GitHub
FreeAn LLM-powered autonomous agent platform
FreeFree tier
Inputs: textOutputs: text
About GitHub
AI Legion is an open-source, LLM-powered autonomous agent platform that enables multiple agents to collaborate on tasks. Built as a framework, it allows users to spin up a configurable number of agents (using GPT-3.5-turbo or GPT-4) that communicate via a console interface. Agents maintain persistent state (memory, goals, notes) in the .store directory, learn from mistakes over time, and can perform actions like web searches (via Google Custom Search API) and file operations. The project is designed for experimentation, debugging, and multi-agent coordination.
Key Features
Multi-agent collaboration: run multiple LLM-powered agents that work together on tasks
Console-based interaction: send messages to all agents in real-time
Agent memory and state persistence: agents store memory, goals, and notes in .store directory
Supports GPT-3.5-turbo and GPT-4 models
Web search capability via Google Custom Search API
Agents learn from mistakes and self-correct over time
State resumable: restart agents and they pick up from where they left off
MIT license open-source codebase
Pros & Cons
Pros
- Completely free and open-source (MIT license)
- Supports multiple LLM backends (GPT-3.5-turbo, GPT-4)
- Agents persist state across restarts for debugging and replay
- Customizable via environment variables and configuration
- Active GitHub repository with 1.4k stars and community forks
Cons
- Requires OpenAI API key and Google Custom Search API key (costs may apply)
- GPT-3.5-turbo agents can get stuck in infinite error loops under certain conditions
- Console-only interface; no graphical or web UI
- Agents may need manual supervision to avoid excessive token consumption
- Setup requires Node.js and multiple API integrations
Best For
Autonomous task completion with multi-agent coordinationExperimentation with LLM agent behavior and learningResearch on open-source autonomous agent frameworksPrototyping collaborative AI systems
FAQ
How do I set up AI Legion?
Clone the repository, install dependencies with 'npm install', rename .env.template to .env, and add your OpenAI API key (and optionally Google Custom Search API keys for web search). Then run with 'npm run start [# of agents] [model]'.
Can agents learn from mistakes?
Yes, agents will often make initial errors (e.g., using actions before knowing parameters) but generally learn and recover. However, gpt-3.5-turbo agents may devolve into endless error loops and require restart.
How is agent state managed?
Each agent stores its state (memory, goals, notes) under the .store directory. You can delete files to reset specific agents or the entire state between runs. On restart, agents resume from where they left off.
What models does AI Legion support?
It supports gpt-3.5-turbo and gpt-4. You specify the model when starting the program.