compute-cfo 0.1.0 → 0.3.0

package/dist/pricing.d.ts

@@ -1,5 +1,5 @@
 /**
- * Model pricing database for OpenAI and Anthropic.
+ * Model pricing database for OpenAI, Anthropic, Google Gemini, and Mistral.
  * Prices are in USD per 1 million tokens. Updated March 2026.
  */
 export interface ModelPrice {

package/dist/pricing.js

@@ -1,6 +1,6 @@
 "use strict";
 /**
- * Model pricing database for OpenAI and Anthropic.
+ * Model pricing database for OpenAI, Anthropic, Google Gemini, and Mistral.
  * Prices are in USD per 1 million tokens. Updated March 2026.
  */
 Object.defineProperty(exports, "__esModule", { value: true });
@@ -52,6 +52,36 @@ const MODEL_PRICES = {
     'claude-3-opus': { inputPerMillion: 15.0, outputPerMillion: 75.0 },
     'claude-3-sonnet': { inputPerMillion: 3.0, outputPerMillion: 15.0 },
     'claude-3-haiku': { inputPerMillion: 0.25, outputPerMillion: 1.25 },
+    // ── Google Gemini ─────────────────────────────────────────
+    'gemini-2.5-pro': { inputPerMillion: 1.25, outputPerMillion: 10.0 },
+    'gemini-2.5-flash': { inputPerMillion: 0.3, outputPerMillion: 2.5 },
+    'gemini-2.5-flash-lite': { inputPerMillion: 0.1, outputPerMillion: 0.4 },
+    'gemini-2.0-flash': { inputPerMillion: 0.1, outputPerMillion: 0.4 },
+    'gemini-1.5-pro': { inputPerMillion: 1.25, outputPerMillion: 5.0 },
+    'gemini-1.5-flash': { inputPerMillion: 0.075, outputPerMillion: 0.3 },
+    'gemini-1.5-flash-8b': { inputPerMillion: 0.0375, outputPerMillion: 0.15 },
+    'gemini-embedding': { inputPerMillion: 0.15, outputPerMillion: 0 },
+    // ── Mistral ───────────────────────────────────────────────
+    'mistral-large-latest': { inputPerMillion: 0.5, outputPerMillion: 1.5 },
+    'mistral-medium-latest': { inputPerMillion: 0.4, outputPerMillion: 2.0 },
+    'mistral-small-latest': { inputPerMillion: 0.03, outputPerMillion: 0.11 },
+    'codestral-latest': { inputPerMillion: 0.3, outputPerMillion: 0.9 },
+    'pixtral-large-latest': { inputPerMillion: 2.0, outputPerMillion: 6.0 },
+    'mistral-nemo': { inputPerMillion: 0.02, outputPerMillion: 0.05 },
+    'pixtral-12b': { inputPerMillion: 0.15, outputPerMillion: 0.15 },
+    // ── Groq ──────────────────────────────────────────────────
+    'llama-3.3-70b-versatile': { inputPerMillion: 0.59, outputPerMillion: 0.79 },
+    'llama-3.1-8b-instant': { inputPerMillion: 0.05, outputPerMillion: 0.08 },
+    'llama-3.1-70b-versatile': { inputPerMillion: 0.59, outputPerMillion: 0.79 },
+    'gemma2-9b-it': { inputPerMillion: 0.2, outputPerMillion: 0.2 },
+    'mixtral-8x7b-32768': { inputPerMillion: 0.24, outputPerMillion: 0.24 },
+    // ── Together AI ───────────────────────────────────────────
+    'meta-llama/Llama-3.3-70B-Instruct-Turbo': { inputPerMillion: 0.88, outputPerMillion: 0.88 },
+    'meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo': { inputPerMillion: 0.18, outputPerMillion: 0.18 },
+    'meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo': { inputPerMillion: 0.88, outputPerMillion: 0.88 },
+    'meta-llama/Meta-Llama-3.1-405B-Instruct-Turbo': { inputPerMillion: 3.5, outputPerMillion: 3.5 },
+    'mistralai/Mixtral-8x7B-Instruct-v0.1': { inputPerMillion: 0.6, outputPerMillion: 0.6 },
+    'Qwen/Qwen2.5-72B-Instruct-Turbo': { inputPerMillion: 1.2, outputPerMillion: 1.2 },
 };
 const ALIASES = {
     'gpt-4o-2024-11-20': 'gpt-4o',
@@ -66,6 +96,19 @@ const ALIASES = {
     'gpt-3.5-turbo-1106': 'gpt-3.5-turbo',
     'o3-2025-04-16': 'o3',
     'o4-mini-2025-04-16': 'o4-mini',
+    // Gemini aliases
+    'models/gemini-2.5-pro': 'gemini-2.5-pro',
+    'models/gemini-2.5-flash': 'gemini-2.5-flash',
+    'models/gemini-2.5-flash-lite': 'gemini-2.5-flash-lite',
+    'models/gemini-2.0-flash': 'gemini-2.0-flash',
+    'models/gemini-1.5-pro': 'gemini-1.5-pro',
+    'models/gemini-1.5-flash': 'gemini-1.5-flash',
+    'models/gemini-1.5-flash-8b': 'gemini-1.5-flash-8b',
+    // Mistral aliases
+    'mistral-large-2501': 'mistral-large-latest',
+    'mistral-medium-2505': 'mistral-medium-latest',
+    'mistral-small-2503': 'mistral-small-latest',
+    'codestral-2501': 'codestral-latest',
 };
 function resolveModel(model) {
     return ALIASES[model] ?? model;

