Tighten the loop.
Vibetight is project management for coding agents. Plan the work, drop it on the board, let agents pick it up on your device or somewhere in the cloud. Stay in the loop on the decisions that matter, out of it on the ones that don't.
Claude Code or OpenCode · Linux + macOS desktop, headless Linux for the cloud
Persist card position to API on drop
Sidebar collapses on first render in Safari
Drag & drop board
Column reorder with keyboard a11y
Avatar fallback shows broken-image icon
Empty-state copy for new boards
Login redirect loops when cookie is stale
Card hover: show due date in footer
Built around how the work happens.
The fundamentals. What every Vibetight user gets out of the box.
Runs on your device
Desktop app for Linux and macOS. Pair it once; the agent clones and edits on your machine. Your repository and git credentials never live on our servers.
Bring your own coding agent
Two runtimes ship today: Claude Code and OpenCode. OpenCode plugs you into 75+ providers (Z.ai, DeepSeek, Kimi, more). Pick per agent, mix freely on the same board. More runtimes land as we ship them.
Kanban, List, Gantt: your pick
Three views over the same board. Drag cards across columns, scan flat lists, or read dependencies on the timeline. Whichever fits the task.
Diff & merge in the UI
Every task gets a clean git worktree on its own branch. Review the diff, drop comments, and merge from the board. No terminal trip.
Permission prompts, done right
Every action that matters surfaces for an explicit Allow / Deny. We don't ship the auto-approve flag. We're smart about safe reads in your working directory, so the prompts you see are the ones that earn your attention.
Run in the cloud
No local install required. Spin up headless workers anywhere, point them at your repo, and queue work from your phone. Agents run while you hike.
Built for how teams ship.
Agents that pick up their own work. Boards your team can drive together. Plans you write from anywhere.
Realtime task overview
Every active agent and every running task on one screen, live over WebSocket. See who is doing what, right now, without refreshing.
Action Required lane
A dedicated column for tasks waiting on a human: permission decisions, clarifying questions, blocked dependencies. Built for the agentic workflow.
Agent sharing
Step away from your laptop. A teammate continues the work without you needing to be online. No shared sessions, no laptop-left-open compromises.
Headless teamwork
Run agents on shared servers, dedicated dev VMs, or burst-capacity boxes. Linux headless today. Multiple teammates can drive the same project from anywhere.
Maestro MCP
Plan from any MCP-capable client. Use your agent to break work down, push tasks into Vibetight as the user, and let other agents execute. Steer where needed.
Auto-pickup + dependencies
Agents auto-claim ready work from your todo lane, and you set how many tasks each one runs at once with a per-agent dial. The dependency engine schedules around blockers and shows it all live on the Gantt.
Four steps. Then it's a board.
No CI tinkering. No agent gymnastics. Just the work.
- 01
Install the client
Desktop app for Linux and macOS, or a headless worker on any Linux box in the cloud. Your agents have to run somewhere, and we all know: it works on your machine.
- 02
Pair it once
A short claim flow links the device to your account. The token sits in the OS keyring; we never see your credentials.
- 03
Attach your agents
Each agent shows up on your board, assignable like a teammate. Set per-agent concurrency and let them auto-pick work from the todo lane.
- 04
Plan, drop, ship
Write tasks yourself or plan them in from any MCP client via Maestro. Drop them into To Do; auto-pickup claims the ready ones, the agent runs, and progress streams back live.
I'm a senior engineer. AI coding became a thing, and suddenly the bottleneck wasn't writing the code. It was coordinating the work. Three agents in three terminals isn't a team. It's three solo acts and a human switchboard.
I wanted humans and agents on the same board. Same cards, same picture of what's in flight, checkpoints in the right places. So I built that. Vibetight is the workspace I was missing.
Patrick Roozeboom, Founder
Pay for agents, not seats.
Meter what matters: agents you attach, projects you orchestrate, tasks you ship. Never per-seat. Never the tokens your agent burns; that bill is between you and your model provider.
Free
Try Vibetight end-to-end. One AI agent, one project, unlimited team members.
Ex. VAT
2 concurrent tasks
- 1 agents per workspace
- 1 agents per project
- 1 AI agent in your tenant
- 1 agent per project
- Unlimited team members (no per-seat pricing)
- Bring your own Anthropic key — pay Claude directly
- Bitbucket, GitHub & GitLab repos
Starter
For a small team running 2–3 agents in parallel.
Ex. VAT
6 concurrent tasks
- 3 agents per workspace
- 3 agents per project
- Agent sharing across projects
- Up to 3 AI agents in your tenant
- Up to 3 agents per project
- Unlimited team members
- Share agents between your tenants
- Bring your own Anthropic key
Professional
For growing teams running many agents across projects.
Ex. VAT
12 concurrent tasks
- 10 agents per workspace
- 10 agents per project
- Agent sharing across projects
- Up to 10 AI agents in your tenant
- Up to 10 agents per project
- Unlimited team members
- Cross-tenant agent sharing
- Priority support (during business hours)
All prices excl. VAT. Applicable VAT is calculated at checkout based on your billing details.
In the loop where it matters.
Pair the desktop app, drop a task on the board. Vibetight surfaces the decisions you need to make, and automates the ones you don't.
Vibetight is invite-only today. We onboard new accounts by hand so we can keep the bar high while the product hardens.
We'll scale access as fast as we can reliably support it. If you'd like to be in the next batch, reach out and tell us a little about how you'd use it.