Skip to main content
The Osmosis AI platform supports automatic synchronization from GitHub repositories. You can store your MCP tools, reward functions, and reward rubrics in your own repository (public or private), and Osmosis will automatically discover and sync them.

Why Use Git Sync?

By connecting your GitHub repository to Osmosis, you can:
  • Version control your evaluation logic alongside your application code
  • Collaborate with your team using familiar Git workflows
  • Automate deployments through CI/CD pipelines
  • Keep repositories private while still syncing with Osmosis

What Can You Sync?

Osmosis automatically discovers and syncs three types of components from your repository:
  1. MCP Tools - Extend AI agent capabilities with custom functions
  2. Reward Functions - Provide deterministic, numeric scoring for LLM outputs
  3. Reward Rubrics - Use LLMs to evaluate outputs with natural language criteria

Example Repository

Check out our reference implementation: osmosis-git-sync-example This example demonstrates the complete folder structure and code artifacts that the Osmosis GitHub integration discovers and syncs.

How It Works

  1. Structure your repository with the required folders
  2. Implement your functions using the Osmosis AI decorators
  3. Push to GitHub (public or private)
  4. Connect your repository to the Osmosis platform
  5. Sync automatically - Osmosis discovers and deploys your components

When to Use Git Sync

Choose Git Sync when you want to:
  • Version control evaluation logic alongside application code
  • Collaborate with your team using Git workflows
  • Deploy automatically through CI/CD
  • Keep code private while using the Osmosis platform
  • Manage multiple rubrics and functions in one place
For quick local testing and development, use the Python SDK directly instead.
1

New to Osmosis?

Start with the Python SDK Introduction to understand core concepts, then return here to learn about Git Sync deployment.
2

Ready to Set Up?

Follow the Folder Structure guide to organize your repository correctly.
3

Building Components

Implement MCP Tools, Reward Functions, or Reward Rubrics based on your needs.
4

Best Practices

Review Best Practices for production-ready implementations.

Next Steps