Download

Prebuilt binaries for macOS, Linux, and Windows. No dependencies, no runtime.

macOS

# after downloading, make it executable and move to PATH
chmod +x wert-darwin-arm64
mv wert-darwin-arm64 /usr/local/bin/wert

Linux

chmod +x wert-linux-amd64
sudo mv wert-linux-amd64 /usr/local/bin/wert

Windows

Rename the downloaded file to wert.exe and move it to a folder in your PATH, e.g. C:\Windows\System32\ or any custom folder you have in %PATH%.

Verify checksum

sha256sum -c checksums.txt --ignore-missing

All releases

Browse all versions and changelogs on GitHub Releases →

Self-update

Once installed, wert can update itself in one command:

wert update

Or type :update in the cmdline inside the TUI. It fetches the latest binary, replaces the current one, and relaunches automatically.

Build from source

Requires Go 1.22+.

Build

git clone https://github.com/mugiwaraluffy56/wert
cd wert
go build -o wert .

# install to PATH
cp wert /usr/local/bin/wert

Cross-compile all platforms

make cross-build   # linux + darwin + windows × amd64/arm64 → dist/

Flags overview

wert serve

Start the server and open the admin TUI. Run this on the host machine.

Basic usage

wert serve --name alice

With admin token

If you set a token, only people who pass it with --token get the admin role. Without a token, the server trusts the first WebSocket connection as admin.

wert serve --name alice --token secret123 --port 9000

Data persistence

The server writes tasks and chat history to a JSON file after every mutation. Members and the server address are in-memory only.

wert serve --name alice --data /var/wert/data.json

REST API

The server exposes a REST API at the same port, used by the MCP server and agents:

wert join

Connect to an existing server as a team member.

Basic usage

wert join --host :8080 --name bob

If the server was started with --token, pass the same token:

wert join --host :8080 --name bob --token secret123

Join approval

First-time members wait for the admin to approve them:

• Member side — animated spinner shows "waiting for admin approval"
• Admin side — a Y/N popup appears. Press Y to approve, N to reject.
• Once approved, the member is remembered. Future reconnects skip the approval step.
• Rejected members see an "Access Denied" screen and must Ctrl+C to exit.

Reconnection

If the connection drops, wert automatically retries up to 15 times (every 3 seconds). The status bar shows progress. TUI state is preserved during reconnection.

Roles

MCP server

Let Claude manage your team's tasks, send messages, and coordinate with other agents.

Start the MCP server

wert mcp --server http://:8080 --agent-name claude

--agent-name registers this instance in the agent capability registry and stamps the agent's name on task updates. Omit it for anonymous access.

Claude Code config

Add wert to ~/.claude/settings.json:

{
  "mcpServers": {
    "wert": {
      "command": "wert",
      "args": ["mcp", "--server", "http://:8080", "--agent-name", "claude"]
    }
  }
}

Ask Claude to run team_context first — it returns all members and tasks in one call.

Available tools

Agent communication

Two AI agents can coordinate through wert: claim tasks, exchange private messages, publish results, and react to events in real time.

How it works

Typical agent workflow

1. 1
   Discover other agents. Call list_agents to see who else is connected and what they can do.
2. 2
   Claim a task. Call claim_task before starting work. Returns a conflict error if another agent already claimed it — prevents duplicate work.
3. 3
   Do the work. Update the task to in_progress with update_task.
4. 4
   Publish the result. Call post_result with a title and markdown content. It appears as a bordered result block in the team chat — distinct from regular messages.
5. 5
   Hand off privately. Use send_direct_message to send instructions only to a specific agent or human. The rest of the team doesn't see it.
6. 6
   Mark done and release. Call update_task to done, then unclaim_task.

Reacting to events with wait_for_change

Instead of polling, agents can block until something happens:

# wait for any event (30s timeout by default)
wait_for_change

# wait only for task updates or agent results
wait_for_change  filter="task_update,agent_result"  timeout=60

Event types: task_create, task_update, task_delete, task_claim, task_unclaim, agent_result, direct_message, chat, member_join, member_leave, agent_online

SSE endpoint (direct HTTP)

Non-MCP clients can subscribe to the event stream directly:

GET /api/watch
GET /api/watch?filter=task_update,agent_result

The server sends text/event-stream. Each event is data: {"type":"...","payload":{...}}. A keepalive comment is sent every 20 seconds.

