omnigent

Omnigent calls itself an open-source meta-harness: a common orchestration layer over agent harnesses such as Claude Code, Codex, Cursor, OpenCode, Hermes, Pi and agents you define yourself. The README says you can swap or combine harnesses without rewriting, enforce policies and sandboxing, and work with the same session from a terminal, a browser, a phone or a native desktop app, with messages, sub-agents, terminals and files kept in sync.

Its pitch has five parts. You can supervise several agents in one session and ask one to review another's work, with custom agents defined in YAML. You can use any model through a first-party API key, a Claude or ChatGPT subscription or a compatible gateway. Sessions can be shared so teammates watch live, co-drive or fork the conversation. Agents can run in disposable cloud sandboxes from Modal, Daytona, Blaxel, E2B, CoreWeave, Kubernetes, Databricks and others, launched from the CLI or provisioned by the server per session. Policies can pause for your approval before risky actions, cap spend or limit which tools an agent can reach, at server, agent or chat level.

Install with a one-line script (`curl ... | sh`), with `uv tool install omnigent` or pip on Python 3.12+, or with Homebrew; optional extras add providers such as Bedrock and Vertex, sandbox providers and storage. A macOS desktop app is offered. It is Apache-2.0 licensed and the last push was 2026-10-15. This review is based on the README and repository data; we have not run it ourselves.