText = '';
+    let finishReason = 'unknown';
+    try {
+        outputText = response.text || '';
+    }
+    catch {
+        const candidates = response.candidates || [];
+        if (candidates[0]?.content?.parts?.[0]?.text) {
+            outputText = candidates[0].content.parts[0].text;
+        }
+    }
+    const candidates = response.candidates || [];
+    if (candidates[0]?.finishReason)
+        finishReason = candidates[0].finishReason;
+    // Extract input preview from contents
+    let inputPreview = '';
+    const contents = params.contents;
+    if (typeof contents === 'string') {
+        inputPreview = truncate(contents);
+    }
+    else if (Array.isArray(contents)) {
+        const last = contents[contents.length - 1];
+        if (typeof last === 'string')
+            inputPreview = truncate(last);
+        else if (last?.parts?.[0]?.text)
+            inputPreview = truncate(last.parts[0].text);
+    }
+    const event = {
+        kind: 'llm_call', provider: 'gemini', model,
+        durationMs: round(performance.now() - startTime),
+        inputTokens, outputTokens, totalTokens,
+        estimatedCostUsd: estimateCost(model, inputTokens, outputTokens),
+        stream: isStream, finishReason,
+        temperature: params.config?.temperature,
+        maxTokens: params.config?.maxOutputTokens,
+        systemPrompt: typeof params.config?.systemInstruction === 'string'
+            ? truncate(params.config.systemInstruction, 200) : undefined,
+        inputPreview, outputPreview: truncate(outputText),
+        messageCount: Array.isArray(contents) ? contents.length : (contents ? 1 : 0),
+        toolUse: !!(params.config?.tools?.length || params.tools?.length),
+        timestamp: Date.now(),
+    };
+    writeLlmEvent(event);
+    if (debug)
+        console.log(`[trickle/llm] Gemini: ${model} (${totalTokens} tokens, ${event.durationMs}ms)`);
+}
+function captureGeminiError(params, err, startTime, debug) {
+    const model = params.model || 'gemini-unknown';
+    writeLlmEvent({
+        kind: 'llm_call', provider: 'gemini', model,
+        durationMs: round(performance.now() - startTime),
+        inputTokens: 0, outputTokens: 0, totalTokens: 0, estimatedCostUsd: 0,
+        stream: false, finishReason: 'error',
+        temperature: params.config?.temperature,
+        maxTokens: params.config?.maxOutputTokens,
+        inputPreview: typeof params.contents === 'string' ? truncate(params.contents) : '',
+        outputPreview: '', messageCount: 0,
+        toolUse: false, timestamp: Date.now(),
+        error: truncate(err?.message || String(err), 200),
+    });
+}
+async function handleGeminiStream(stream, params, startTime, debug) {
+    if (!stream || !stream[Symbol.asyncIterator])
+        return stream;
+    const chunks = [];
+    const origIterator = stream[Symbol.asyncIterator].bind(stream);
+    let lastUsage = null;
+    stream[Symbol.asyncIterator] = function () {
+        const iter = origIterator();
+        return {
+            async next() {
+                const result = await iter.next();
+                if (!result.done) {
+                    const chunk = result.value;
+                    try {
+                        if (chunk.text)
+                            chunks.push(chunk.text);
+                    }
+                    catch { }
+                    if (chunk.usageMetadata)
+                        lastUsage = chunk.usageMetadata;
+                }
+                else {
+                    // Stream finished
+                    const model = params.model || 'gemini-unknown';
+                    const inputTokens = lastUsage?.promptTokenCount || 0;
+                    const outputTokens = lastUsage?.candidatesTokenCount || 0;
+                    writeLlmEvent({
+                        kind: 'llm_call', provider: 'gemini', model,
+                        durationMs: round(performance.now() - startTime),
+                        inputTokens, outputTokens, totalTokens: inputTokens + outputTokens,
+                        estimatedCostUsd: estimateCost(model, inputTokens, outputTokens),
+                        stream: true, finishReason: 'stop',
+                        temperature: params.config?.temperature,
+                        maxTokens: params.config?.maxOutputTokens,
+                        inputPreview: typeof params.contents === 'string' ? truncate(params.contents) : '',
+                        outputPreview: truncate(chunks.join('')),
+                        messageCount: 0, toolUse: false, timestamp: Date.now(),
+                    });
+                    if (debug)
+                        console.log(`[trickle/llm] Gemini stream: ${model} (${outputTokens} tokens)`);
+                }
+                return result;
+            },
+            return: iter.return?.bind(iter),
+            throw: iter.throw?.bind(iter),
+        };
+    };
+    return stream;
+}
+// ────────────────────────────────────────────────────
 // Helpers
 // ────────────────────────────────────────────────────
 function round(n) {

