// the find
mprove-io/mprove
Open Source Agentic Business Intelligence with Malloy Semantic Layer :tada:
Mprove is a self-hosted BI platform that uses Malloy as its semantic layer instead of LookML or dbt metrics. It sits in the same space as Looker or Metabase but bets on Malloy — Google's open-source data modeling language — as the query abstraction. Aimed at data engineers who want version-controlled metric definitions and don't want to pay Looker prices.
Malloy as the semantic layer is a genuine differentiator — it handles fan-out joins and aggregation correctly by default, which is a real pain point in LookML and raw SQL BI tools. The git-backed project model (branches, commits visible in the tree structure) means metric definitions are treated like code, not config stored in a database you can't diff. Docker Compose setup is straightforward and the e2e test suite is extensive — the controllers directory shows nearly every endpoint has test coverage including error cases. Helm chart support means it's actually deployable in a real infra setup, not just a demo.
336 stars for a tool this complex is a warning sign — Malloy itself is still niche and Mprove doubles down on that bet, so you're adopting two non-mainstream tools simultaneously. The 'agentic' label in the description is doing a lot of marketing work; nothing in the README or directory tree suggests what the agent actually does beyond standard BI. Password reset requires SMTP configured before you can invite other users, which is a friction wall for evaluation. TypeScript monorepo with NestJS backend and what looks like a separate Angular/React frontend means the operational surface area is large — six services in the docker-compose command — for something you'd need to maintain yourself.