Databricks released Omnigent, an open‑source meta‑harness that turns individual AI agents into interchangeable parts of a unified system. Engineers today often juggle Claude Code, Codex, Pi, and other agents, copying outputs between terminals, docs, and Slack. Each harness works in isolation, making composition, control, and collaboration cumbersome.
Omnigent solves this by providing a common interface above command‑line agents and SDKs. It wraps any harness—Claude Code, Codex, Pi, OpenAI Agents, Claude Agents SDK, or a custom YAML agent—so the user‑facing API is the same: messages and files go in, text streams and tool calls come out. This lets you swap models or harnesses with a one‑line change instead of re‑integrating each tool.
A runner creates a sandboxed session with a uniform API, while a server supplies policies and sharing. Starting a session launches a terminal command and a local web UI at localhost:6767; the same session appears in the browser or on a mobile device, keeping terminals, files, sub‑agents, and chat in sync. The CLI installs as either omnigent or omni and detects existing model credentials on first run.
Composition means combining models, harnesses, and techniques without rewriting code. Control is enforced at the meta‑harness layer through stateful policies—e.g., pausing an agent after a $100 spend or requiring human approval before a git push that follows an npm install. Collaboration lets teammates share live agent sessions via URL, watch, comment, co‑drive, or fork the conversation, eliminating copy‑pasting between tools.
An OS sandbox (Omnibox) hides secrets like GitHub tokens, injecting them only through an approved egress proxy. Policies are defined in YAML and stack across server, agent, and session levels, with stricter session rules evaluated first.
Getting started requires Python 3.12+, Node.js 22 LTS, and tmux. One command installs everything, after which you configure model credentials and run a custom agent YAML file with a single omnigent run command.
Omnigent is Apache 2.0 licensed, runs locally or on disposable cloud sandboxes (Modal, Daytona), and targets platforms like Fly.io, Railway, and Render.
#AI #DevTools #AgentFramework #Omnigent #MLOps #Collaboration