With AI becoming table stakes for SaaS, every company wants "chat with your data" features. But building this properly is harder than it looks. Many developers think they can just pipe user questions through GPT to generate SQL and call it done. Turns out that's nowhere near sufficient for production use.
Real data analysis requires iterative exploration, consistent metric definitions, proper visualizations, and handling of complex schemas. Without these, users get frustrated with wrong answers, inconsistent results, and raw data dumps instead of insights. Teams end up spending months building what should be a solved problem.
We solve this by using an agentic approach where the AI explores your data iteratively, maintains context across queries, and automatically generates interactive visualizations.
camelAI connects directly to your database (Postgres, MySQL, Snowflake, BigQuery, ClickHouse, Supabase, and more). You teach it your metrics through reference queries and knowledge base entries. We handle row-level security through session variables, so each user only sees their data. Everything is delivered through a customizable iframe - no backend work required.
We created a playground using the entire Hacker News dataset so you can see what the embed looks like (no signup needed): https://camelai.com/hackernews/. Try asking things like "What are the most discussed AI companies this year?" or "Show me posting patterns for Show HN posts".
We're currently in beta and it's free while we gather feedback. Our planned pricing will be usage-based at 25% markup on tokens used, plus a monthly fee for iframe embed. Setup takes about 5 minutes with our quickstart guide: https://docs.camelai.com/dev_api/camel-api-quickstart
Early customers include fintech and retail analytics platforms who use us to handle ad-hoc data requests from their users.
We're excited to see what the HN community builds with this and would love feedback on the agent's performance and what features you'd need!