SafeAICoder

A framework for safety analysis of LLM-generated code in multi-agent systems.

Requirements

• python3
• docker
• docker-compose-plugin
• ollama

How to run locally

• Run make up to start all services at once
  • Put your Ollama API token in OLLAMA_API_KEY env variable before running services
  • You'll also need to pull models from inside ollama container
  • If you are using WSL 2 on Windows with systemd disabled, start docker service manually if necessary

$ export OLLAMA_API_KEY="<your token goes here>"
$ ollama pull qwen3-coder:480b-cloud  # or any other model of your preference
$ sudo service docker restart

• Run make down to tear down all services at once