Install Project Osmos in GitHub Copilot CLI, Codex, or Claude Code and use natural language to run data engineering tasks in Microsoft Fabric.
Project Osmos is an AI-powered data engineering workflow for Fabric. Describe the outcome you want, and the agent can:
- Explore data in your Lakehouse.
- Create and test Fabric notebooks.
- Clean, join, aggregate, and validate data.
- Create or update Delta tables and workspace artifacts.
- Run longer workflows while you monitor progress from a local dashboard.
The plugin on your machine starts and monitors the task. Agent reasoning, Spark execution, and OneLake data access run inside Microsoft Fabric.
You need:
| Requirement | Details |
|---|---|
| Project Osmos access | Project Osmos must be enabled for your account. |
| Fabric workspace | The workspace must be assigned to Fabric capacity and contain a Lakehouse. |
| Workspace permissions | Contributor or higher on the target workspace. |
| Fabric Copilot setting | User can use Copilot and other features powered by Azure OpenAI must be enabled for the tenant or workspace. |
| Azure CLI | Install Azure CLI and sign in with an identity that can access the workspace. |
| AI coding client | Install GitHub Copilot CLI, Codex, or Claude Code. |
Use the commands for your preferred client.
copilot plugin marketplace add microsoft/project-osmos
copilot plugin install project-osmos@project-osmoscodex plugin marketplace add microsoft/project-osmos
codex plugin add project-osmos@project-osmosclaude plugin marketplace add microsoft/project-osmos
claude plugin install project-osmos@project-osmosRestart the client after the first installation. In Claude Code, you can instead run /reload-plugins.
az login
az account showConfirm that az account show displays the expected account and tenant.
| Client | Command |
|---|---|
| GitHub Copilot CLI | copilot |
| Codex | codex |
| Claude Code | claude |
Use a prompt that names Project Osmos and describes the goal:
Use Project Osmos to transform data in my Fabric lakehouse.
The skill asks for:
- The full browser URL of your Fabric Lakehouse.
- Your complete data engineering instruction.
- Any optional constraints or additional context.
- Confirmation of the recommended operating settings before the run starts.
Be specific about source data, transformations, outputs, and validation. For example:
Load Orders and Customers from my lakehouse, remove Orders rows with missing
customer IDs, join the tables, calculate monthly revenue by customer segment,
write the result as a Delta table, and validate the source and output row counts.
Another example:
Create a sales sample CSV in a workspace folder, create a matching Delta table,
and build and test a notebook that ingests the CSV into the table.
Project Osmos runs in Fabric and complex tasks can take up to a day. After the task starts, the plugin displays the task ID and creates a local dashboard under:
./.dataprojects/<task-id>/
You can close the terminal or shut down your machine without stopping the Fabric task. When you return, reopen your client and ask it to resume the existing Project Osmos task. Include the task ID if the client cannot identify the previous run.
Copilot CLI and Claude Code include a startup update check. Codex updates configured Git marketplaces in the background. You can also update manually.
copilot plugin marketplace update project-osmos
copilot plugin update project-osmos@project-osmoscodex plugin marketplace upgrade project-osmosclaude plugin marketplace update project-osmos
claude plugin update project-osmos@project-osmosRestart the client after an update so the new skill content is loaded.
| Issue | What to do |
|---|---|
| Project Osmos is not available after installation | Restart the client, then run the install commands again and check for an installation error. |
| Azure authentication fails | Run az login, then verify the expected account and tenant with az account show. |
| Workspace or capacity lookup fails | Confirm the Lakehouse URL is correct, the workspace has Fabric capacity, and your Azure identity can access it. |
| Lakehouse lookup or task creation fails | Open the Lakehouse in Fabric and copy the complete browser URL again. |
| The task runs for a long time | Spark startup, planning, and complex transformations can take time. Continue monitoring the existing task rather than creating another one. |
| The local dashboard or poller stopped | Resume the existing task with the same task ID. Do not restart intake or create a duplicate task. |
| The installed plugin appears outdated | Run the manual update command for your client, restart it, and retry. |
skills/project-osmos/contains the Project Osmos skill and operational references.- The client marketplace manifests make the same plugin available to Copilot CLI, Codex, and Claude Code.
- Copilot CLI uses
hooks.jsonfor its startup update check; Claude Code carries its native startup hook in its marketplace manifest. .github/workflows/validates and publishes the plugin.
Maintainer and contribution details are in CONTRIBUTING.md.
- Read PRIVACY.md for privacy and telemetry information.
- For general feedback, contact project-osmos@microsoft.com.
- Report security vulnerabilities through SECURITY.md.
- Do not post tokens, tenant details, workspace or Lakehouse IDs, certificate material, or private data in issues or logs.
This project is licensed under the MIT License and follows the Microsoft Open Source Code of Conduct.
Microsoft, Microsoft Fabric, GitHub Copilot, OneLake, and Azure may be trademarks or registered trademarks of Microsoft Corporation. See the Microsoft Trademark and Brand Guidelines.