@venturekit-pro/ai

Warning: This package is in active development and not production-ready. APIs may change without notice.

AI utilities for VentureKit — embeddings, vector stores, RAG pipelines, and agents with tool use.

Installation

npm install @venturekit-pro/ai@dev

Optional Peer Dependencies

Install the providers you need:

# OpenAI (embeddings, agents)
npm install openai

# AWS Bedrock (embeddings)
npm install @aws-sdk/client-bedrock-runtime

# Pinecone (vector store)
npm install @pinecone-database/pinecone

Overview

@venturekit-pro/ai provides:

• Embeddings — generate vector embeddings via OpenAI or AWS Bedrock
• Vector stores — store and query vectors with Pinecone, pgvector, or in-memory
• RAG pipelines — chunking, retrieval, and context building for retrieval-augmented generation
• Agents — AI agents with tool use via OpenAI function calling

Embeddings

import { createEmbedder } from '@venturekit-pro/ai';

const embedder = createEmbedder({
  provider: 'openai',
  model: 'text-embedding-3-small',
  apiKey: process.env.OPENAI_API_KEY,
});

const vector = await embedder.embed('What is VentureKit?');
const vectors = await embedder.embedBatch(['Question 1', 'Question 2']);

Vector Stores

import { createVectorStore } from '@venturekit-pro/ai';

const store = createVectorStore({
  provider: 'pinecone',
  indexName: 'my-index',
  apiKey: process.env.PINECONE_API_KEY,
});

await store.upsert([{ id: 'doc-1', vector, metadata: { title: 'Guide' } }]);
const results = await store.query(queryVector, { topK: 5 });

RAG Pipeline

import { createRagPipeline, chunkText } from '@venturekit-pro/ai';

const rag = createRagPipeline({
  embedder,
  vectorStore: store,
  chunkSize: 500,
  chunkOverlap: 50,
});

// Ingest documents
const chunks = chunkText(documentText, { size: 500, overlap: 50 });
await rag.ingest(chunks);

// Query with context
const context = await rag.retrieve('How do I deploy?', { topK: 3 });

Agents

import { createAgent, defineTool } from '@venturekit-pro/ai';

const searchTool = defineTool({
  name: 'search',
  description: 'Search the knowledge base',
  parameters: { query: { type: 'string', description: 'Search query' } },
  handler: async ({ query }) => {
    return await rag.retrieve(query);
  },
});

const agent = createAgent({
  model: 'gpt-4',
  apiKey: process.env.OPENAI_API_KEY,
  tools: [searchTool],
  systemPrompt: 'You are a helpful assistant.',
});

const response = await agent.run('How do I set up authentication?');

Chat Completion

Provider-neutral single- or multi-turn text completion across Anthropic, OpenAI, and Google. Construct one client per (provider, apiKey) and call .complete(); supply pricing to get a USD cost stamp back.

import { createChatClient } from '@venturekit-pro/ai';

const client = createChatClient({ provider: 'anthropic', apiKey: process.env.ANTHROPIC_API_KEY! });

const result = await client.complete({
  model: 'claude-sonnet-4-5-20250929',
  messages: [
    { role: 'system', content: 'You are a concise assistant.' },
    { role: 'user', content: 'Three words about cats.' },
  ],
  maxTokens: 100,
  pricing: { costPer1kInputUsd: 0.003, costPer1kOutputUsd: 0.015 },
});

console.log(result.output, result.usage, result.cost);

Pair with withFallback / retryWithBackoff from the fallback module to add primary→backup routing or exponential-backoff retries.

Image Generation

Provider-neutral text-to-image (currently Google Imagen), returning raw image bytes. The caller owns prompt composition (brand/style suffixes) and cost attribution — the result carries provider/model echo and latency only.

import { createImageClient } from '@venturekit-pro/ai';

const client = createImageClient({ provider: 'google', apiKey: process.env.GOOGLE_API_KEY! });

const { images } = await client.generate({
  model: 'imagen-4.0-ultra-generate-001',
  prompt: 'A sunlit Casablanca apartment, editorial photographic style.',
  aspectRatio: '16:9',
  count: 1,
});

// images[0].bytes (Uint8Array), images[0].mimeType

API Reference

See the API reference for full documentation.

License

Apache-2.0 — see LICENSE for details.