Private by default
Keep the code transparent, review how the product works, and choose a self-hosted path when privacy matters most.
Private analytics with agent-grade reasoning
Roam Your Data
Ask once. Let the agent run the steps, trace the SQL, and turn the result into a visual answer you can trust.
Privacy-first deployment options Multi-step, multi-SQL analysis Automatic chart visualization Open source and inspectable
What users care about
People do not buy dashboards. They buy trust, speed, and clarity.
The product story should start with privacy, safety, and better answers, then show why the agent and the open-source path make those promises more believable.
Agent that thinks in steps
Let the agent break a question into multiple steps, run multiple SQL queries, and build a sharper answer instead of guessing from one shot.
Answers you can see
Turn raw query output into charts, evidence, and explanations that are easier to trust and easier to share.
Showcase
Show the thinking, not just the interface.
The strongest screenshots here are the ones that prove the agent can plan, query, and visualize instead of just showing a pretty shell.
Agent planning view
Show how the agent breaks a hard business question into smaller steps before it starts querying.
Multi-SQL reasoning
Show the sequence of SQL runs, intermediate evidence, and how the answer is assembled across queries.
Visual answer output
Show the final chart, summary, and table output in a way that feels decision-ready.
FAQ
Answer the objections before people ask them.
Will this expose sensitive data to a black box?
The product is positioned around privacy, inspectability, and deployment control, so teams can decide how much they want to run themselves and how closely they want to inspect the flow.
What makes it better than a chat box on top of SQL?
It does more than translate one prompt into one query. The agent can think in steps, run multiple SQL queries, gather evidence, and build a stronger answer from that trail.
Do I get charts or just tables and SQL?
The goal is to return something decision-ready: visible reasoning, supporting SQL, and chartable output instead of a raw dump that still needs manual cleanup.
How hard is it to try?
The goal is to keep it lightweight to inspect and run. Clone the repo, connect a database, and start exploring how the agent plans, queries, and visualizes answers.
Open source