# Vibemux

> Agent work console that routes AI coding tasks from a distributed kanban board to real worker machines, returning branches, execution logs, and reviewable diffs under human approval.

Vibemux is a control plane for AI coding work that runs against real repositories on real machines. Humans put tasks on a distributed kanban board, assign them to role-based agents, and route execution to any online worker node — a laptop, a cloud VM, a spare office PC, or a private company server — then get back branches, commits, execution logs, and reviewable diffs. The site labels the product an alpha release that is rapidly iterating, with the UI and capabilities still changing.

## What It Is

The product's framing is that AI tools can already produce output, but managing the work is the missing layer. The site argues that tasks scatter across one-off chats, execution depends on a single machine, progress is hard to trace, and results are hard to coordinate. Vibemux positions itself as the execution and delivery layer rather than a replacement for existing models or agent runtimes — the FAQ states plainly that it does not try to replace your existing agents, and that you can connect different agents or models to it.

## How the Workflow Runs

The documented loop is create a task, pick an agent, route a worker, open an isolated workspace, execute with logs, review the result, then complete or retry. A dashboard consolidates projects, tasks, agents, workers, and pending reviews into one operational view, showing task progress, worker node status, branch isolation state, and items awaiting human confirmation. A group-chat surface puts teammates, agents, and executors in the same thread, where an agent is invoked by @mention and each session can select its own executor.

## Agent Roles

Rather than one general assistant, Vibemux assigns work to specialized roles. Developer agents build features, fix bugs, and prepare patches. Tester agents run environments, inspect browsers, and log issues. Reviewer agents review changes, flag risks, and produce checklists. Doc Writer agents handle documentation, release notes, and change summaries. Researcher agents gather options, competitor information, and supporting material. Operator agents support launches, growth work, and recurring workflows. The site states that humans retain ownership of goals, priority, and final approval.

## Distributed Worker Nodes

Workers turn any online machine into schedulable execution capacity. Per the install documentation, a connect command is generated in the console and run on the target machine, with local installs available for macOS, Linux, and Windows plus a Docker option that runs the worker inside a node:22-bookworm-slim container. Local installs register a service with auto-restart and auto-update; Docker relies on the container restart policy. The console shows a node inventory with IP, memory, CPU, and status, so long-running jobs can go to cloud machines while private repositories stay on local hardware.

## Safety and Human Approval

Agent-generated code always runs in temporary isolated branches so it does not touch the mainline repository directly, and every delivery passes through human review inside Vibemux before a one-click merge back to main. The FAQ is explicit that agents can execute, retry, and report but that final confirmation, takeover, and approval stay with the team — describing the product as an orchestration and delivery console rather than an autonomy fantasy. A mobile view lets operators create tasks, assign agents, inspect logs, approve, or request retries away from a desk.

## Beyond Software Delivery

Software delivery is described as the first wedge rather than the only one. The same board-driven control plane is presented as applicable to product planning covering ideas, PRDs, and user feedback; content work including docs, blogs, and release notes; and operations work such as launch checklists, growth experiments, and customer feedback triage. The common thread is any knowledge workflow that needs traceable execution and a review path.

## Features
- Distributed kanban board for AI agent task management
- Role-based agents: Developer, Tester, Reviewer, Doc Writer, Researcher, Operator
- Task routing to any online worker node
- Worker nodes on macOS, Linux, Windows, or Docker with auto-restart and auto-update
- Node inventory showing IP, memory, CPU, and status
- Isolated branch execution for all agent-generated code
- Mandatory human review and one-click merge back to main
- Execution logs, sessions, and artifacts kept traceable
- Group chat with teammates, agents, and executors in one thread
- Per-session executor routing within a shared context
- Mobile console for creating tasks, viewing logs, and approving work
- Pending-approval queue with diffs, test records, and risk notes
- Multi-project workspaces with team collaboration
- Global skill catalog and MCP registry
- Bilingual interface (English and Chinese)

## Integrations
GitHub, GitLab, Vercel, Docker, Creem

## Platforms
WEB, MACOS, LINUX, WINDOWS

## Pricing
Freemium — Free tier available with paid upgrades

## Links
- Website: https://vibemux.com
- Documentation: https://vibemux.com/docs/worker-install
- EveryDev.ai: https://www.everydev.ai/tools/vibemux