omnigent

Omnigent is an open-source AI agent framework and meta-harness that provides a common orchestration layer over various coding agents, including Claude Code, Codex, Cursor, OpenCode, Hermes, Pi, and custom agents. It allows users to swap or combine harnesses without rewriting code, enforce policies and sandboxing, and collaborate in real time from any device. Key capabilities include: - Cross-device session continuity: Sessions sync across terminal, browser, phone, and a native desktop app. - Multi-agent supervision: Mix different agents in the same session, allowing them to review each other's work or split tasks based on their strengths. - Model flexibility: Supports first-party API keys, Claude/ChatGPT subscriptions, or any compatible gateway (OpenRouter, LiteLLM, Ollama, vLLM, Azure, Databricks). - Team collaboration: Share live sessions, co-drive agents on a host machine, or fork conversations. Supports multi-user accounts with invite-only access and OIDC authentication (Google, GitHub, Okta, Microsoft). - Cloud sandbox execution: Run sessions in disposable sandboxes across providers like Modal, Daytona, E2B, Kubernetes, CoreWeave, and Databricks. - Policy governance: Create rules to pause for approval before risky actions, cap spending, or limit tool access. Policies stack across server-wide, per-agent, and per-session levels. Custom agents are defined in YAML files, specifying prompts, tools (local Python functions, MCP servers), and sub-agents. The framework supports deployment via Docker Compose, Render, Railway, Fly.io, Hugging Face Spaces, Cloudflare, and Databricks Apps. It runs on macOS, Linux, and Windows (with some feature limitations on Windows), requiring Python 3.12+. The project is licensed under Apache 2.0 and is currently in alpha status.