package/dist/observe-register.js

@@ -1552,6 +1552,15 @@ if (enabled) {
             }
             catch { /* not critical */ }
         }
+        // Google Gemini SDK
+        if (request === '@google/genai' && !expressPatched.has('@google/genai')) {
+            expressPatched.add('@google/genai');
+            try {
+                const { patchGemini } = require(path_1.default.join(__dirname, 'llm-observer.js'));
+                patchGemini(exports, debug);
+            }
+            catch { /* not critical */ }
+        }
         // Resolve to absolute path for dedup — do this FIRST since bundlers like
         // tsx/esbuild may use path aliases (e.g., @config/env) that don't start
         // with './' or '/'. We need the resolved path to decide if it's user code.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "trickle-observe",
-  "version": "0.2.119",
+  "version": "0.2.120",
   "description": "Runtime type observability for JavaScript applications",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",

package/src/llm-observer.ts

@@ -30,6 +30,12 @@ const PRICING: Record<string, { input: number; output: number }> = {
   'claude-3-5-sonnet-20241022': { input: 3, output: 15 },
   'claude-3-5-haiku-20241022': { input: 0.8, output: 4 },
   'claude-3-haiku-20240307': { input: 0.25, output: 1.25 },
+  'gemini-2.5-flash-lite': { input: 0.1, output: 0.4 },
+  'gemini-2.5-flash': { input: 0.3, output: 2.5 },
+  'gemini-2.5-pro': { input: 1.25, output: 10 },
+  'gemini-2.0-flash': { input: 0.1, output: 0.4 },
+  'gemini-1.5-flash': { input: 0.075, output: 0.3 },
+  'gemini-1.5-pro': { input: 1.25, output: 5 },
 };
 
 function getLlmFile(): string {
@@ -480,6 +486,217 @@ function captureAnthropicError(params: any, err: any, startTime: number, debug:
   });
 }
 