package/dist/wrapper.d.ts

@@ -1,5 +1,5 @@
 /**
- * Drop-in wrapper for OpenAI and Anthropic SDK clients.
+ * Drop-in wrapper for OpenAI, Anthropic, Google Gemini, and Mistral SDK clients.
  */
 import { CostTracker } from './tracker';
 export interface WrapOptions {

package/dist/wrapper.js

@@ -1,6 +1,6 @@
 "use strict";
 /**
- * Drop-in wrapper for OpenAI and Anthropic SDK clients.
+ * Drop-in wrapper for OpenAI, Anthropic, Google Gemini, and Mistral SDK clients.
  */
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.wrap = wrap;
@@ -25,7 +25,16 @@ function wrap(client, options) {
     if ('messages' in client && typeof client.messages?.create === 'function') {
         return wrapAnthropic(client, tracker);
     }
-    throw new TypeError(`Unsupported client type. Supported: OpenAI, Anthropic`);
+    if ('models' in client && typeof client.models?.generateContent === 'function') {
+        return wrapGemini(client, tracker);
+    }
+    if ('chat' in client && typeof client.chat?.complete === 'function') {
+        return wrapMistral(client, tracker);
+    }
+    throw new TypeError(`Unsupported client type. Supported: OpenAI, Anthropic, Google Gemini, Mistral`);
+    // Note: Groq and Together AI use OpenAI-compatible APIs, so they match the
+    // first condition (chat.completions.create) and are wrapped as OpenAI clients.
+    // The provider will be detected as "openai" in cost events.
 }
 function wrapOpenAI(client, tracker) {
     const originalCreate = client.chat.completions.create.bind(client.chat.completions);
@@ -35,6 +44,35 @@ function wrapOpenAI(client, tracker) {
         const model = rest.model ?? 'unknown';
         const start = performance.now();
         const response = await originalCreate(rest);
+        // Handle streaming responses
+        if (rest.stream && response && typeof response[Symbol.asyncIterator] === 'function') {
+            const originalIterator = response[Symbol.asyncIterator].bind(response);
+            let usage = null;
+            response[Symbol.asyncIterator] = async function* () {
+                for await (const chunk of originalIterator()) {
+                    if (chunk?.usage)
+                        usage = chunk.usage;
+                    yield chunk;
+                }
+                const latencyMs = Math.round((performance.now() - start) * 10) / 10;
+                const inputTokens = usage?.prompt_tokens ?? 0;
+                const outputTokens = usage?.completion_tokens ?? 0;
+                const costUsd = (0, pricing_1.getCost)(model, inputTokens, outputTokens);
+                const event = {
+                    timestamp: new Date().toISOString(),
+                    provider: 'openai',
+                    model,
+                    operation: 'chat.completions',
+                    inputTokens,
+                    outputTokens,
+                    costUsd,
+                    latencyMs,
+                    tags,
+                };
+                tracker.record(event);
+            };
+            return response;
+        }
         const latencyMs = Math.round((performance.now() - start) * 10) / 10;
         const usage = response?.usage;
         const inputTokens = usage?.prompt_tokens ?? 0;
@@ -78,6 +116,94 @@ function wrapOpenAI(client, tracker) {
         },
     });
 }
+function wrapGemini(client, tracker) {
+    const originalGenerateContent = client.models.generateContent.bind(client.models);
+    const trackedGenerateContent = async (params) => {
+        const { compute_cfo_tags, ...rest } = params ?? {};
+        const tags = compute_cfo_tags && typeof compute_cfo_tags === 'object' ? { ...compute_cfo_tags } : {};
