agentl
v0.1.2
Published
High-performance Agent SDK with Tree-Shaking memory, Parallel Tool Execution, and Shell access. An OpenClaw replacement.
Downloads
287
Readme
Agent-L 🚀🤖
High-Performance, Production-Grade Agentic Engine.
Agent-L is a lightweight, deterministic SDK designed to replace bloated agent frameworks (like OpenClaw). It's optimized for speed (Parallel Tools), efficiency (Tree Shaking Memory), and power (Local Shell Access).
Key Features
- ⚡ Parallel Tool Execution: Concurrently execute multiple tool calls using
Promise.all. - 🌳 Tree Shaking Context: Intelligent pruning that always keeps System Prompt + relevant memories + recent flow.
- 🧠 Semantic Memory: Zero-cost local embeddings or persistent Qdrant storage.
- 🐚 Local Shell Access: Sandboxed or direct terminal execution capabilities.
- 🛡️ Production Grade: Auto-retry with exponential backoff, SSE streaming, and observability hooks.
- 🔌 Multi-Channel: Built-in support for Telegram and Webhooks.
Installation
npm install agentlQuick Start
import { runAgent, OpenRouterProvider, ToolRegistry, ShellTool } from 'agentl';
const provider = new OpenRouterProvider({ apiKey: 'YOUR_API_KEY' });
const registry = new ToolRegistry();
registry.register(ShellTool.definition);
const response = await runAgent(provider, {
messages: [{ role: 'user', content: 'List files in the current directory' }],
onStream: (chunk) => process.stdout.write(chunk),
}, registry);
console.log(response.content);Advanced Usage
Telegram Bot
import { TelegramConnector } from 'agentl';
const tg = new TelegramConnector(process.env.TG_TOKEN, provider, registry);
tg.launch();Docker Deployment
docker-compose up --buildLicense
MIT
