A new kind of task manager has arrived for developers and small teams, merging the rhythm of agile cycles with the calmness of a personal notebook. gitDone packets replace sprints, AI digests your inbox, and a public REST API lets your tools join the conversation. Free to start. Owned by you.
Send a task to the AI — your local agent launches Claude Code and you watch the work live.
Every task has a "Send to AI" button. The gitdone-agent on your machine picks it up, launches Claude Code with the task's full context, and the status moves on its own: analysis, coding, tests. You see every step on the board without touching a terminal.
While the agent works, the AI Console streams everything live — every action, every file, every command. When it's done, the token spend is recorded as a note on the task itself, so you know what the run cost.
When it hits a question only a human can answer, the AI stops in "AI question" status and asks in a note on the task. You reply — gitDone automatically re-dispatches it and the work continues. The final check is always yours: the task lands in "Testing", and only you move it to DONE.
Whatever the moment asks for — board, list, calendar, or quarterly roadmap.
Linear has cycles. Shortcut has sprints. gitDone has packets — flexible bundles of tasks with optional start and end dates. Activate one to focus your dashboard; complete it to roll unfinished work into the next. A burndown chart lives inside each expanded packet card.
The execution page offers a Kanban for workflow, a List for fast scanning, and a Calendar for date-based planning. Each remembers your last choice. Drag a task across a calendar day to reschedule it; click its status chip in the list to advance it without opening the modal.
Zoom out to three months, six months, or a year and see every packet as a bar on the timeline, grouped by project. Today is a red line. Done work fills with darker shade. It's the view you show stakeholders without explaining anything.
Markdown, mentions, and shared projects without leaving the task.
Comments accept full Markdown — bold, italics, inline code, fenced blocks, lists, tables, blockquotes. The renderer is GFM-flavoured. No HTML, nothing renders that the writer didn't type.
Type @ and an autocomplete drops in with your friends. Tab to select, Enter to send. The mentioned user receives a bell notification, and — if they configured a Slack webhook in settings — the message follows them into Slack with the task reference embedded.
Invite a friend, assign them a role, set their per-action permissions: can they add tasks, approve them, move into testing, mark them done, delete? Members see the shared project alongside their own; tasks created by anyone show up everywhere appropriately.
GitHub commits, Slack pings, REST endpoints, and AI assistants over MCP.
Link a project to a repo and commit messages containing gd-N auto-update task status. Open a PR, the task shows the PR; merge it, the task moves to Done. Same webhook secret per project, verified with HMAC-SHA256.
Paste an Incoming Webhook URL in Settings → Slack; @mentions arrive in the channel of your choice. For everything else, register an outgoing webhook with HMAC signing and we'll POST signed JSON on every task event.
REST endpoints at /api/v1 with an OpenAPI 3.1 spec for Postman and Zapier. AI assistants speak the Model Context Protocol on /api/mcp — ten tools for tasks, projects, and packets. Both authenticate with the same Bearer API key.
Sign up free with email or Google. Hook up the local agent to have Claude Code write your code — or start without it; the rest works just as well.