Free, open source, and fully local.
Markdown tasks,
built for AI agents.
Keep long-running work clear and moving. Every task is a Markdown file that you and your agents can read, update, and hand off from the web app, TUI, or CLI.
Free · no login · your project stays local
The web app is the lightweight experience for trying Kandown or checking a board. Use the TUI and CLI for agents, automation, and the full workflow.
npm install -g kandown- Plain files
- One Markdown file per task. Easy to read, edit, and version.
- Made for agents
- Durable context, clear next steps, and useful completion reports.
- Works everywhere
- Web board, full terminal UI, and scriptable CLI.
Video 01 · Hero workflow · 16:10 · 12 to 15 seconds
Recording brief
Show the full Kandown workflow in one clean loop
Create a task in the web board, add subtasks, assign an agent, move the task into progress, save a completion report, and finish in Done. Return to the opening board state so the loop feels seamless.
Expected file: /demo.webm
Long-running work keeps its memory
Some tasks take hours, days, or several agents. Kandown keeps the plan, progress, blockers, and completion reports in files that survive every session.
Start with the full picture
Project rules, instructions, active tasks, blockers, and priorities arrive in one command.
Know what comes next
Kandown finds the highest-priority task that is ready to start.
Leave a useful handoff
Agents check off subtasks and record what changed, so progress survives the session.
Use any agent you want
Launch Claude Code, Codex, Gemini CLI, Goose, Aider, or OpenCode. Shell commands and MCP work too.
Video 02 · Agent handoff · 16:10 · 10 to 12 seconds
Recording brief
Show one task surviving two different agent sessions
Start the task with Codex in the TUI. Complete one subtask and save a report, then end the session. Launch Claude on the same task, show it reading the saved context, finish the remaining work, and move the task to Done.
Expected file: /agent-handoff.webm
Choose your workflow
Kandown ships one opinion about how agents should work: plan, take one task, check off subtasks, write a report. It is a good opinion, and it should still be a choice. Workflow selection is being designed now and lands before 1.0.
Pick a method, not a religion
A PRD-first flow, strict TDD, a research loop, or almost no ceremony. Choose one and every agent on the project follows the same plan, the same review checkpoints, and the same definition of done.
Today's protocol becomes one entry
The rules Kandown ships stay the default. They stop being the only option, and nothing changes for a project that never picks anything else.
Some rules are not opinions
Dependency gating, the Markdown round-trip, and how archiving works are the data model, not a method. Every workflow inherits those and none can override them.
Bring your own, or install one
A workflow is Markdown and a manifest. Fork one, adapt it to your team, keep it in the repo, or install one written by somebody else.
$ kandown workflow list default · plan, one task, subtask reports ai-dev-tasks · PRD, then tasks, one at a time tdd · red, green, refactor, per subtask $ kandown workflow use ai-dev-tasks Workflow set for this project. # Instructions come from the CLI, never a file# in your repo that can drift out of date.Your board lives in your project
Every task is a Markdown file you can open, edit, search, and commit with git. Kandown turns those files into a visual board without hiding them in a database.
Video 03 · Markdown sync · 16:10 · 8 to 10 seconds
Recording brief
Prove that the file and the board are the same thing
Record a split view with a real task file in the editor and Kandown beside it. Change the title, priority, assignee, checklist, and status in Markdown. Show each change appearing instantly on the real board.
Expected file: /markdown-sync.webm
One board, wherever you work
Plan in the browser, manage tasks from the terminal, or automate everything from scripts. Every interface reads and writes the same Markdown files, so your board never drifts.
Video 04.01 · Web board interface · 16:10 · 6 to 8 seconds
Recording brief
Record the real web board
Create a task, drag it between columns, open its details, update a subtask, and find another task with search.
$ kandown
Expected file: /interface-web.webm
Structure that survives the work
Clear handoffs
Every agent sees the plan, current progress, blockers, and reports left by the session before it.
Honest dependencies
Blocked work cannot reach Done until every task it depends on is resolved, from any interface.
Completion reports
A checked subtask can explain what changed, giving the next person or agent more than an empty checkmark.
Search everything
Find text across titles, descriptions, subtasks, tags, assignees, and priorities. Filter or group the result in seconds.
Local and private
No account, no telemetry, and no hosted database. Task commands work offline and your files stay inside the project.
38 themes, 48 languages
Choose a familiar theme, tune your own, and use the board in the language that feels natural to your team.
Start with two commands.
Kandown runs locally and keeps every task inside your project. No account, no hosted service, and no migration if you ever stop using it.
No install or login. Opens on a sample board, then works directly with local projects in compatible browsers. TUI and agent launching are not available on the web.
Full local experience
npm install -g kandownkandown init && kandown