Alpha

Alpha / Rapid iteration

The control plane for your AI workforce

The AI-Native Organization OS

One system where your team and agents work together — projects, tasks, agent sessions, and documents shared in one place, not scattered across tools.

Agents can keep running on your Mac, in the cloud, or on a private server. Your team can assign tasks, share context, and review results in one workspace.

Open Source
Privacy First
Self-hosted
Cross-platform

auth-flow-fix

Task linkedworktreemain
AI ChatFix auth callback bugMacBook-Pro · online
68%

Workspace Status

The session tree keeps the main path on the left while the file tree shows the current patch surface on the right.

Project Owner

Keep this workspace on the main patch path. Leave the file tree open on the right so I can track the patch and changed files together.

09:41 · workspace main session

AX
WD

Developer

I wrote the callback state persistence patch in this worktree. Next I will add the Safari redirect regression.

09:43 · patch in worktree

WD

Reviewer

The file tree on the right shows changes concentrated in auth/callback.ts and session-store.ts, so the risk surface is fairly contained.

09:46 · diff review

@DeveloperMacBook-Profiles split
Files
apps/web/src/routes
auth/callback.ts
session-store.ts
apps/server/src/routes
auth-session.ts
tests/e2e
safari-redirect.spec.ts

Diff Summary

+ Additions

18

- Deletions

6

Files

3

Execution Snapshot

Terminal Results

2 cmdspassed
$pnpm typecheck
apps/web ... ok
apps/server ... ok
$pnpm --filter web test safari-redirect
1 passed
0 failed

01 / Control Plane Status

Global Work Status

Keep kanban tasks, agent sessions, worker nodes, branch isolation, and human approvals in one control plane. Humans see the process and keep final judgment.

Tasks in Flight

12 Tasks Running
Task: AUTH_CALLBACK_FIX78%
Task: BROWSER_REGRESSION42%
Task: RELEASE_CHECKLIST12%

Worker Node Status

Branch Isolation

main
feature/task-chat-sessionIsolated
patch/worker-runtimeMerged
docs/release-notesIsolated

Awaiting Human Review

Pending Diff

// Agent proposes worker execution config changes

- worker: macbook-local

+ worker: cloud-vm

- review_required: false

+ review_required: true

// Reasoning: route long-running work to cloud capacity while keeping private repo changes under human approval.

02 / Pain Point

Agents can do the work — humans still lack the console.

Task Intake

Scattered Tasks

A chat can handle one request, but it does not carry priorities, status, and collaboration over time.

Worker Routing

Machine-bound Execution

When one computer goes offline, the work stops. Long jobs, private repos, and shared team capacity need routing.

Human Review

Opaque Results

Without unified logs, sessions, artifacts, and approval, humans cannot confidently confirm, retry, or take over.

Phone / Mobile Support

A mobile work console in your pocket.

Stay precise even while commuting. Create new dev tasks fast and check live diffs and execution logs from your phone.

Desktop is for heavy editing. Mobile is for critical control. You do not need to stay at your desk to keep the agent workflow moving.

Create Task
AI Chat
Terminal
Git Ops
09:41
5G88%

Live Task

Mobile Command

Payment callback fix / Reviewer waiting for approval

3 workers online, 1 task waiting for your approval, and logs are still streaming back.

Direct actions from mobile

Create task
Assign agent
View logs
Approve / Retry

Live Updates

Human In Loop

Check Status Fast

See which task is blocked, which worker is online, and which result is waiting for approval the moment you unlock your phone.

Act Immediately

Request a retry, switch the assigned agent, or take over at the exact moment work starts drifting.

Keep the Flow Moving

The workflow does not lose momentum just because the human operator is away from a laptop.

Safety and control: more than just AI.

Isolated Branch Execution

AI-generated code always runs in temporary isolated branches, so it does not directly contaminate your mainline repository.

Final Human Approval

Every code delivery still goes through human review inside Wemux before a one-click merge back to the main branch.

Workflow

Humans set goalsAgents executeWorkers deliverHumans approve

03 / Agent Roles

Configure dedicated agents for each kind of work.

Developer Agent

Build features, fix bugs, and prepare patches.

Tester Agent

Run environments, inspect browsers, and log issues.

Reviewer Agent

Review changes, flag risks, and produce checklists.

Doc Writer Agent

Update docs, release notes, and change summaries.

Researcher Agent

Research options, competitors, and supporting materials.

Operator Agent

Support launches, operations, growth, and recurring workflows.

04 / Distributed Execution

Turn every online machine into execution capacity.

You do not need to watch one computer until a task finishes. Deploy workers anywhere and Wemux routes work to the right online node.

Run long jobs on cloud machines, private repos on local hardware, and shared tasks on team workers.

WemuxControl

MacBook

192.168.1.24

Memory18 / 32 GBCPU41%

Auth callback fix

Cloud VM

10.18.0.12

Memory46 / 64 GBCPU72%

Browser regression

Office PC

172.16.8.33

Memory9 / 16 GBCPU24%

Docs cleanup

Home Server

192.168.1.88

Memory28 / 64 GBCPU55%

Research crawl

Backup Mini

100.96.4.17

Memory6 / 16 GBCPU8%

Waiting for route

Distributed Task Routing

Open Source

Open source. Self-hosted. Yours.

Wemux is open source under the Apache-2.0 license. The control plane, execution nodes, workspaces, chat and channel integrations are fully self-hostable — your model keys stay on your machines, and code runs in your own repositories.

Humans direct, agents execute

You set goals and priorities. Agents handle execution and delivery. Key results come back for your approval.

Bring your own keys

Configure OpenCode / Claude Code / Codex runtimes on your own machines — keys never leave them.

Multi-node mesh

Connect any machines, mesh them with easytier, and route work by capability.

Built with the community

Issues, PRs and discussions welcome. See CONTRIBUTING for the contribution guide.

High value · Wide coverage · Low latency

Pricing that grows with you

Coming soon

The community edition is ready

The community edition is free to self-host. Hosted options, when available, are described on the product pages.

Start for free

06 / Use Cases

Not just for development.

Development is the first workflow, not the only one.

Kanban + agents + execution workers can support many knowledge workflows.

Software Development

Requirements, bugs, refactors, testing, and review.

Product Planning

Ideas, PRDs, UI/UX work, and user feedback.

Content Writing

Docs, blogs, release notes, and research summaries.

Operations

Launch checklists, growth experiments, and customer feedback triage.

07 / FAQ

Questions teams usually ask first.

FAQ-01

Does Wemux replace existing AI tools?

No. Wemux is the operating system for AI-native organizations. It organizes tasks, agents, workers, logs, and approvals into one workflow while still letting you connect different agents or models.

FAQ-02

Is it only for software development?

No. Development is the first workflow, but writing, research, testing, operations, launch checklists, and feedback triage can use the same board-driven process.

FAQ-03

Where can worker nodes run?

Workers can run on your main machine, a cloud server, a spare computer, a team machine, or a home server. Any online worker becomes schedulable execution capacity.

FAQ-04

Will agents make final decisions for me?

No. Agents can execute and produce results, but final confirmation, retries, takeover, and completion stay under human control.

FAQ-05

What can I do from mobile?

You can create tasks, assign agents, inspect logs, track status, review results, request retries, or take over work while away from your desk.

AI-native organization OS

From single AI tools to the operating system for your organization

From a conversation to a traceable task, to execution in real environments and result review.

Humans and agents collaborate: you own goals and judgment, agents execute and deliver, and every result comes back for your confirmation.

Agents & humans collaborateHumans set the goalsAgents own execution