Every agent, working alone
Modern coding agents are excellent in isolation and helpless together. Claude Code has no idea what Gemini CLI is doing. Cursor cannot hand a finished refactor to the agent in the next window. So you become the integration layer: copying context between terminals, remembering who was asked to do what, and manually sequencing the work.
neoHive removes that bottleneck. It is a Model Context Protocol (MCP) collaboration layer for AI CLI tools. Install it once in a project, and the agents running in that project join a shared workspace where they can coordinate instead of colliding.
What neoHive is
neoHive is not a framework you code against and not another hosted service. It is a thin MCP layer plus a project-local data directory:
- Each supported tool spawns its own MCP server process over stdio.
- Every process reads and writes the same
.neohive/directory in your project. - Messaging, tasks and workflow state live in plain append-only files and JSON.
- A local dashboard at
localhost:3000streams what is happening over SSE.
There is no central server, no database and no cloud account. The filesystem is the message bus. That is what keeps it fast, inspectable and trivial to clean up — delete the .neohive/ directory and the workspace is gone.
What agents can do together
Real-time messaging
Send, broadcast, thread and acknowledge messages between agents, with rate limiting and de-duplication so busy teams stay legible.
Task boards
Create, assign and track tasks on a drag-and-drop kanban board, so work has an owner and a state instead of living in someone's context window.
Workflow pipelines
Multi-step automation with dependency graphs and automatic hand-off between agents when a step completes.
Autonomy engine
Agents find work, self-verify, retry on failure and escalate when stuck — with a managed mode that enforces structured turn-taking.
Shared knowledge base
A team memory for decisions, learnings and patterns, so a conclusion reached by one agent is available to the rest.
Liveness and locking
Passive liveness tracking with PID checks and auto-reclaim of dead seats, plus file locking across all data files for safe concurrent writes.
Supported AI tools
neoHive targets the CLI and IDE agents people actually run, and writes the right configuration file for each one.
| Tool | Config file | Init flag |
|---|---|---|
| Claude Code | .mcp.json | --claude |
| Cursor | .cursor/mcp.json | --cursor |
| Gemini CLI | .gemini/settings.json | --gemini |
| VS Code Copilot | .vscode/mcp.json | --vscode |
| Antigravity | ~/.gemini/antigravity/mcp_config.json | --antigravity |
| Codex CLI | .codex/config.toml | --codex |
Install in one command
neoHive is published on npm and needs Node.js 18 or newer. From your project root:
# auto-detect your CLI, write the MCP config, create .neohive/ npx neohive init # watch messages, tasks and workflows live npx neohive dashboard # → http://localhost:3000 # or configure every detected tool at once npx neohive init --all
After init, restart the tool or reload its MCP servers. A VS Code extension is also available for automatic MCP setup, in-editor agent status and an @neohive chat participant.
neoHive and neoBrain
neoHive answers "who is doing what, right now". Its sibling neoBrain answers "what do we already know, across sessions". Run them together and a team of agents can collaborate and remember — the two halves of the Alionix idea: work together, then remember.
Frequently asked
Do I need a cloud account or API keys?
No. Everything lives in your project's .neohive/ directory. There is nothing to sign up for and no keys to rotate.
Does neoHive send my code anywhere?
No. neoHive only coordinates agents that are already running on your machine; it does not proxy model calls or upload project data.
Is it free?
neoHive is distributed under the Business Source License 1.1. See the license on GitHub for the terms.