ai-functions 2.1.3 → 2.3.0

package/src/batch/cloudflare.js

@@ -1,287 +0,0 @@
-/**
- * Cloudflare AI Gateway Batch Adapter
- *
- * Implements batch processing using Cloudflare AI Gateway's batch endpoint.
- * Cloudflare's batch API is newer but integrates well with Workers.
- *
- * Note: Cloudflare's batch API is evolving. This adapter implements
- * the current batch capabilities and can be extended as new features land.
- *
- * @see https://developers.cloudflare.com/ai-gateway/
- *
- * @packageDocumentation
- */
-import { registerBatchAdapter, } from '../batch-queue.js';
-// ============================================================================
-// Cloudflare Client Configuration
-// ============================================================================
-let accountId;
-let gatewayId;
-let apiToken;
-let baseUrl = 'https://api.cloudflare.com/client/v4';
-/**
- * Configure the Cloudflare client
- */
-export function configureCloudflare(options) {
-    if (options.accountId)
-        accountId = options.accountId;
-    if (options.gatewayId)
-        gatewayId = options.gatewayId;
-    if (options.apiToken)
-        apiToken = options.apiToken;
-    if (options.baseUrl)
-        baseUrl = options.baseUrl;
-}
-function getConfig() {
-    const accId = accountId || process.env.CLOUDFLARE_ACCOUNT_ID;
-    const gwId = gatewayId || process.env.CLOUDFLARE_AI_GATEWAY_ID || process.env.AI_GATEWAY_ID;
-    const token = apiToken || process.env.CLOUDFLARE_API_TOKEN;
-    if (!accId) {
-        throw new Error('Cloudflare account ID not configured. Set CLOUDFLARE_ACCOUNT_ID or call configureCloudflare()');
-    }
-    if (!gwId) {
-        throw new Error('Cloudflare AI Gateway ID not configured. Set CLOUDFLARE_AI_GATEWAY_ID or call configureCloudflare()');
-    }
-    if (!token) {
-        throw new Error('Cloudflare API token not configured. Set CLOUDFLARE_API_TOKEN or call configureCloudflare()');
-    }
-    return { accountId: accId, gatewayId: gwId, apiToken: token };
-}
-async function cloudflareRequest(method, path, body) {
-    const config = getConfig();
-    const url = `${baseUrl}${path}`;
-    const response = await fetch(url, {
-        method,
-        headers: {
-            Authorization: `Bearer ${config.apiToken}`,
-            'Content-Type': 'application/json',
-        },
-        body: body ? JSON.stringify(body) : undefined,
-    });
-    if (!response.ok) {
-        const error = await response.text();
-        throw new Error(`Cloudflare API error: ${response.status} ${error}`);
-    }
-    const data = await response.json();
-    if (!data.success) {
-        throw new Error(`Cloudflare API error: ${JSON.stringify(data.errors)}`);
-    }
-    return data.result;
-}
-// ============================================================================
-// In-memory job storage (for polling)
-// ============================================================================
-const pendingJobs = new Map();
-let jobCounter = 0;
-// ============================================================================
-// Cloudflare Batch Adapter
-// ============================================================================
-/**
- * Cloudflare batch adapter
- *
- * Note: Cloudflare's AI Gateway doesn't have a native batch API like OpenAI/Anthropic.
- * This adapter implements batch processing by:
- * 1. Sending requests concurrently through the gateway
- * 2. Utilizing Cloudflare's caching and rate limiting
- * 3. Tracking job state locally (or in D1/KV for production)
- *
- * For true async batch processing, consider using Cloudflare Queues + Workers.
