Databricks' Omnigent puts Claude Code, Codex, Cursor and Pi under one control layer
Omnigent, an open-source meta-harness from Databricks, orchestrates multiple coding agents with shared policies, sandboxes and live session sharing.
Omnigent is an open-source "meta-harness" that runs Claude Code, Codex, Cursor, OpenCode, Hermes, Pi and custom YAML-defined agents behind one orchestration layer. If you already bounce between two or three coding agents and copy context by hand, this is aimed at you. The repo says you can mix harnesses in one session, have one agent review another's work, set policies that pause for approval, cap spend or limit tools, and run sessions in cloud sandboxes such as Modal, E2B, Daytona or Kubernetes.
Databricks is behind it. Its AWS docs describe a managed version tied to workspace identity and Databricks Sandboxes, gated behind a workspace preview. Per Efficiently Connected, Databricks says its own engineers were running four to five agents at once and copy-pasting output between tools. That report also notes that roadmap items such as automatic optimization are not shipped yet.
The open-source install needs Python 3.12+, Node 22+ and tmux, plus bubblewrap on Linux. Native Windows runs in a degraded mode without filesystem or network sandboxing. The bundled Polly agent splits a task across worktrees and sends each diff to a reviewer from a different vendor, as a hands-on walkthrough shows. No source shows that cross-agent review catches more bugs than a single agent.
The main risk is the adapters. A 0.16.0 bug report describes Codex agents failing in network-denying sandboxes, while the Claude harness handles the same case. How quickly fixes land when vendor CLIs change is unclear, and so is any long-term maintenance commitment for the open-source version. For now, treat it as a team tool and expect setup overhead if you work solo.
Sources
- GitHub - omnigent-ai/omnigent: Omnigent is an open-source AI agent framework and meta-harness: orchestrate Claude Code, Codex, Cursor, Pi, and custom agents — swap harnesses without rewriting, enforce policies and sandboxing, and collaborate in real time from any device. · GitHub
- Omnigent on Databricks | Databricks on AWS · docs.databricks.com
- Omnigent: Databricks Open Sources an AI Agent Meta-Harness - Efficiently Connected · Efficiently Connected
- Part 1: A practical introduction to Omnigent · Databricks Decoded
- [Bug] 0.16.0: Codex agents with a network-denying sandbox cannot start with a ChatGPT subscription login · Issue #8737 · omnigent-ai/omnigent · GitHub