+        let model = rest.model ?? 'unknown';
+        if (typeof model === 'string' && model.startsWith('models/')) {
+            model = model.slice('models/'.length);
+        }
+        const start = performance.now();
+        const response = await originalGenerateContent(rest);
+        const latencyMs = Math.round((performance.now() - start) * 10) / 10;
+        const usage = response?.usageMetadata;
+        const inputTokens = usage?.promptTokenCount ?? 0;
+        const outputTokens = usage?.candidatesTokenCount ?? 0;
+        const costUsd = (0, pricing_1.getCost)(model, inputTokens, outputTokens);
+        const event = {
+            timestamp: new Date().toISOString(),
+            provider: 'google',
+            model,
+            operation: 'generate_content',
+            inputTokens,
+            outputTokens,
+            costUsd,
+            latencyMs,
+            tags,
+        };
+        tracker.record(event);
+        return response;
+    };
+    return new Proxy(client, {
+        get(target, prop) {
+            if (prop === 'models') {
+                return new Proxy(target.models, {
+                    get(modelsTarget, modelsProp) {
+                        if (modelsProp === 'generateContent')
+                            return trackedGenerateContent;
+                        return modelsTarget[modelsProp];
+                    },
+                });
+            }
+            return target[prop];
+        },
+    });
+}
+function wrapMistral(client, tracker) {
+    const originalComplete = client.chat.complete.bind(client.chat);
+    const trackedComplete = async (params) => {
+        const { compute_cfo_tags, ...rest } = params ?? {};
+        const tags = compute_cfo_tags && typeof compute_cfo_tags === 'object' ? { ...compute_cfo_tags } : {};
+        const model = rest.model ?? 'unknown';
+        const start = performance.now();
+        const response = await originalComplete(rest);
+        const latencyMs = Math.round((performance.now() - start) * 10) / 10;
+        const usage = response?.usage;
+        const inputTokens = usage?.prompt_tokens ?? 0;
+        const outputTokens = usage?.completion_tokens ?? 0;
+        const actualModel = response?.model ?? model;
+        const costUsd = (0, pricing_1.getCost)(actualModel, inputTokens, outputTokens);
+        const event = {
+            timestamp: new Date().toISOString(),
+            provider: 'mistral',
+            model: actualModel,
+            operation: 'chat.complete',
+            inputTokens,
+            outputTokens,
+            costUsd,
+            latencyMs,
+            tags,
+        };
+        tracker.record(event);
+        return response;
+    };
+    return new Proxy(client, {
+        get(target, prop) {
+            if (prop === 'chat') {
+                return new Proxy(target.chat, {
+                    get(chatTarget, chatProp) {
+                        if (chatProp === 'complete')
+                            return trackedComplete;
+                        return chatTarget[chatProp];
+                    },
+                });
+            }
+            return target[prop];
+        },
+    });
+}
 function wrapAnthropic(client, tracker) {
     const originalCreate = client.messages.create.bind(client.messages);
     const trackedCreate = async (params) => {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "compute-cfo",
-  "version": "0.1.0",
+  "version": "0.3.0",
   "description": "Cost tracking, attribution, and budget enforcement for AI inference APIs",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",