osmosis-ai 0.2.18

A Python SDK for Osmosis LLM training workflows: reward/rubric validation, rollout, platform CLI, and evaluation tools.

osmosis-ai

⚠️ Warning: osmosis-ai is still in active development. APIs may change between versions.

Python SDK for Osmosis AI, a platform for training LLMs with reinforcement learning. Implement an AgentWorkflow in Python, add a concrete Grader for serve/eval flows, drive them locally with the CLI, then connect a RolloutServer to managed training.

Quick start

Step	What you do
Define agents	One AgentWorkflow subclass (+ optional AgentWorkflowConfig) in your repo. For rollout serve, the entrypoint must also expose a concrete Grader (typically with a GraderConfig).
Layout	Use a rollout pack directory under rollouts/<name>/ when loading by rollout name; the CLI adds that directory to sys.path.
Serve	osmosis rollout serve serve.toml — HTTP server for TrainGate. The entrypoint module must expose a concrete Grader (typically with a GraderConfig); config is TOML ([serve], [server], [debug]).
Evaluate	osmosis eval run eval.toml — same execution stack as training, with optional pass@k and caching.

Example repositories: osmosis-git-sync-example (synced agent repo patterns) · osmosis-remote-rollout-example (reference server usage — align with current SDK exports when upgrading).

Documentation index: docs/README.md

Installation

Requires Python 3.12+. For development setup, see CONTRIBUTING.md.

• An LLM API key (e.g., OpenAI, Anthropic, Groq) — required for osmosis eval run when using hosted models. See supported providers.
• Osmosis account (optional) — needed for osmosis auth login, workspace management, and platform-backed commands such as datasets, models, and training runs. Sign up at platform.osmosis.ai.

pip

pip install osmosis-ai            # Core SDK
pip install osmosis-ai[server]    # + FastAPI rollout server (pulls in platform extra)
pip install osmosis-ai[full]      # Same as [server] (all packaged optional features)

uv

uv add osmosis-ai                 # Core SDK
uv add osmosis-ai[server]         # + FastAPI rollout server (pulls in platform extra)
uv add osmosis-ai[full]           # Same as [server] (all packaged optional features)

Testing and evaluation

• Eval — graded runs, pass@k, cache/resume with osmosis eval run
• CLI reference

Contributing

See CONTRIBUTING.md for development setup, testing, linting, and PR guidelines.

License

MIT License - see LICENSE file for details.