Capability registry

Agents announce what they can do on startup. The registry is visible in the Members screen under "AI Agents".

# register via MCP tool
register_capabilities  capabilities="code_review,testing,analysis"

# or directly via REST
POST /api/agents  {"name": "claude", "capabilities": ["code_review", "analysis"]}

Registrations are in-memory only — agents re-register on each startup. The --agent-name flag on wert mcp handles this automatically.

Multi-agent config example

Two Claude instances with different roles in ~/.claude/settings.json:

{
  "mcpServers": {
    "wert-reviewer": {
      "command": "wert",
      "args": ["mcp", "--server", "http://localhost:8080", "--agent-name", "reviewer"]
    },
    "wert-deployer": {
      "command": "wert",
      "args": ["mcp", "--server", "http://localhost:8080", "--agent-name", "deployer"]
    }
  }
}

Navigation

Move between screens with number keys, tab, or the cmdline.

Screen numbers

Keyboard shortcuts

Header

The top bar always shows your username, role, online member count, and unread chat badge. The current screen name is shown next to "wert".

Home screen

Your landing screen with stats and quick navigation menu.

The Home screen shows the wert logo, a live summary (open tasks, done tasks, online members), and a navigation menu. It's purely informational — no input box.

Getting around from Home

Press 2–6 to jump directly to a screen, or open the cmdline with Shift+; and type a screen number.

Chat

Real-time team chat with @mentions and message history.

Sending messages

Type in the input box at the bottom and press Enter. Messages are broadcast to everyone connected. The last 100 messages are persisted on the server.

@mentions

Include @username in a message to mention someone. They'll see a notification badge in the status bar even if they're on a different screen.

Unread badge

When you're on any screen other than Chat and a new message arrives, an unread count badge appears in the header next to "2:Chat". It clears when you visit Chat.

Sending from Workstation

The Workstation screen has its own message input at the bottom. Messages sent there go to the same shared chat channel.

Tasks

Create, assign, track, and update tasks across the team.

Task statuses

Sub-tabs and filters

Press number keys 1–5 while on the Tasks screen (when the input box is empty) to jump directly to a filter tab:

You can also use [ and ] to cycle through filters left and right.

Creating tasks (admin)

# basic
/assign @bob "Fix login bug"

# with description and priority
/assign @alice "Deploy to staging" "Run deploy.sh on prod box" high

# with due date
/assign @bob "Write release notes" "" medium due:2025-01-31

Updating tasks

Use the short 8-character ID prefix shown in the task list:

/wip a1b2c3d4
/done a1b2c3d4
/blocked a1b2c3d4

Deleting tasks (admin)

/delete a1b2c3d4

Notification

When a new task is assigned to you, a notification appears in the status bar regardless of which screen you're on.

GitHub

View your org's repos, open PRs, and open issues — live from the GitHub API.

Quick setup

You need a Classic PAT with two scopes. See the GitHub setup guide → for a full step-by-step walkthrough.

/github setup --token ghp_xxxxxxxxxxxx --org your-org-or-username

Refresh

/github refresh

Sub-tabs

Press number keys 1–4 while on the GitHub screen (when the input box is empty) to jump to a tab:

You can also use [ / ] to cycle through tabs.

Personal accounts

wert automatically falls back from the org: search qualifier to user:, and then to a per-repo search if needed. Personal accounts and orgs both work.

Workstation

A dedicated split-pane workspace with a message input. Screen 5.

Default layout

By default, Workstation shows Tasks on the left and Chat on the right, with a message input at the bottom. Navigate here with 5 or :5.

Split pane commands

Open the cmdline (Shift+;) and use sp to set a custom layout:

Screen numbers for panes

Message input

The input box at the bottom of Workstation sends messages to the shared chat channel — same as the Chat screen's input.

