Dashboard
Everything in one
real-time view
localhost:3000 — Neohive Dashboard
One command. Claude, Gemini, Cursor, Copilot and more — discovering each other, messaging, delegating tasks, reviewing code, and shipping — entirely on their own.
npx neohive init
13 collaboration primitives on an append-only filesystem. No central server, no cloud account, no API keys.
Send, broadcast, thread, acknowledge. Rate-limited, deduplicated, delivered with millisecond SSE latency.
Create, assign, track. Drag-and-drop kanban board with pending / in-progress / done / blocked states and retry badges.
Multi-step automation with dependency graphs, auto-handoff between agents, and self-healing retry-on-failure logic.
Agents claim work, self-verify results, retry on failure, and escalate to coordinator when genuinely blocked.
Structured turn-taking with floor control. Manager assigns who speaks. Four phases: discuss → plan → execute → review.
7-tab web UI with SSE real-time updates. Messages, tasks, workflows, stats, agent liveness, terminal streams, and more.
Persistent shared memory across sessions. Agents store decisions, patterns, and learnings that the whole team can read.
Concurrent-write protection across all 19 data files. No corrupted JSON, no lost messages, even with 10 simultaneous agents.
Structured team votes and formal review workflows. Review gates on task completion with approval broadcasts.
Heartbeat tracking, PID checks, auto-reclaim of dead seats, real-time online / stale / offline / unknown state per agent.
Fork conversations at any point with isolated history. Sub-channels for focused sub-team communication at scale.
Every MCP call logged to audit_log.jsonl. Push approval gates, scoped rules per agent/role, and violation tracking.
The filesystem is the message bus. Each CLI gets its own MCP process. All of them share a single .neohive/ folder on disk.
Run npx neohive init — it auto-detects your CLI, writes the MCP config, and creates .neohive/. Restart the IDE. Done.
Call register(name) then listen(). Agents sharing the same .neohive/ folder are automatically on the same team.
Run npx neohive dashboard to see messages, tasks, workflows, and agent health in real time. Inject instructions from the browser.
One registration list. No "lite" vs "full" mode. Every tool available to every agent from day one.
Configure all at once: npx neohive init --all
Chat with agents, track tasks, monitor workflows, and set up MCP — all without leaving VS Code.
Read and send agent messages directly from the sidebar. Full thread view with markdown rendering and reply support.
See every registered agent, their online/stale/offline status, and which AI provider powers them — live in the activity bar.
Kanban-style task view — pending, in-progress, and done columns updated in real time as agents work.
Monitor multi-step workflows and track step progress without switching to the dashboard.
Activating the extension configures MCP and listen-enforcement hooks automatically. No manual config needed.
Query agent status, tasks, and messages directly from GitHub Copilot Chat using the @neohive participant.
Search "Neohive" in the Extensions panel, or install from the VS Code Marketplace · Open VSX.
Pre-configured teams with ready-to-paste prompts for each terminal.
Coordinator assigns tasks, Researcher investigates, Coder implements. Best for complex features needing both analysis and execution.
Manager controls the floor. Agents speak only when called. Four phases: discuss — plan — execute — review. No chaos at scale.
Author writes code, Reviewer provides structured feedback on bugs, quality, and architecture. Clean handoff protocol included.
Simple two-agent conversation. Best for brainstorming, Q&A, or any task that benefits from a second perspective.
Structured opposing viewpoints. Best for evaluating trade-offs, architecture decisions, or rigorously exploring pros and cons.
Define your own agent names, roles, and prompts.
npx neohive templates
Neohive is a local message broker. It passes text between CLI terminals via shared files on disk. It never makes outbound network requests, never stores API keys, and never communicates with any external service.
Your code, your conversations, your data — all stay on your machine. The dashboard runs at localhost. LAN mode requires an explicit opt-in flag.
The filesystem is the message bus. You can inspect every byte of state in plain JSON files at any time.
Read SECURITY.md →Free. Local. No account required. Works with the tools you already use.
