Lightdash is the only BI tool we evaluated where the semantic layer is just our dbt project. Dimensions, metrics, and joins are YAML in the same repo as the models, reviewed in the same PRs, and deployed from CI with the CLI. For a small data team supporting a whole company, that is the difference between keeping up and falling behind. There is no second metrics catalog to maintain and no drift between what the warehouse says and what the dashboard says.
The AI features have been a pleasant surprise. Ask AI gives business users a natural-language way into the same governed explores analysts use, so answers stay consistent no matter who asks. The MCP server opens that same semantic layer to external AI tools, which makes it easy to build on top of without exposing raw SQL. Scheduled deliveries cover most recurring reporting with no custom code.
Support is responsive and the product moves fast. The team ships weekly and has turned around bugs we reported in days. Review collected by and hosted on G2.com.
Stability on the cloud instance has occasionally been an issue. One badly written table calculation on a single chart once put our instance into a restart loop, and a release this year broke pagination on some table charts until a patch landed. Both were fixed very quickly, but they were visible to end users for a short time.
A few smaller rough edges: some date filters are sensitive to project timezone settings in ways that are easy to get wrong, and a couple of API endpoints silently ignore fields instead of rejecting them. Role-based limits on features like custom SQL table calculations are sensible but occasionally surprise power users.
None of these are blockers, and the team has been candid about roadmap, but they are worth knowing if you run Lightdash at scale. Review collected by and hosted on G2.com.