@weaveai/core

The framework-agnostic core of Weave — universal AI operations without framework dependencies.

Features

• Universal AI operations (generate, classify, extract, etc.)
• Provider-agnostic (OpenAI, Anthropic, Google, local, custom)
• Tool integration framework
• Memory and state management
• Built-in observability and cost tracking
• Streaming support (onChunk)
• Type-safe with TypeScript strict mode

Installation

yarn add @weaveai/core
# or
npm install @weaveai/core

Quick Start (async init)

import { Weave } from '@weaveai/core';

const weave = await Weave.createAsync({
  provider: { type: 'openai', apiKey: process.env.OPENAI_API_KEY! },
});

const { text, tokenCount } = await weave.generate('Write a haiku about AI', {
  streaming: true,
  onChunk: (chunk) => process.stdout.write(chunk),
});
console.log('\nTokens:', tokenCount);

API Reference

Weave

Main entry point for all AI operations.

const weave = await Weave.createAsync(config: WeaveConfig);

Config Options:

• provider (ProviderConfig): AI provider configuration (OpenAI, Anthropic, Google, etc.)
• cache (boolean): Enable response caching
• logging (LogLevel): Log level ('debug' | 'info' | 'error')
• timeout (number): Request timeout in ms

Operations

generate(prompt, options?)

Generate text from a prompt.

const { text } = await weave.generate('Write a poem about AI', {
  temperature: 0.8,
  maxTokens: 500,
  streaming: true,
  onChunk: (chunk) => console.log(chunk),
});

classify(text, labels, options?)

Classify text into predefined categories.

const result = await weave.classify('This product is amazing!', [
  'positive',
  'negative',
  'neutral',
]);

extract(text, schema, options?)

Extract structured data from unstructured text.

const schema = {
  name: 'string',
  email: 'string',
  phone: 'string?',
};

const data = await weave.extract(resume, { schema });