- */
-const cloudflareAdapter = {
-    async submit(items, options) {
-        const config = getConfig();
-        const jobId = `cf_batch_${++jobCounter}_${Date.now()}`;
-        const model = options.model || 'mistral/mistral-7b-instruct-v0.1';
-        // Store job state
-        pendingJobs.set(jobId, {
-            items,
-            options,
-            results: [],
-            status: 'pending',
-            createdAt: new Date(),
-        });
-        // Process requests concurrently (Cloudflare handles rate limiting)
-        const completion = processCloudflareRequests(jobId, items, config, model, options);
-        const job = {
-            id: jobId,
-            provider: 'cloudflare',
-            status: 'pending',
-            totalItems: items.length,
-            completedItems: 0,
-            failedItems: 0,
-            createdAt: new Date(),
-            webhookUrl: options.webhookUrl,
-        };
-        return { job, completion };
-    },
-    async getStatus(batchId) {
-        const job = pendingJobs.get(batchId);
-        if (!job) {
-            throw new Error(`Batch not found: ${batchId}`);
-        }
-        const completedItems = job.results.filter((r) => r.status === 'completed').length;
-        const failedItems = job.results.filter((r) => r.status === 'failed').length;
-        return {
-            id: batchId,
-            provider: 'cloudflare',
-            status: job.status,
-            totalItems: job.items.length,
-            completedItems,
-            failedItems,
-            createdAt: job.createdAt,
-            completedAt: job.completedAt,
-        };
-    },
-    async cancel(batchId) {
-        const job = pendingJobs.get(batchId);
-        if (job) {
-            job.status = 'cancelled';
-        }
-    },
-    async getResults(batchId) {
-        const job = pendingJobs.get(batchId);
-        if (!job) {
-            throw new Error(`Batch not found: ${batchId}`);
-        }
-        return job.results;
-    },
-    async waitForCompletion(batchId, pollInterval = 1000) {
-        const job = pendingJobs.get(batchId);
-        if (!job) {
-            throw new Error(`Batch not found: ${batchId}`);
-        }
-        while (job.status !== 'completed' && job.status !== 'failed' && job.status !== 'cancelled') {
-            await new Promise((resolve) => setTimeout(resolve, pollInterval));
-        }
-        return job.results;
-    },
-};
-// ============================================================================
-// Processing
-// ============================================================================
-async function processCloudflareRequests(jobId, items, config, model, options) {
-    const job = pendingJobs.get(jobId);
-    if (!job) {
-        throw new Error(`Job not found: ${jobId}`);
-    }
-    job.status = 'in_progress';
-    // Process all requests concurrently with concurrency limit
-    const CONCURRENCY = 10;
-    const results = [];
-    for (let i = 0; i < items.length; i += CONCURRENCY) {
-        const batch = items.slice(i, i + CONCURRENCY);
-        const batchResults = await Promise.all(batch.map(async (item) => {
-            try {
-                const result = await processCloudflareItem(item, config, model);
-                return result;
-            }
-            catch (error) {
-                return {
-                    id: item.id,
-                    customId: item.id,
-                    status: 'failed',
-                    error: error instanceof Error ? error.message : 'Unknown error',
-                };
-            }
-        }));
-        results.push(...batchResults);
-        job.results = results;
-    }
-    job.status = results.every((r) => r.status === 'completed') ? 'completed' : 'failed';
-    job.completedAt = new Date();
-    return results;
-}
-async function processCloudflareItem(item, config, model) {
-    // Route through AI Gateway
-    const gatewayUrl = `https://gateway.ai.cloudflare.com/v1/${config.accountId}/${config.gatewayId}`;
-    // Determine provider from model
-    let provider = 'workers-ai';
-    let endpoint = '';
-    if (model.startsWith('openai/') || model.startsWith('gpt-')) {
-        provider = 'openai';
-        endpoint = '/chat/completions';
-    }
-    else if (model.startsWith('anthropic/') || model.startsWith('claude-')) {
-        provider = 'anthropic';
-        endpoint = '/messages';
-    }
-    else if (model.startsWith('@cf/') || model.startsWith('workers-ai/')) {
-        provider = 'workers-ai';
-        endpoint = `/ai/run/${model.replace('workers-ai/', '').replace('@cf/', '')}`;
-    }
-    else {
-        // Default to OpenAI-compatible
-        provider = 'openai';
-        endpoint = '/chat/completions';
-    }
-    const url = `${gatewayUrl}/${provider}${endpoint}`;
-    const messages = [
-        ...(item.options?.system ? [{ role: 'system', content: item.options.system }] : []),
-        { role: 'user', content: item.prompt },
-    ];
-    const body = {
-        model: model.replace(`${provider}/`, ''),
-        messages,
-        max_tokens: item.options?.maxTokens || 4096,
-        temperature: item.options?.temperature,
-    };
-    // Add JSON mode if schema is provided
-    if (item.schema) {
-        body.response_format = { type: 'json_object' };
-    }
-    const response = await fetch(url, {
-        method: 'POST',
-        headers: {
-            'cf-aig-authorization': `Bearer ${config.apiToken}`,
-            'Content-Type': 'application/json',
-        },
-        body: JSON.stringify(body),
-    });
-    if (!response.ok) {
-        const error = await response.text();
-        throw new Error(`Cloudflare Gateway error: ${response.status} ${error}`);
-    }
-    const data = await response.json();
-    // Extract content based on response format
-    let content;
-    if (data.choices?.[0]?.message?.content) {
-        // OpenAI format
-        content = data.choices[0].message.content;
-    }
-    else if (data.content?.[0]?.text) {
-        // Anthropic format
-        content = data.content[0].text;
-    }
-    else if (data.response) {
-        // Workers AI format
-        content = data.response;
-    }
-    let result = content;
-    // Try to parse JSON if schema was provided
-    if (item.schema && content) {
-        try {
-            result = JSON.parse(content);
-        }
-        catch {
-            // Keep as string
-        }
-    }
-    return {
-        id: item.id,
-        customId: item.id,
-        status: 'completed',
-        result,
-        usage: data.usage
-            ? {
-                promptTokens: data.usage.prompt_tokens,
-                completionTokens: data.usage.completion_tokens,
-                totalTokens: data.usage.total_tokens,
-            }
-            : undefined,
-    };
-}
-// ============================================================================
-// Register Adapter
-// ============================================================================
-registerBatchAdapter('cloudflare', cloudflareAdapter);
-export { cloudflareAdapter };

package/src/batch/google.js

@@ -1,359 +0,0 @@
-/**
- * Google GenAI (Gemini) Adapter
- *
- * Implements processing using Google's Generative AI API (Gemini models).
