Meta AI SDK

TypeScript SDK and CLI for building AI-powered developer tools, agents, and integrations with Meta AI.

NPM Version License: MIT

Buy Me a Coffee
--- ## Overview **Meta AI SDK** provides a unified TypeScript interface for building, integrating, and automating with Meta AI models. It includes: - SDK for agent orchestration, tool integration, and context management - Utilities for safe, context-aware automation - CLI for advanced developer workflows (coming soon) --- ## Features - **MetaAI Class**: Simple interface for interacting with Meta AI models - **Agent Framework**: Compose multi-agent systems with planning, tool selection, and feedback loops - **Safe Execution**: Built-in guardrails to prevent destructive actions - **Streaming & Context**: Stream model outputs and manage working memory - **Extensible Tools**: Integrate your own tools, functions, and APIs - **Beautiful Terminal Output**: Colorful, structured, and readable logs --- ## Installation Add to your project: ```bash npm install npm i meta-ai-sdk ``` > Requires Node.js 18+ and TypeScript 5+ for SDK usage. --- ## Usage ### Basic Usage with MetaAI ```typescript import { MetaAI } from '@orello/meta-sdk'; const meta = new MetaAI({ apiKey: 'YOUR_API_KEY', model: 'meta-llama/Llama-2-7b-chat-hf', }); async function main() { const response = await meta.generateContent({ prompt: 'Summarize the latest AI research trends.', maxTokens: 256, temperature: 0.7, }); console.log(response.result); } main(); ``` ### Streaming Responses ```typescript import { MetaAI } from '@orello/meta-sdk'; const meta = new MetaAI({ apiKey: 'YOUR_API_KEY' }); async function streamExample() { const stream = await meta.generateContentStream({ prompt: 'Write a poem about the ocean.', maxTokens: 100, }); for await (const chunk of stream) { process.stdout.write(chunk.text); } } streamExample(); ``` --- ## Agentic Workflow Example The SDK supports multi-agent orchestration: 1. **Information Gathering**: Collect files, logs, docs, and context 2. **Planning**: Generate a markdown implementation plan and decompose into tasks 3. **Tool Detection**: Select tools/functions and create execution plans 4. **Execution & Feedback**: Run plans, detect errors, and retry with feedback loops --- ## Contributing Contributions are welcome! Please open an issue or submit a pull request. --- ## License MIT License