Members

See who's on the team, their role, task counts, and registered AI agents. Screen 6.

Navigate here with 6 or :6. Each row shows:

• Online indicator — green * when connected, grey o when offline
• Username and role — admin shown in red
• Task counts — open and done tasks assigned to that member

AI agents section

Below the human members, a separator shows registered AI agents with their capabilities. Agents appear when they connect via wert mcp --agent-name and disappear when they disconnect.

In Workstation panes

You can show the Members view as a pane inside Workstation using screen number 6:

:sp h6    # members pane, full width
:sp h3 h6 # tasks left, members right

Cmdline

A Vim-style floating command bar accessible from every screen.

Activating

Press Shift+; (i.e. :) from any screen. The cmdline appears as a centered floating box overlaid on the current content. Press Shift+; again or Esc to dismiss it.

Navigation commands

Workstation commands

Split command syntax

v = vertical split (top / bottom), h = horizontal split (left / right), number = screen.

sp v3 h2   # tasks top, chat left + github right (bottom)
sp h3      # tasks only, full pane
sp h6 h3   # members left, tasks right

Split panes

Arrange multiple screen views side-by-side inside the Workstation.

How it works

The sp command builds a binary pane tree. Each argument adds a split:

• First arg — splits the full area. Direction + screen number for the new pane.
• Second arg — further splits the second leaf of the first split.

Examples

sp h3 — Tasks, full width

sp v2 — current top, Chat bottom

sp h3 h2 — Tasks left, Chat right

sp v3 h2 — Tasks top, Chat/GitHub bottom

Reset

Type :cl in the cmdline to go back to the default Tasks + Chat layout.

Command reference

All slash commands and cmdline commands in one place.

Slash commands (input box)

Cmdline commands (Shift+;)

Task priorities

GitHub setup guide

Step-by-step: create a Classic PAT with the right scopes so wert can read your org's repos, PRs, and issues.

Step-by-step

1. 1
   Open GitHub in your browser. Make sure you're signed in to the account that is a member of your org.
2. 2
   Go to token settings.
   Click your profile photo (top-right) → Settings → scroll all the way to the bottom of the left sidebar → Developer settings → Personal access tokens → Tokens (classic).
3. 3
   Generate a new classic token.
   Click Generate new token → Generate new token (classic).
   GitHub may ask you to confirm your password.
4. 4
   Fill in the token details.

   • Note: anything descriptive, e.g. wert-local
   • Expiration: set to 90 days or No expiration (your call)
5. 5
   Select exactly these two scopes:

   repo REQUIRED

   Full control of private repositories. This lets wert list your repos and read PR/issue data from private repos. Check the top-level repo checkbox — it automatically selects all sub-scopes (repo:status, repo:deployment, public_repo, repo:invite, security_events).

   read:org REQUIRED

   Read org and team membership. This is what enables the GitHub Search API to accept org:yourorg as a search qualifier. Without this, you get a 422 Validation Failed error even if you own the org. It's under the admin:org section — expand it and check only read:org.

   ℹ You do not need admin:org, write:org, or any other scope. Just repo + read:org.
6. 6
   Click "Generate token" at the bottom of the page.
   GitHub shows you the token exactly once. Copy it now — you cannot see it again.

   ghp_xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

7. 7
   If your org uses SAML SSO — authorize the token.
   Next to your newly created token in the list, click Configure SSO → Authorize next to your org name. If you skip this, GitHub will reject all API calls for that org even with the right scopes.

   ⚠ If you don't see a "Configure SSO" button, your org doesn't use SAML SSO — skip this step.
8. 8
   Connect wert to GitHub.
   In the wert TUI, go to screen 4 (GitHub) or use the input on any screen:

   /github setup --token ghp_xxxxxxxxxxxx --org your-org-name

   The --org value is the org slug (the part after github.com/), not the display name. For personal accounts, use your GitHub username.

9. 9
   You're done. wert fetches your repos, open PRs, and open issues. Use /github refresh any time to reload the data.

   ✓ wert automatically tries three search strategies: org: → user: → per-repo. So even if one approach fails for your setup, it falls through to the next.

Troubleshooting