- * Google doesn't have a native batch API like OpenAI/Anthropic, so this
- * implements concurrent processing for the flex tier.
- *
- * @see https://ai.google.dev/gemini-api/docs
- *
- * @packageDocumentation
- */
-import { registerBatchAdapter, registerFlexAdapter, } from '../batch-queue.js';
-// ============================================================================
-// Google GenAI Client Configuration
-// ============================================================================
-let googleApiKey;
-let googleBaseUrl = 'https://generativelanguage.googleapis.com/v1beta';
-// AI Gateway configuration (optional - for routing through Cloudflare AI Gateway)
-let gatewayUrl;
-let gatewayToken;
-/**
- * Configure the Google GenAI client
- */
-export function configureGoogleGenAI(options) {
-    if (options.apiKey)
-        googleApiKey = options.apiKey;
-    if (options.baseUrl)
-        googleBaseUrl = options.baseUrl;
-    if (options.gatewayUrl)
-        gatewayUrl = options.gatewayUrl;
-    if (options.gatewayToken)
-        gatewayToken = options.gatewayToken;
-}
-function getConfig() {
-    // Check for AI Gateway configuration
-    const gwUrl = gatewayUrl || process.env.AI_GATEWAY_URL;
-    const gwToken = gatewayToken || process.env.AI_GATEWAY_TOKEN;
-    // If using gateway, we don't need a direct API key
-    if (gwUrl && gwToken) {
-        return {
-            apiKey: '',
-            baseUrl: googleBaseUrl,
-            gatewayUrl: gwUrl,
-            gatewayToken: gwToken,
-        };
-    }
-    const key = googleApiKey || process.env.GOOGLE_API_KEY || process.env.GEMINI_API_KEY;
-    if (!key) {
-        throw new Error('Google API key not configured. Set GOOGLE_API_KEY or GEMINI_API_KEY, or use AI_GATEWAY_URL and AI_GATEWAY_TOKEN');
-    }
-    return { apiKey: key, baseUrl: googleBaseUrl, gatewayUrl: undefined, gatewayToken: undefined };
-}
-// ============================================================================
-// In-memory job tracking
-// ============================================================================
-const pendingJobs = new Map();
-let jobCounter = 0;
-// ============================================================================
-// Google GenAI Batch Adapter
-// ============================================================================
-/**
- * Google GenAI batch adapter
- *
- * Note: Google doesn't have a native batch API like OpenAI/Anthropic.
- * This adapter implements batch processing via concurrent requests.
- * For true async batch processing, consider using Google Cloud Batch
- * with Vertex AI.
