Ask your data anything.
Upload a CSV or connect your SQL database. Ciba AI maps your schema, plans a validated query, and answers in plain English — with the SQL it ran and per-message token cost.
SELECT region, SUM(revenue) FROM orders WHERE ts >= '2025-10-01' GROUP BY 1;
From schema to answer
in under four seconds
Connect
Point Ciba AI at PostgreSQL, MySQL, MariaDB, or upload a CSV. We map your tables, columns, and foreign-key relationships into a join graph in the background — the connection itself returns instantly.
Ask
Type a question. The agent plans a structured query against your schema, validates it against the join graph, then runs it read-only — scoped to just the tables your question needs.
Answer
Get a prose answer with the actual figures, the exact SQL that ran, the result rows, and per-message token cost — in one view. Multi-step questions chain queries automatically.
Built for serious
data teams
No hallucinated SQL
The model never writes raw SQL. It emits a structured query that we validate against your real schema and compile to parameterised SQL — invented tables, columns, or joins are rejected before anything runs.
Schema-aware joins
We build a join graph from your foreign keys, so the agent picks valid joins from your actual relationships — no manual modelling required.
Curated metrics
Define a metric like "revenue" once. The agent references it by name, so business definitions stay authoritative instead of being re-derived each query.
Per-message cost
Every answer records tokens in/out, latency, and the model used — so cost is visible per question, not just per month.
Role-based access
Owner, Admin and User roles, enforced server-side. Manage members, view org usage analytics, control who sees what.
Self-hosted
Ships as a Docker stack behind nginx. Deploy in your VPC; your data never leaves your network.
Your data
stays yours
Self-hosted by default. Your data never leaves your network.
- Self-hostedDeploy behind your nginx. Zero egress to third parties.
- Read-only, alwaysEvery customer connection is forced read-only with a per-statement timeout — no writes, no runaway queries.
- Scoped query planningThe model plans from schema metadata and only the slice of tables your question needs — not your whole database.
- Usage loggingTokens, latency, model and cost recorded per request, surfaced in org analytics.
- Role-based accessOwner, Admin, User roles enforced server-side with org-level management.
Stop writing SQL
for every question.
Connect your data and start asking questions in plain English.