# Hybro

> Open-source platform that connects local and remote AI agents over the A2A protocol so they can collaborate with human approval.

Hybro, from Hybro AI, is an open-source agent interoperability platform released under the Apache License 2.0. It lets local and remote AI agents communicate through the Agent2Agent (A2A) protocol, with a web app for chatting with agents, inspecting them, and approving their work. It runs as a single Docker container started by a `hybro` command.

## What It Is

Hybro is a hybrid multi-agent orchestration engine and dashboard. Users register A2A agents, group them into subnets, and start chats where they mention specific agents or let Hybro pick from all available ones. Replies stream live, and the interface shows which agents were called. The backend is an async FastAPI service with MongoDB persistence, and the frontend is a Next.js application.

## How Collaboration and Oversight Work

A single message can be handed to several agents, and Hybro brings their answers together. Chats can use a Fast mode for simple questions or an Ultimate mode for larger tasks that need planning. When an agent needs more information or approval, the chat pauses and waits for the user to answer, approve, or reject. An Inbox collects sessions that need attention, including sessions opened by external clients. The Network page maps registered agents and the subnets that group them, and an A2A Agent Inspector helps test agent capabilities.

## Connecting Agents and Clients

The project's A2A Adapter is an open-source Python SDK and CLI that wraps agents such as Claude Code, Codex, LangGraph, and CrewAI so they join the network as A2A agents. Hybro also exposes an MCP gateway over Streamable HTTP, and the Access page provides setup snippets for clients such as Claude Code, Cursor, Codex, and VS Code. The MCP gateway has no authentication and is served only on loopback listeners. Five sample agents are bundled: Echo, Weather, Story, Travel Planner, and Image Generator.

## Setup Path

Install with a shell script or npm, then run `hybro`. Docker is required on amd64 or arm64. A model provider (OpenAI, Anthropic, or DeepSeek) is configured in Settings; Echo Agent and MCP work without one. Configuration lives in `~/.hybro/`. A source checkout can be built and run alongside a released install, and `hybro upgrade` handles updates and migration from 0.3.

## Features
- Hybrid execution of local and remote agents
- Native A2A protocol interoperability
- Multi-agent execution rooms
- Live SSE streaming and turn timelines
- A2A Agent Inspector
- Human approval and input requests
- Inbox for sessions needing attention
- Network topology and subnets
- Built-in MCP gateway
- Single-container deployment
- Bundled sample agents
- Model provider setup in browser (OpenAI, Anthropic, DeepSeek)

## Integrations
Claude Code, Codex, LangGraph, CrewAI, LangChain, n8n, OpenClaw, Ollama, Hermes, Cursor, VS Code, OpenAI, Anthropic, DeepSeek, MCP, A2A

## Platforms
MACOS, LINUX, WEB, API, VSC_EXTENSION, CLI

## Pricing
Open Source

## Version
v0.4.4

## Links
- Website: https://hybro.ai
- Documentation: https://www.hybro.ai/docs
- Repository: https://github.com/hybroai/hybro
- EveryDev.ai: https://www.everydev.ai/tools/hybro