Purpose
AI Legion is an LLM-based autonomous agent platform that provides a framework for multiple autonomous agents to collaborate on tasks.
Main capabilities
- Multiple autonomous agents can work together to complete tasks.
- Interact through the console with all current agents, and the input content is sent to them as messages.
- Agent state is persistently stored in the .store directory, and state can be selectively cleared to control the starting point of operation.
- Supports replay during debugging: delete events after the point where a problem occurred, modify the code and restart, and replay that moment until it is fixed.
- You can have the agent implement a feature, then after completion restart and tell it that the feature has been loaded, and then ask it to try it.
How to use
- Environment requirement: at least Node 10.
- Rename the .env.template file in the project root directory to .env and add keys.
- Enable the Google Custom Search API for a Google Cloud account, for example by visiting https://console.cloud.google.com/apis/library/customsearch.googleapis.com。
- Start the program.
- Interact with the agents through the console; any input content will be sent as a message to all current agents.
Applicable scenarios and limitations
- Applicable scenarios: exploratory development and debugging that require multiple agents to collaborate on tasks.
- Limitations and cautions: newly started agents often make mistakes and produce errors, for example trying to use actions before requesting help to understand the parameters, using raw text responses instead of actions in the correct format (or wrapping a code block containing valid actions in raw text), or using multi-line argument values without wrapping them with multi-line delimiters ( % ff9d7713-0bb0-40d4-823c-5a66de48761b ). This is a normal adjustment period when they are learning to operate themselves, and they can usually recover from errors, but the gpt-3.5-turbo agent sometimes falls into an endless error loop and cannot locate the problem. It is strongly recommended not to leave it unattended, because such infinite loops will quickly consume tokens, especially when stuck on high-token-consumption actions such as writeFile.
- State management: each agent's state is stored in the .store directory. For example, Agent 1 has its own storage. You can selectively clear state by deleting any file, an entire agent folder, or the entire .store; otherwise, after restarting, the agent will continue from where it left off last time.
Information not yet provided
The project does not provide specific information such as fees, pricing, platform support scope, or performance commitments.
