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.

Osmosis Platform is the web dashboard for managing everything related to reinforcement learning training of LLMs. It handles GPU provisioning, training orchestration, metrics collection, and model management — so you can focus on defining agent behavior and evaluation logic.

Core Capabilities

Workspaces

Organize your team, projects, and resources in isolated workspaces with role-based access control.

Training Runs

Submit, monitor, and manage RL training runs with configurable hyperparameters and strategies.

Datasets

Upload and validate JSONL, CSV, or Parquet datasets up to 5 GB for training.

Models

Import base models from HuggingFace and manage trained checkpoints.

Monitoring

Track real-time training metrics, reward trends, and checkpoint progress.

Git Integration

Connect your GitHub repository to sync AgentWorkflows, Graders, and tools automatically.

How It Works

The typical workflow from setup to trained model follows four stages:
1

Initialize your workspace

Use the CLI to create a workspace and initialize your local project directory.
osmosis workspace create my-workspace
osmosis init my-project
2

Define your AgentWorkflow and Grader

Write your agent behavior in an AgentWorkflow and define evaluation logic in a Grader. Push your code to GitHub.
3

Sync and submit training

The platform syncs your code via Git Integration. Upload a dataset and submit a training run:
osmosis dataset upload data/train.jsonl
osmosis train submit configs/training/default.toml
4

Monitor results and export

Track training progress in real time on the platform dashboard. When training completes, export your trained model to HuggingFace.

Ready to Get Started?

Platform Quickstart

Go from zero to your first training run in minutes.