Skip to main content
Follow the steps below to go from zero to your first training run on the Osmosis Platform.

​
Step 1: Create Your Account

  1. Go to platform.osmosis.ai and sign up
  2. Create a new workspace (or join an existing one via team invite)
  3. You’ll land on the workspace dashboard

​
Step 2: Install the SDK

Install the Osmosis Python SDK and authenticate: 
pip install osmosis-ai
osmosis login
The osmosis login command opens your browser to authenticate with the platform and stores your CLI token locally.

​
Step 3: Create a Project

From your workspace dashboard:
  1. Click New Project
  2. Give it a name and description
  3. Choose your rollout mode — this determines how you’ll connect your code:
    • Local Rollout: Connect a GitHub repository with reward functions and tools
    • Remote Rollout: Point to a rollout server you host

​
Step 4: Choose Your Path

Local Rollout

Push reward functions, rubrics, and MCP tools to GitHub — Osmosis syncs and runs everything automatically.

Remote Rollout

Build and host your own agent server with full control over rollout logic and tool execution.

​
Step 5: Configure Training

Once your code is connected (via Local Rollout or Remote Rollout):
  1. Navigate to Training Runs in your project
  2. Click New Training Run
  3. Select your base model, dataset, reward functions, and tools
  4. Configure hyperparameters (or use defaults)
  5. Click Start Training
See Training Runs for detailed configuration options.

​
Step 6: Monitor and Export

While training runs:
  • Watch real-time metrics on the Monitoring dashboard
  • Review training logs and checkpoints
  • When satisfied, merge a checkpoint and export to Hugging Face

​
Next Steps

Training Runs

Configure models, datasets, rewards, and hyperparameters

Python SDK

Install the SDK for building reward functions and agent loops