The missing piece in a self-hosted CalDAV stack: a web UI for Tasks.org

Update, 8 October 2026: The app is being rewritten from scratch, with a new design and the shared CalDAV libraries @spy4x/caldav and @spy4x/time/ical. The current version does not work with Stalwart, can lose data when you save a task, and has no service worker. There is no public demo. Do not install the current version. This post describes what I set out to build in July 2026, not what the repo does today. Progress is in the repository.

Tasks.org syncs Android tasks to CalDAV cleanly. There is no web UI for that data. After a year of reaching for my phone to edit a todo at my desk, I built one. Here is the gap I noticed in the self-hosted ecosystem, and how I closed it without breaking the standards-first contract most of us signed up for.

Tasks.org is the strongest Android task manager and it syncs to CalDAV. The data lives on a server you control. The moment you sit at a desk, that data is unreachable from a browser — no conformant client exists for the web.

The data is fine. The standard is fine. The clients are missing.

Nextcloud is a whole suite, not a CalDAV server. People who chose Radicale or Stalwart on purpose shouldn't need to deploy a Nextcloud instance to edit a todo on desktop. The fix had to assume the CalDAV server is the only piece already running.

The goal is a task UI for whichever CalDAV server you already run. I used the current version against Radicale. Stalwart does not work with it today. Nextcloud and Baikal are untested.

VTODO files live on the CalDAV server. The web app is a client. In the current version, SQLite in the API container holds user accounts and AES-GCM-encrypted server credentials and nothing else. Switch CalDAV servers and no todos move.

That property is the design goal. The current version did not meet it: it does not work with Stalwart, and it can lose data when you save a task. The rewrite exists to fix that.

The read/write split maps cleanly onto CalDAV's GET versus PROPPATCH/PUT/DELETE. Each side tests in isolation; the CalDAV adapter is the only piece that knows the protocol. The idea was that Radicale and Stalwart would both speak through the same PROPFIND/PROPPATCH/PUT/DELETE interface, so adding a third server is a new adapter class, not a chain of if (serverType === ...) blocks. The rewrite does not use a CQRS layer.

The CalDAV XML responses vary more than they should. Radicale uses the default namespace and emits lowercase <d:response>. Stalwart emits uppercase <D:response> and <A:response> for the caldav namespace. Substring-matching "calendar" in the response body matches the user's principal home on top of actual calendars. The parsers have to be case-insensitive, prefix-tolerant, and filter calendars by <(prefix:)?calendar/> inside <resourcetype>. Stalwart still fails in the current version.

VTODO state changes constantly — status, priority, due date, percent complete, complete vs active, filters, sort, tag toggles. Signals give surgical UI updates without a virtual DOM tree of hooks to debug. A hook-based version of this project would have spent most of its life chasing re-render storms.

The hooks-vs-signals debate is a real one and the right answer depends on the app. For a heavily stateful list of items with frequent mutations and dense filtering, signals win on clarity and on raw update count. The codebase is roughly half the size it would be in React.

The current version deploys with deno task deploy, which rsyncs the project to a Hetzner box behind Traefik and runs docker compose up -d --build. It is not a single binary. It sits behind the same Traefik as the CalDAV server, and its SQLite file can be backed up by the same restic job as the CalDAV data.

There is no Kubernetes. There is no service mesh. The whole stack fits on a single $20/month box and stays there. A real engineer reading this knows that's the right shape for a personal productivity tool.

What the current version does: lists CalDAV calendars with their colors and renders a touch-first task list with filters, tags and due dates, plus a Kanban view. Credentials are encrypted with AES-GCM at rest.

What it doesn't do: it does not work with Stalwart, it can lose data when you save a task, and it has no service worker, so it is a web app you open in a browser, not an installable offline app. It has no sync engine (it is a client, not a server), no recurring-task editor and no native mobile wrapper.

I built it for myself. The rewrite is for the same reason.

Nothing to test yet. The rewrite is not ready, and I will not promise features or dates. If you want to follow it, watch the repository.

Self-hosting is a stance, not a chore. Every SaaS-shaped tool that wants to be the source of truth is a small erosion of that stance. The gap in the self-hosted CalDAV stack was real, and fixing it didn't need to be enterprise-grade — it just needed to honour the contract: the server you trust, the format you control, the data you already have.

If you want a desktop UI for the same data your Android phone sees, watch the repository for the rewrite. I would not install the current version.