Live runs come from a small demo server on your own machine. It starts each experiment
in Docker, with the paper's settings, and this page shows the run as it happens.
What you need
- Linux with Docker, able to run privileged containers (the network is emulated with Mininet).
- The keys for the models you want to use. They stay on your machine: the experiment
container only gets a throwaway token for a proxy the server runs next to it.
- For Qwen, a vLLM server serving it.
Start the server
git clone https://github.com/fchamicapereira/agentic-networks
cd agentic-networks
export ANTHROPIC_API_KEY=... # Claude
export OPENAI_API_KEY=... # GPT
export TOGETHER_API_KEY=... # GLM
python3 demo/server.py --port 8800 --vllm-host <vLLM server>
Then come back here, choose Live, and connect to the same port. Your browser may ask
whether this page may access devices on your local network: allow it, since the server is
on your machine.
The server checks which models answer when it starts (a tiny request to each; Recheck
does it again). Runs use the models' APIs and cost what they cost.