omnigent-ai/omnigent

Omnigent is a tool that sits on top of several AI coding assistants and gives you one common way to work with all of them. It supports Claude Code, Codex, Pi, and agents you write yourself, and it lets you swap between them or combine them without rewriting your setup. The README describes it as a single control layer that sits over and manages other AI agents. It is currently in an alpha stage, so it is early software. One of its main ideas is that a working session follows you across devices. You can start a task in your terminal, continue it in a web browser, and pick it up on your phone, with messages, files, and sub-agents staying in sync. It also lets you supervise more than one agent at once, for example asking one agent to review the work of another, or splitting a job across agents that are each good at different things. Omnigent works with any model through four kinds of credentials: a direct vendor API key, a Claude or ChatGPT subscription, any compatible gateway such as OpenRouter or Ollama, or a Databricks workspace. You can set a default model per agent and switch models in the middle of a session. Other features include real-time collaboration, where teammates can join a shared session to watch or co-drive the agent, and the option to run agents in disposable cloud sandboxes from providers like Modal and Daytona, so you do not need your own laptop running. There is also a policy system for governing agents: you can pause for approval before risky actions, cap spending, or limit which tools an agent can reach. Installation is a single command, and the project needs Python 3.12 or newer plus a few other tools. Two example agents ship with it: Polly, an orchestrator that plans and delegates coding work to sub-agents and routes each result to a reviewer from a different vendor, and Debby, a brainstorming partner that sends every question to both a Claude and a GPT model and shows the answers side by side. The project is released under the Apache 2.0 License.