+// ────────────────────────────────────────────────────
+// Google Gemini SDK instrumentation (@google/genai)
+// ────────────────────────────────────────────────────
+
+export function patchGemini(geminiModule: any, debug: boolean): void {
+  if (!geminiModule || getattr(geminiModule, '_trickle_llm_patched')) return;
+  setattr(geminiModule, '_trickle_llm_patched', true);
+
+  // @google/genai exports GoogleGenAI class
+  // Usage: const ai = new GoogleGenAI({ apiKey }); ai.models.generateContent({...})
+  const GoogleGenAI = geminiModule.GoogleGenAI || geminiModule.default?.GoogleGenAI;
+  if (typeof GoogleGenAI !== 'function') {
+    if (debug) console.log('[trickle/llm] Gemini: GoogleGenAI class not found');
+    return;
+  }
+
+  try {
+    // GoogleGenAI creates models as own property in the constructor.
+    // Patch the GoogleGenAI constructor to wrap generateContent after creation.
+    const origGoogleGenAIInit = GoogleGenAI.prototype.constructor;
+
+    // Use a post-construction hook: after new GoogleGenAI() creates the instance
+    // with models.generateContent as an own property, wrap that method.
+    const tmpClient = new GoogleGenAI({ apiKey: 'trickle-probe' });
+    const ModelsClass = Object.getPrototypeOf(tmpClient.models)?.constructor;
+
+    if (ModelsClass) {
+      const origModelsInit = ModelsClass;
+      // Patch the Models constructor to wrap generateContent after instance creation
+      const origConstruct = ModelsClass.prototype.constructor;
+
+      // We can't replace the ES6 class constructor, so instead we use a
+      // post-construction approach: hook into GoogleGenAI's prototype to
+      // patch models on each new client instance.
+      const origGAIProto = GoogleGenAI.prototype;
+      const origInitDescriptors = Object.getOwnPropertyDescriptors(origGAIProto);
+
+      // Define a lazy wrapper: first time models.generateContent is called,
+      // install the instrumentation wrapper
+      function wrapModelsInstance(models: any): void {
+        if (!models || models.__trickle_patched) return;
+        models.__trickle_patched = true;
+
+        if (typeof models.generateContent === 'function') {
+          const origGenerate = models.generateContent.bind(models);
+          models.generateContent = function patchedGenerateContent(...args: any[]) {
+            const params = args[0] || {};
+            const startTime = performance.now();
+            const result = origGenerate(...args);
+            if (result && typeof result.then === 'function') {
+              return result.then((response: any) => {
+                captureGeminiResponse(params, response, startTime, false, debug);
+                return response;
+              }).catch((err: any) => {
+                captureGeminiError(params, err, startTime, debug);
+                throw err;
+              });
+            }
+            return result;
+          };
+        }
+
+        if (typeof models.generateContentStream === 'function') {
+          const origStream = models.generateContentStream.bind(models);
+          models.generateContentStream = function patchedStream(...args: any[]) {
+            const params = args[0] || {};
+            const startTime = performance.now();
+            const result = origStream(...args);
+            if (result && typeof result.then === 'function') {
+              return result.then((stream: any) => handleGeminiStream(stream, params, startTime, debug));
+            }
+            return result;
+          };
+        }
+      }
+
+      // Intercept the GoogleGenAI constructor to patch each instance's models
+      // Since we can't replace ES6 class constructors, we use a Proxy
+      const proxyHandler: ProxyHandler<any> = {
+        construct(target: any, args: any[], newTarget: any): object {
+          const instance = Reflect.construct(target, args, newTarget) as any;
+          if (instance.models) wrapModelsInstance(instance.models);
+          return instance as object;
+        }
+      };
+      const ProxiedGoogleGenAI = new Proxy(GoogleGenAI, proxyHandler);
+
+      // Replace on the module (try both export styles)
+      try { geminiModule.GoogleGenAI = ProxiedGoogleGenAI; } catch {}
+      try { if (geminiModule.default?.GoogleGenAI) geminiModule.default.GoogleGenAI = ProxiedGoogleGenAI; } catch {}
+
+      // Also patch the already-created probe instance (in case someone imported before us)
+      // This is a no-op since the probe is discarded.
+
+      if (debug) console.log('[trickle/llm] Patched Gemini SDK');
+    }
+  } catch (e: any) {
+    if (debug) console.log('[trickle/llm] Gemini patch probe failed:', e.message);
+  }
+}
+
+function captureGeminiResponse(params: any, response: any, startTime: number, isStream: boolean, debug: boolean): void {
+  const usage = response.usageMetadata || {};
+  const model = params.model || 'gemini-unknown';
+  const inputTokens = usage.promptTokenCount || 0;
+  const outputTokens = usage.candidatesTokenCount || 0;
+  const totalTokens = usage.totalTokenCount || inputTokens + outputTokens;
+
+  let outputText = '';
