Hi everyone,
I self-host Vikunja for my own projects, and I kept running into the same
chore: turning a big, vague idea into a clean set of individual tasks by
hand. So I built a small open-source tool that does the decomposition with
an LLM — but keeps me fully in control.
What it does
You describe a project in plain language, and the AI breaks it down into
structured tasks. It can ask clarifying questions first if your description
is ambiguous. Then — and this was the whole point for me — nothing is written
to Vikunja until you review and confirm. Every task name, description,
comment and tag is editable, and you can drop any
works for editing existing projects too, not just creating new ones.
So the AI never silently changes your projects — it proposes, you approve.
How it connects to Vikunja
It talks to your instance through the Vikunja API using your own token, so
nothing is locked in. You bring your own LLM provider — Anthropic, OpenAI,
Google, DeepSeek, or fully local via Ollama (in which case nothing leaves
your machine).
It ships as Docker Compose behind nginx with TLS. Stack is Kotlin/Spring on
the backend (a gateway + a separate AI service) with a React frontend.
It’s still early and rough in places, and I’d really value feedback from
people who actually use Vikunja day to day:
- Does the clarifying-questions step sound useful, or would it just add
friction for you? - Would you rather see something like this as a standalone app (how it works
now), or integrated more directly — e.g. as an MCP server?
Repo with setup instructions:
Thanks for taking a look!