- */
-const googleAdapter = {
-    async submit(items, options) {
-        const jobId = `google_batch_${++jobCounter}_${Date.now()}`;
-        const model = options.model || 'gemini-2.0-flash';
-        // Store job state
-        pendingJobs.set(jobId, {
-            items,
-            options,
-            results: [],
-            status: 'pending',
-            createdAt: new Date(),
-        });
-        // Process requests concurrently
-        const completion = processGoogleRequestsConcurrently(jobId, items, model, options);
-        const job = {
-            id: jobId,
-            provider: 'google',
-            status: 'pending',
-            totalItems: items.length,
-            completedItems: 0,
-            failedItems: 0,
-            createdAt: new Date(),
-            webhookUrl: options.webhookUrl,
-        };
-        return { job, completion };
-    },
-    async getStatus(batchId) {
-        const job = pendingJobs.get(batchId);
-        if (!job) {
-            throw new Error(`Batch not found: ${batchId}`);
-        }
-        const completedItems = job.results.filter((r) => r.status === 'completed').length;
-        const failedItems = job.results.filter((r) => r.status === 'failed').length;
-        return {
-            id: batchId,
-            provider: 'google',
-            status: job.status,
-            totalItems: job.items.length,
-            completedItems,
-            failedItems,
-            createdAt: job.createdAt,
-            completedAt: job.completedAt,
-        };
-    },
-    async cancel(batchId) {
-        const job = pendingJobs.get(batchId);
-        if (job) {
-            job.status = 'cancelled';
-        }
-    },
-    async getResults(batchId) {
-        const job = pendingJobs.get(batchId);
-        if (!job) {
-            throw new Error(`Batch not found: ${batchId}`);
-        }
-        return job.results;
-    },
-    async waitForCompletion(batchId, pollInterval = 1000) {
-        const job = pendingJobs.get(batchId);
-        if (!job) {
-            throw new Error(`Batch not found: ${batchId}`);
-        }
-        while (job.status !== 'completed' && job.status !== 'failed' && job.status !== 'cancelled') {
-            await new Promise((resolve) => setTimeout(resolve, pollInterval));
-        }
-        return job.results;
-    },
-};
-// ============================================================================
-// Google GenAI Flex Adapter
-// ============================================================================
-/**
- * Google GenAI Flex Adapter
- *
- * Implements concurrent processing for medium-sized batches.
- * Uses the Gemini API for fast turnaround.
- */
-const googleFlexAdapter = {
-    async submitFlex(items, options) {
-        const model = options.model || 'gemini-2.0-flash';
-        const CONCURRENCY = 10;
-        const results = [];
-        // Process items concurrently
-        for (let i = 0; i < items.length; i += CONCURRENCY) {
-            const batch = items.slice(i, i + CONCURRENCY);
-            const batchResults = await Promise.all(batch.map(async (item) => {
-                try {
-                    return await processGoogleItem(item, model);
-                }
-                catch (error) {
-                    return {
-                        id: item.id,
-                        customId: item.id,
-                        status: 'failed',
-                        error: error instanceof Error ? error.message : 'Unknown error',
-                    };
-                }
-            }));
-            results.push(...batchResults);
-        }
-        return results;
-    },
-};
-// ============================================================================
-// Processing
-// ============================================================================
-async function processGoogleRequestsConcurrently(jobId, items, model, options) {
-    const job = pendingJobs.get(jobId);
-    if (!job) {
-        throw new Error(`Job not found: ${jobId}`);
-    }
-    job.status = 'in_progress';
-    const CONCURRENCY = 10;
-    const results = [];
-    for (let i = 0; i < items.length; i += CONCURRENCY) {
-        const batch = items.slice(i, i + CONCURRENCY);
-        const batchResults = await Promise.all(batch.map(async (item) => {
-            try {
-                return await processGoogleItem(item, model);
-            }
-            catch (error) {
-                return {
-                    id: item.id,
-                    customId: item.id,
-                    status: 'failed',
-                    error: error instanceof Error ? error.message : 'Unknown error',
-                };
-            }
-        }));
-        results.push(...batchResults);
-        job.results = results;
-    }
-    job.status = results.every((r) => r.status === 'completed') ? 'completed' : 'failed';
-    job.completedAt = new Date();
-    return results;
-}
-async function processGoogleItem(item, model) {
-    const config = getConfig();
-    // Check if using AI Gateway
-    if (config.gatewayUrl && config.gatewayToken) {
-        return processGoogleItemViaGateway(item, config, model);
-    }
-    // Build the model name (add models/ prefix if not present)
-    const modelName = model.startsWith('models/') ? model : `models/${model}`;
-    const url = `${config.baseUrl}/${modelName}:generateContent?key=${config.apiKey}`;
-    // Build messages
-    const contents = [];
-    // Add system instruction as a user message if provided (Gemini handles this differently)
-    if (item.options?.system) {
-        contents.push({
-            role: 'user',