+  let finishReason = 'unknown';
+  try {
+    outputText = response.text || '';
+  } catch {
+    const candidates = response.candidates || [];
+    if (candidates[0]?.content?.parts?.[0]?.text) {
+      outputText = candidates[0].content.parts[0].text;
+    }
+  }
+  const candidates = response.candidates || [];
+  if (candidates[0]?.finishReason) finishReason = candidates[0].finishReason;
+
+  // Extract input preview from contents
+  let inputPreview = '';
+  const contents = params.contents;
+  if (typeof contents === 'string') {
+    inputPreview = truncate(contents);
+  } else if (Array.isArray(contents)) {
+    const last = contents[contents.length - 1];
+    if (typeof last === 'string') inputPreview = truncate(last);
+    else if (last?.parts?.[0]?.text) inputPreview = truncate(last.parts[0].text);
+  }
+
+  const event: LlmEvent = {
+    kind: 'llm_call', provider: 'gemini', model,
+    durationMs: round(performance.now() - startTime),
+    inputTokens, outputTokens, totalTokens,
+    estimatedCostUsd: estimateCost(model, inputTokens, outputTokens),
+    stream: isStream, finishReason,
+    temperature: params.config?.temperature,
+    maxTokens: params.config?.maxOutputTokens,
+    systemPrompt: typeof params.config?.systemInstruction === 'string'
+      ? truncate(params.config.systemInstruction, 200) : undefined,
+    inputPreview, outputPreview: truncate(outputText),
+    messageCount: Array.isArray(contents) ? contents.length : (contents ? 1 : 0),
+    toolUse: !!(params.config?.tools?.length || params.tools?.length),
+    timestamp: Date.now(),
+  };
+  writeLlmEvent(event);
+  if (debug) console.log(`[trickle/llm] Gemini: ${model} (${totalTokens} tokens, ${event.durationMs}ms)`);
+}
+
+function captureGeminiError(params: any, err: any, startTime: number, debug: boolean): void {
+  const model = params.model || 'gemini-unknown';
+  writeLlmEvent({
+    kind: 'llm_call', provider: 'gemini', model,
+    durationMs: round(performance.now() - startTime),
+    inputTokens: 0, outputTokens: 0, totalTokens: 0, estimatedCostUsd: 0,
+    stream: false, finishReason: 'error',
+    temperature: params.config?.temperature,
+    maxTokens: params.config?.maxOutputTokens,
+    inputPreview: typeof params.contents === 'string' ? truncate(params.contents) : '',
+    outputPreview: '', messageCount: 0,
+    toolUse: false, timestamp: Date.now(),
+    error: truncate(err?.message || String(err), 200),
+  });
+}
+
+async function handleGeminiStream(stream: any, params: any, startTime: number, debug: boolean): Promise<any> {
+  if (!stream || !stream[Symbol.asyncIterator]) return stream;
+
+  const chunks: string[] = [];
+  const origIterator = stream[Symbol.asyncIterator].bind(stream);
+  let lastUsage: any = null;
+
+  stream[Symbol.asyncIterator] = function () {
+    const iter = origIterator();
+    return {
+      async next() {
+        const result = await iter.next();
+        if (!result.done) {
+          const chunk = result.value;
+          try { if (chunk.text) chunks.push(chunk.text); } catch {}
+          if (chunk.usageMetadata) lastUsage = chunk.usageMetadata;
+        } else {
+          // Stream finished
+          const model = params.model || 'gemini-unknown';
+          const inputTokens = lastUsage?.promptTokenCount || 0;
+          const outputTokens = lastUsage?.candidatesTokenCount || 0;
+          writeLlmEvent({
+            kind: 'llm_call', provider: 'gemini', model,
+            durationMs: round(performance.now() - startTime),
+            inputTokens, outputTokens, totalTokens: inputTokens + outputTokens,
+            estimatedCostUsd: estimateCost(model, inputTokens, outputTokens),
+            stream: true, finishReason: 'stop',
+            temperature: params.config?.temperature,
+            maxTokens: params.config?.maxOutputTokens,
+            inputPreview: typeof params.contents === 'string' ? truncate(params.contents) : '',
+            outputPreview: truncate(chunks.join('')),
+            messageCount: 0, toolUse: false, timestamp: Date.now(),
+          });
+          if (debug) console.log(`[trickle/llm] Gemini stream: ${model} (${outputTokens} tokens)`);
+        }
+        return result;
+      },
+      return: iter.return?.bind(iter),
+      throw: iter.throw?.bind(iter),
+    };
+  };
+  return stream;
+}
+
 // ────────────────────────────────────────────────────
 // Helpers
 // ────────────────────────────────────────────────────

package/src/observe-register.ts

@@ -1537,6 +1537,15 @@ if (enabled) {
       } catch { /* not critical */ }
     }
 
+    // Google Gemini SDK
+    if (request === '@google/genai' && !expressPatched.has('@google/genai')) {
+      expressPatched.add('@google/genai');
+      try {
+        const { patchGemini } = require(path.join(__dirname, 'llm-observer.js'));
+        patchGemini(exports, debug);
+      } catch { /* not critical */ }
+    }
+
     // Resolve to absolute path for dedup — do this FIRST since bundlers like
     // tsx/esbuild may use path aliases (e.g., @config/env) that don't start
     // with './' or '/'. We need the resolved path to decide if it's user code.