Set the goal. Agents finish it.

Niuniu is a local-first AI dev workstation: write the work as issues, and many AI Agents plan, execute, and verify in parallel across projects and repositories — only coming back to you when they're truly blocked.

↓ Download Personal View Self-Hosted Team →

Screenshot of the Niuniu workstation: parallel task list, chat panel, commit feed, and terminal

Live demo

Many agents, running in parallel

Turn single-session Claude Code's “one chat you drive” into many workspaces driving many agents at once — each picking its own flow from your kanban stages, looping autonomously.

Niuniu demo animation: 3 workspaces each running an agent, each picking its own kanban stages, looping autonomously in parallel — 3 workspaces, each driving one issue; agents advance in parallel, fully isolated, finishing on their own. Illustrative animation (not a screen capture of the desktop app), shown to convey the parallel scheduling model.

Core capability

Goal in. Done out.

You define the goal; the agent owns plan → execute → verify on its own, and only comes back to you when it actually needs you.

1
Write the goal

File the work as an issue — one sentence is enough; no need to spell out the code.
2
Agent plans

Inside the workspace, the agent reads context, breaks the work down, and proposes a plan.
3
Parallel execution

Many workspaces pursue their own goals in parallel; each agent works directly in its own git worktree.
4
Self-verification

After changes, the agent runs tests, reads the diff, self-reviews, and iterates on failures.
5
Only when blocked

You only get pinged for real blockers — missing credentials, decisions, risky actions. Otherwise it ships quietly.

Pick your deployment

Two flavors — from solo desktop to enterprise self-host.

🖥

Personal

Local-first · 3 platforms · Zero deps

✓ macOS / Windows / Linux
✓ git worktree isolation
✓ SQLite backend

Download →

🐳

Self-Hosted Team

Docker · On-prem · Unlimited

✓ docker compose deploy
✓ Private data
✓ Custom audit

View install docs →

Core capabilities

🪟

Goal-driven autonomy

Write down the goal; the agent plans, executes, and verifies on its own — only pinging you when actually blocked.

🗂

Parallel workspaces

One workspace = one git worktree + isolated agent session. Many run in parallel without stepping on each other.

🤖

Built-in kanban

Issues, checklists, comments, timeline — all in one place.

⚙

Dual agent paths

PTY terminal or structured proxy — pick what fits the task.

👥

Harness orchestration

AI auto-selects the flow from your kanban stages (implement / AI review / human review / done) — automated, gated, observable.

🔌

MCP-native

Claude Code reads Niuniu state directly via MCP.

Why Niuniu

Niuniu vs single-session Claude Code / Cursor

Claude Code / Cursor — even with parallel subagents — is still one session you drive; Niuniu is the workstation that schedules many independent workspaces, each with its own team of agents, from a board.

	🐮 Niuniu	Single-session Claude Code / Cursor
Unit of work	A goal / issue	A single conversation
Concurrency	Many workspaces running many agents in parallel	One session you drive, can spawn parallel subagents
Isolation	Each workspace gets its own git worktree + session — no stepping on each other	Shared workspace, easy to collide
Autonomy	Plans → executes → verifies on its own; pings you only when blocked	You drive every step
Cross-project / repo	Native, scheduled from one kanban board	Manual context switching
Progress tracking	Built-in kanban: issues / checklists / comments / timeline	Scattered across chat history
Execution flow	AI auto-selects the flow from your kanban stages (implement / AI review / human review / done)	Fixed manual steps
State access	Claude Code reads Niuniu context directly over MCP	None

Comparison reflects publicly available product behavior and is for reference only.

One workstation. Many parallel AI Agent sessions.

Community & traction

Developers are shipping in parallel with Niuniu

Open source, local-first, actively iterating — follow the real progress on GitHub.

MIT: Open-source license, public code
3: Platforms · macOS / Windows / Linux
100%: Local-first · data stays on your machine by default
∞: Unlimited workspaces

Single-session Claude Code only let me watch one task at a time. With three workspaces running at once I ship roughly twice as much before end of day.

Indie developer Full-stack · SaaS side project

One git worktree per workspace means agents never step on each other — parallel refactors no longer risk breaking someone else's branch.

Backend engineer Mid-size dev team

The self-hosted team edition runs on our own machines, so code never leaves the network. Our compliance team finally signed off.

Engineering lead Fintech startup

Ready to get started?

Download the desktop edition, or explore Self-Hosted Team.

↓ Download Read the docs →