-            parts: [{ text: `System instruction: ${item.options.system}\n\nUser request: ${item.prompt}` }],
-        });
-    }
-    else {
-        contents.push({
-            role: 'user',
-            parts: [{ text: item.prompt }],
-        });
-    }
-    const body = {
-        contents,
-        generationConfig: {
-            maxOutputTokens: item.options?.maxTokens || 8192,
-            temperature: item.options?.temperature,
-        },
-    };
-    // Add JSON mode if schema is provided
-    if (item.schema) {
-        body.generationConfig = {
-            ...body.generationConfig,
-            responseMimeType: 'application/json',
-        };
-    }
-    const response = await fetch(url, {
-        method: 'POST',
-        headers: {
-            'Content-Type': 'application/json',
-        },
-        body: JSON.stringify(body),
-    });
-    if (!response.ok) {
-        const error = await response.text();
-        throw new Error(`Google GenAI API error: ${response.status} ${error}`);
-    }
-    const data = (await response.json());
-    // Extract content
-    const content = data.candidates?.[0]?.content?.parts?.[0]?.text;
-    let result = content;
-    // Try to parse JSON if schema was provided or content looks like JSON
-    if (content && (item.schema || content.trim().startsWith('{') || content.trim().startsWith('['))) {
-        try {
-            result = JSON.parse(content);
-        }
-        catch {
-            // Keep as string
-        }
-    }
-    return {
-        id: item.id,
-        customId: item.id,
-        status: 'completed',
-        result,
-        usage: data.usageMetadata
-            ? {
-                promptTokens: data.usageMetadata.promptTokenCount,
-                completionTokens: data.usageMetadata.candidatesTokenCount,
-                totalTokens: data.usageMetadata.totalTokenCount,
-            }
-            : undefined,
-    };
-}
-/**
- * Process a Google GenAI item via Cloudflare AI Gateway
- * Gateway URL format: {gateway_url}/google-ai-studio/v1beta/models/{model}:generateContent
- */
-async function processGoogleItemViaGateway(item, config, model) {
-    // AI Gateway URL for Google AI Studio
-    // Format: {gateway_url}/google-ai-studio/v1beta/models/{model}:generateContent
-    const modelName = model.startsWith('models/') ? model.replace('models/', '') : model;
-    const url = `${config.gatewayUrl}/google-ai-studio/v1beta/models/${modelName}:generateContent`;
-    // Build messages
-    const contents = [];
-    if (item.options?.system) {
-        contents.push({
-            role: 'user',
-            parts: [{ text: `System instruction: ${item.options.system}\n\nUser request: ${item.prompt}` }],
-        });
-    }
-    else {
-        contents.push({
-            role: 'user',
-            parts: [{ text: item.prompt }],
-        });
-    }
-    const body = {
-        contents,
-        generationConfig: {
-            maxOutputTokens: item.options?.maxTokens || 8192,
-            temperature: item.options?.temperature,
-        },
-    };
-    if (item.schema) {
-        body.generationConfig = {
-            ...body.generationConfig,
-            responseMimeType: 'application/json',
-        };
-    }
-    const response = await fetch(url, {
-        method: 'POST',
-        headers: {
-            'cf-aig-authorization': `Bearer ${config.gatewayToken}`,
-            'Content-Type': 'application/json',
-        },
-        body: JSON.stringify(body),
-    });
-    if (!response.ok) {
-        const error = await response.text();
-        throw new Error(`Google GenAI via Gateway error: ${response.status} ${error}`);
-    }
-    const data = (await response.json());
-    const content = data.candidates?.[0]?.content?.parts?.[0]?.text;
-    let result = content;
-    if (content && (item.schema || content.trim().startsWith('{') || content.trim().startsWith('['))) {
-        try {
-            result = JSON.parse(content);
-        }
-        catch {
-            // Keep as string
-        }
-    }
-    return {
-        id: item.id,
-        customId: item.id,
-        status: 'completed',
-        result,
-        usage: data.usageMetadata
-            ? {
-                promptTokens: data.usageMetadata.promptTokenCount,
-                completionTokens: data.usageMetadata.candidatesTokenCount,
-                totalTokens: data.usageMetadata.totalTokenCount,
-            }
-            : undefined,
-    };
-}
-// ============================================================================
-// Register Adapters
-// ============================================================================
-registerBatchAdapter('google', googleAdapter);
-registerFlexAdapter('google', googleFlexAdapter);
-export { googleAdapter, googleFlexAdapter };

package/src/batch/index.js

@@ -1,30 +0,0 @@
-/**
- * Batch Adapters Index
- *
- * Import specific adapters to register them:
- *
- * @example
- * ```ts
- * // Import to register the OpenAI batch adapter
- * import 'ai-functions/batch/openai'
- *
- * // Import to register the Anthropic batch adapter
- * import 'ai-functions/batch/anthropic'
- *
- * // Import to register the Cloudflare adapter
- * import 'ai-functions/batch/cloudflare'
- *
- * // Import to register the AWS Bedrock adapter
- * import 'ai-functions/batch/bedrock'
- *
- * // Or import the in-memory adapter for testing
- * import 'ai-functions/batch/memory'
- * ```
- *
- * @packageDocumentation
- */
-export * from './openai.js';
-export * from './anthropic.js';
-export * from './cloudflare.js';
-export * from './bedrock.js';
-export * from './memory.js';