What Kolibri is
Kolibri tracks work: tasks on boards, sprints, milestones, a wiki, and the conversation around all of it. If you have used Jira, Linear, Plane or OpenProject, nothing on the surface will surprise you.
Three things underneath are different, and they are the reason the surface behaves the way it does.
It reads from your device, not from a server
Section titled “It reads from your device, not from a server”Every screen is drawn from a full copy of the workspace held in your browser. Filtering a board of two thousand tasks is not a request; grouping them by assignee is not a request; opening a task is not a request. There is no spinner because there is nothing to wait for.
The same mechanism is what makes the app work with no connection at all. Changes you make go into a queue and are sent when there is something to send them to — from a train, from a plane, from a hotel that wants your email address before it will route a packet.
When two people changed the same task while apart, Kolibri merges field by field: one person changing the title and another changing the priority both keep their change, rather than the later save flattening the earlier one. A page body is merged differently again — character by character, so two people typing in the same paragraph both keep their sentences.
It is one process and one file
Section titled “It is one process and one file”The whole server is a single Node process with a SQLite database beside it. No Postgres, no Redis, no queue worker, no search cluster. That is not a limitation you are working around — it is why an instance can live on the smallest virtual machine a host sells, why a backup is a file you can copy, and why upgrading is pulling an image.
For you as a user this shows up in one place: things are fast, and the instance is unlikely to be down.
An assistant is a first-class user
Section titled “An assistant is a first-class user”Kolibri speaks the Model Context Protocol natively. An AI client — Claude, or anything else that speaks MCP — can be pointed at your instance and given a token, and from then on it can list projects, read the backlog, file tasks, move them through the workflow, write pages and log time. Not by driving a browser, and not through a scraped API: through tools the server exposes on purpose.
What it may do is exactly what the token says. A token issued with read-only scope is refused by every tool that writes, so handing an assistant a view of the backlog does not hand it a pen.
What is in it
Section titled “What is in it”| Tasks | Sub-tasks, blocking relations, priorities, estimates, labels, due dates, assignees, custom fields, attachments, comments |
| Views | List, board, table and calendar over the same work, grouped and filtered however you like, saved under a name and shared |
| Planning | Cycles, modules, a timeline where dragging a task moves what it blocks, baselines, work-in-progress limits, four charts per project |
| Pages | A nested markdown wiki two people can edit at once, with version history, comments and read-only share links |
| Automation | Templates, repeating tasks, and rules that file work when something happens |
| People | Chat, an in-app inbox with optional email, push and Telegram, intake forms for people with no account |
| Elsewhere | A calendar subscription, CSV and JSON import, an assistant over MCP, a REST API and webhooks |
Three languages throughout — English, German and French — including the emails, each written in the recipient’s own.
What it is not
Section titled “What it is not”Being clear about this is cheaper for everybody than finding out in month two.
- It is not a timesheet system. Time is recorded and added up; there is no week view, no submit and no approval step. See time tracking.
- There is no cost or billing. Entries carry a
billableflag that nothing reads yet. Rates, budgets and cost reports do not exist. - There is no forecasting. The burn-up chart stops at today rather than drawing a line into the future, because a line into the future looks like a promise.
- Estimates are in points, not hours, and are deliberately never compared with logged time.
- Documents are markdown pages, not a document editor. No page layout, no tables of contents generated across a space, no PDF authoring — though any page prints.
- There is no email-in. You cannot file a task by writing to an address. A commit message can reach a task; an email cannot.