omnigent

Omnigent puts one layer over all your AI coding agents so you can swap between them without rewriting anything. Claude Code, Codex, Pi, your own custom agents, it runs them through a single harness with shared policy controls, so the agent choice becomes a config detail instead of a rewrite. The standout feature is a security sandbox that hides your credentials and locks down filesystem and network access while an agent runs.

Setup is light for what it does: install the CLI through uv or Homebrew, and you get terminal and web UIs plus runners that execute locally or on Modal or Daytona. You bring your own model credentials. The governance and sandboxing are the real reason to look, they address the thing that makes people nervous about letting agents run loose on their machine.

Developers juggling multiple coding agents, or anyone uneasy about an agent seeing their secrets, are the target. If you've settled on one agent and trust it, this is overhead you don't need. The value scales with how much you care about isolation and the ability to switch tools without lock-in.

The catch: it's alpha. The README says so, and that means rough edges, breaking changes, and features still landing. It's free and Apache-licensed, but you're adopting an early project to wrap other early projects. Promising, but not something to build a team's critical workflow on yet.