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.

​
What is an Osmosis Workspace?

An Osmosis workspace is a local project directory for developing AgentWorkflow and Grader code. It provides a standardized structure for organizing rollouts, configurations, datasets, and AI coding assistant instructions — everything you need to iterate on agent behavior locally before syncing to the platform for training. Workspaces are created by the osmosis init command and connected to the Osmosis Platform via Git Sync.
There are two kinds of “workspace” in the Osmosis ecosystem:
  • Local Workspace — the directory on your machine created by osmosis init. This is where you write and test your AgentWorkflows, Graders, and configs.
  • Platform Workspace — a remote organizational space on the Osmosis Platform for managing datasets, training runs, models, and team members. Managed via osmosis workspace commands (see CLI Installation).
This page documents the local workspace.

​
Getting Started

Create a new workspace with osmosis init: 
osmosis init my-project
cd my-project
This creates a my-project/ directory with the following structure: 
my-project/
├── .osmosis/            # Workspace metadata and AI skills
├── rollouts/            # AgentWorkflow + Grader code
├── configs/             # Training, eval, and serve configs
├── data/                # Local test datasets
├── pyproject.toml       # Python project config
├── AGENTS.md            # AI coding assistant instructions
├── CLAUDE.md            # Claude Code instructions
└── .gitignore
The workspace is also initialized as a git repository with an initial commit, ready to push to GitHub. See Structure & Configuration for a detailed breakdown of every file and directory.

​
The --here Flag

To initialize the current directory as a workspace instead of creating a subdirectory: 
mkdir my-project && cd my-project
osmosis init my-project --here
The current directory must be empty (an existing .git/ directory is allowed).

​
Name Validation

Workspace names must be lowercase alphanumeric characters and hyphens only, with a maximum of 64 characters. Names cannot start or end with a hyphen. 
Valid:    my-project, agent-v2, data-extraction-tool
Invalid:  My-Project, -agent, agent_, my project

​
Next Steps

Write your AgentWorkflow

Define the agent loop that will run during training rollouts.

Write your Grader

Create evaluation logic to score agent outputs for RL training.

Run eval locally

Test your agent against a dataset before submitting to training.

Sync to Platform

Push to GitHub and let the platform sync your rollout code automatically.