Kavla is a self-hostable infinite-canvas data workspace pairing tldraw with an in-browser DuckDB and an OpenAI-compatible agent for exploring data.
Yes, Kavla is open source under the Apache-2.0 license.
Kavla is free to use.
Yes, Kavla can be self-hosted (the source is available under the Apache-2.0 license).
Kavla is a self-hostable, infinite-canvas data workspace: it puts a tldraw canvas, an in-browser DuckDB, and an OpenAI-compatible agent on one surface, so you can lay data out spatially, query it locally, and let an agent help drive the analysis. Open source: yes (Apache-2.0); self-hostable; free OSS. It is an early project (~19 stars, ~75 commits) from a single maker, and it works with the Codex CLI and other OpenAI-compatible endpoints.
Most data exploration happens in a notebook or a BI tool's rigid grid. Kavla swaps that for an infinite canvas (built on tldraw): you arrange queries, results, and notes spatially, the way you'd sketch on a whiteboard. Underneath, an in-browser DuckDB runs the SQL locally, so exploration stays fast and nothing has to leave the page, and an OpenAI-compatible agent — compatible with the Codex CLI — sits alongside to help write queries and move the analysis forward. The whole thing is self-hostable and Apache-2.0, so the canvas, the data, and the agent wiring are yours to run. Open source: yes (Apache-2.0); self-hostable; free OSS. It is early and low-traction today, so treat it as a tool for exploration and prototyping rather than governed production reporting.
A lot of GTM analysis is exploratory — pulling a CSV, joining two tables, eyeballing a cohort — and that work fits a spatial canvas better than a locked dashboard. Kavla's combination is appealing for that: local DuckDB speed, a whiteboard-style layout, and an agent to take some of the SQL off your hands, all self-hosted and open so sensitive data stays in the browser. For an analyst or small RevOps team that wants to prototype on owned data with an agent in the loop, it is worth a look. The honest read: this is an early, ~19-star, single-maker project — a promising exploration tool, not a governed BI platform — so use it for prototyping, bring your own model endpoint, and don't expect production-grade sharing, permissions, or large-warehouse scale yet.