AgentJet

Run `npx skills add modelscope/AgentJet --all -g -y` in your terminal to install all skills in this suite globally.

AgentJet is a reinforcement learning framework that fine-tunes the models behind your agent workflows. It supports distributed swarm training, so you can launch training from a laptop while GPU servers do the heavy lifting.

Yes. Start the swarm server on a remote GPU machine with `ajet-swarm start`, then run a lightweight client script from any laptop to drive the training.

It works with agents built on the OpenAI SDK, AgentScope, LangChain, or plain HTTP requests, as long as the agent can call an OpenAI-compatible endpoint.

Use the included tmux monitoring skill to watch training logs at increasing intervals, detect errors early, and get fix suggestions without sitting in front of the console.