Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.osmosis.ai/llms.txt

Use this file to discover all available pages before exploring further.

This guide walks you through the full path from installation to a running training job on the Osmosis Platform.
1

Install the CLI

Install the Osmosis CLI via pip:
pip install osmosis-ai
See CLI Installation for detailed setup instructions and requirements.
2

Authenticate

Log in to connect the CLI to your Osmosis account:
osmosis auth login
This opens your browser to authenticate and stores your CLI token locally.
3

Create or switch workspace

Create a new workspace for your team, or switch to an existing one:
osmosis workspace create my-workspace
4

Initialize your local workspace

Set up the local directory structure that the platform expects:
osmosis init my-project
This creates the standard workspace layout with directories for AgentWorkflows, Graders, configs, and datasets. See Workspace Overview for details on the directory structure.
5

Write your AgentWorkflow and Grader

Define your agent’s behavior in an AgentWorkflow and your evaluation logic in a Grader:
# rollouts/my-rollout/main.py
from osmosis_ai.rollout import AgentWorkflow, AgentWorkflowContext, Grader, GraderContext

class MyWorkflow(AgentWorkflow):
    async def run(self, ctx: AgentWorkflowContext) -> None:
        response = await litellm.acompletion(
            model="openai/gpt-5.2",
            messages=ctx.prompt,
        )
        return response

class MyGrader(Grader):
    async def grade(self, ctx: GraderContext) -> None:
        for sample_id, sample in ctx.samples.items():
            last_msg = sample.messages[-1]["content"] if sample.messages else ""
            reward = 1.0 if ctx.label and ctx.label.strip() in last_msg else 0.0
            ctx.set_sample_reward(sample_id, reward)
See Rollout Overview for the full API reference and Command Reference for all available CLI commands.
6

Upload a dataset

Upload a JSONL, CSV, or Parquet file with the required columns (system_prompt, user_prompt, ground_truth):
osmosis dataset upload data/train.jsonl
7

Connect GitHub and sync

Install the Osmosis GitHub App on your repository to enable automatic syncing of your AgentWorkflows, Graders, and configs. See Git Sync for setup instructions.
8

Submit a training run

Submit your training run using a TOML configuration file:
osmosis train submit configs/training/default.toml
The platform provisions GPUs, syncs your code, and begins training automatically.
9

Monitor results

Track training progress in real time:
osmosis train status my-run
osmosis train metrics my-run
Or open the Osmosis Platform dashboard to view metrics, reward trends, and checkpoints visually.

​
Next Steps

Training Runs

Learn about training configuration, statuses, and management.

Datasets & Models

Understand dataset formats, validation, and model management.