Skip to content

osmosis-ai/osmosis

v0.1.1MIT

Evaluate an AI agent on your own examples, compare models on benchmarks, and train and deploy a checkpoint in your Osmosis workspace.

Osmosis

Evaluate an AI agent on examples from your project, compare it with a baseline, and use the results to decide whether to improve the data, change the agent, or train a model.

In Claude Code with the Osmosis marketplace configured, install with:

/plugin install osmosis@osmosis

Try it: “I'm new to Osmosis. Check what I already have and guide me through my first evaluation.” If you already have results, try “Compare my existing evaluation results and recommend what to try next.”

Try the demo: “Show me the bundled Osmosis sample evaluation. Which support-ticket agent would you choose, and why?” Compare five synthetic tickets without installing the CLI, signing in to Osmosis, or running a model evaluation.

You can plan an experiment or analyze local files without an account. Native installs include skills and a connection to the public Osmosis documentation; the connection searches documentation and accepts documentation feedback. It does not access your workspace. For execution, the agent helps you set up the CLI and, for managed runs, a workspace. Model calls, sandboxes, and managed runs can incur charges.

Osmosis · Open your workspace · Documentation and support · Contact Osmosis