Overview
What is Omnigent?
Omnigent brings Codex, Claude Code, Cursor, Pi, and other coding agents together in one interface. Run agents on your machine or in the cloud, where they can keep working even when your laptop is off. Collaborate with teammates and follow the work from your browser, desktop app, or phone.
Omnigent on Databricks connects this experience to your workspace identity and model access through Foundation Model APIs and Unity Gateway.
Omnigent on Databricks is in Beta. For detailed guides and reference, see the open-source Omnigent documentation.
When to use it
- Work beyond the terminal. Read conversations, inspect code changes, and steer agents from a browser or desktop app.
- Use multiple coding agents in one place. Bring Codex, Claude Code, Cursor, Pi, and 10+ other harnesses into a single interface. Choose whichever agent fits each task, switch between agents as you work, and manage all their sessions together.
- Collaborate on a live session. Share a session with teammates so they can follow the work or help drive it.
- Take your agents anywhere. Start a session on your laptop, then check progress and send instructions from your phone using the mobile app.
Omnigent is an open-source meta-harness: a common layer over the runtimes that execute agents. It adds shared tools, policies, and collaboration across supported harnesses. See the Omnigent documentation for the full feature set.
Build apps with Omnigent
Use Omnigent while building your application: bring a repository, describe a task, and work with agents on the changes. A DevHub template can provide the starting instructions for an app that uses Databricks Apps, Lakebase, or Agent Bricks.
Databricks operates the managed Omnigent server, which coordinates your sessions using your workspace identity. You choose the host where the agent executes commands and accesses files:
| Host | When to choose it |
|---|---|
| Databricks Sandbox | Use a managed cloud environment that keeps running when your laptop is off. Available in supported AWS, Azure, and GCP regions; requires the Sandbox preview. Model access goes through Unity Gateway. |
| Your laptop or VM | Use files, tools, or networks available on your own machine. Register it with the Omnigent CLI and keep the machine powered on with the host process running. |
Both choices use the same managed server and workspace UI. Follow the Omnigent quickstart for host setup and requirements. Review the Sandbox limitations to ensure it works for your use case.
Get started
Your workspace needs the Omnigent preview enabled and a region that supports Unity Gateway. Databricks Sandbox has additional preview and region requirements. See availability and limitations.
After checking the prerequisites in the quickstart:
- Open
<workspace-url>/omnigentin your browser. - Select New session.
- Choose Sandbox, if available, or a machine you have registered as a host.
- Describe the task you want the agent to work on.
Where to next
- Programmatic usage explains how to run tasks from scripts with Databricks authentication.
- Identity and access explains sign-in, session visibility, and sharing permissions.
- Supported harnesses lists the open-source project's agent runtimes and their capabilities.
- Custom agents explains how to define an agent's instructions, tools, harness, and policies.
- Managed limitations describes which policies and model credentials are supported on Databricks.