Vel Agent Runtime

A production-ready AI agent runtime aligned with 12-Factor Agent principles. Built for reliability, scalability, and maintainability.

Get started now View on GitHub

Features

Dual Execution Modes: Streaming (SSE) and non-streaming (JSON) responses
Multiple LLM Providers: OpenAI, Google Gemini, and Anthropic Claude with plug-and-play architecture
Flexible Prompts: Jinja2 templating with XML formatting, environment-based configuration, and version control
Stream Protocol: Vercel AI SDK-compatible event system for provider-agnostic streaming
Tool System: JSON schema-validated tools with async support
Persistent Storage: PostgreSQL for durability, Redis for caching
Production-Ready: FastAPI service with health checks and 12-Factor alignment

Quick Start

Installation

# Clone and install
git clone https://github.com/rscheiwe/vel
cd vel
pip install -e .

# Set up environment
cp .env.example .env
# Edit .env with your API keys

Basic Usage

import asyncio
from vel import Agent

async def main():
    agent = Agent(
        id='chat-general:v1',
        model={'provider': 'openai', 'model': 'gpt-4o'},
        tools=['get_weather']
    )

    answer = await agent.run({'message': 'What is the weather?'})
    print(answer)

if __name__ == '__main__':
    asyncio.run(main())