Work management we built, and run our own work on

1. 01

   Work scattered across tools

   Tasks lived on a board, the context in notes, the updates in chat, and the small follow-ups disappeared, so progress had to be chased by hand.

2. 02

   Too light or too heavy

   Simple boards fell apart once the work got real, and the tools that could model anything needed long set-up and someone to administer them.

3. 03

   AI agents outside the record

   Agents can do real work, but they need to read the task, act within the right permissions, show their progress, and leave a trail the team can review.

A card, from request to done

1. 1

   Capture

   A card is added on a board, captured from the scratchpad, mirrored in from a client's board, or created by an agent, up to 50 at a time. It picks up the workspace's lists, labels, and fields.

2. 2

   Plan

   It gets a start and due date and links to the cards it waits on. The Timeline runs a working-day schedule, highlights the critical path, and flags any card planned to start before its blocker finishes.

3. 3

   Delegate

   It's assigned to a teammate or handed to an AI agent. The agent opens a run, and everyone on the board sees a live chip on the card that turns amber when it's waiting on a person.

4. 4

   Build

   A branch named with the card key links its pull request to the card, which shows the repository, the lines changed, and whether it's draft, open, merged, or closed.

5. 5

   Move

   When the card enters review, an automation can assign the reviewer, set a due date, post a comment, or call a webhook, and the change can be posted to Slack, Discord, Telegram, or Zulip.

6. 6

   Review

   The run's outcome stays in the card's activity, and Work Done and its report show what moved forward, with agent work told apart from the team's own.

• Shared workspaces

  Boards in a workspace share lists, labels, custom fields, checklist templates, and automations, so reporting stays consistent. Standalone boards keep their own set-up for work that doesn't fit.

• Views of the same work

  Kanban, Table, Calendar, Timeline, and Dependencies on a board, plus My Cards, Team Cards, and Work Done across boards, with a ranked Up next queue for each person.

• Dependencies & timeline

  Blocking links across boards, traced in a map, timeline, or numbered steps. A Gantt-style Timeline shows the critical path and how far each other card could slip.

• Automations

  A When trigger and ordered Do actions: 16 triggers, from cards entering lists to blocked work and slipping plans, and 14 actions, from assigning people to calling a webhook.

• AI agents on the board

  An MCP server with 104 tools, a CLI on the same tool layer, an agent skill, and an Agent Workflow template. Runs show live progress, and comments carry a badge naming the agent.

• GitHub pull requests

  A card key such as KAN-123 in a branch, title, or description links the pull request to the card. Its state stays current, and commit links render as readable change summaries.

• Notifications where you want them

  Each notification type can go to email, browser push, ntfy, Gotify, or your own signed webhook, with exceptions per workspace. Card activity can be posted to team chat.

• Any screen, any language

  The full product works on phones and tablets and installs as an app, with recently opened boards readable offline. The app, emails, and push notifications come in 11 languages.

• Team members

  • One ranked Up next queue across every board
  • Context, files, and history kept on the card
  • Notifications by email, push, or their own channel

• Team leads

  • Team overview of what everyone has in progress
  • Work Done reports for stand-ups and reviews
  • Blocked, at-risk, and critical-path work in view

• Admins

  • Workspace set-up shared across every board
  • Automations that run across a whole workspace
  • Read-only or write access for each agent connection

• AI agents

  • 104 MCP tools, or the same tools from a CLI
  • Runs that show live progress and blockers
  • Event subscriptions for new cards, comments, and moves

• Clients & guests

  • Access to only the boards they're invited to
  • Mirrored boards that keep internal fields private
  • Comments and files that reach the team live

• Developers

  • A REST API, signed webhooks, and a TypeScript SDK
  • Pull requests linked to cards by card key
  • A self-hosted option with no plan limits

1. May–June 2026

   Launch

   Built from May and used for our own work from the start. The public site began on 1 June, and v1.0 shipped on 30 June with the MCP server and its tests in place.

2. July–August 2026

   Teams & integrations

   Multiple organisations, card keys, personal notification channels, Up next, the scratchpad, idempotent API writes, and the CLI and SDK.

3. September 2026

   Agents on the board

   Agent runs and attribution, CSV import, undo, themes, agents setting up workspaces from templates, and an end-to-end test suite.

4. October–November 2026

   Planning & delivery

   The Agent Workflow template, prebuilt self-host images, dependencies, time in progress, 11 languages, then the Timeline and GitHub pull requests on cards.

• The MCP server and the CLI share one tool layer, so anything an agent can do over MCP it can also do from a shell, and read-only credentials are refused writes by the server itself.
• Every change is validated, checked against the person's access, recorded in the card's activity, and then broadcast, with events kept in a durable outbox so other processes and webhooks deliver them too.
• An agent run that stops reporting is marked stalled after 15 minutes, so an interrupted agent can't leave a card looking active.
• One shared package defines the database schema, request and response shapes, and realtime events, so the API, web app, MCP server, and CLI change together.
• More than 3,500 automated tests. Unit, web, MCP, CLI, and SDK suites and a three-way parallel end-to-end suite run on every push.
• Performance changes are measured before and after: in one pass, the time for notifications to reach a 200-member team fell from 14 seconds to about 3 at the 95th percentile.

• TypeScript
• Angular
• Fastify
• Socket.IO
• PostgreSQL
• Drizzle ORM
• Valkey
• MCP
• Docker
• Playwright