Reeve
Skip the query. Get the answer.
From business question to a path through your data, in one conversation.
Reeve answers questions about how your organisation's data fits together: which tables link to which, how strongly, and where no reliable path exists.
No SQL, no waiting on an engineer. Guided data navigation over the relationship graph Relait discovers.
What an answer looks like
You: There's no documented link between the patient portal and billing – is there a real one in the data?
Reeve: Nothing documented connects them. The data does.
PORTAL_ACCOUNTS – 99% – PAT_MASTER – 97% – CLAIMS
Path reliability 97% – the weakest hop, never a multiplied product.
Every answer is drawn from the live graph and carries its own data-freshness stamp.
What you get
Plain-English path discovery
Ask how two things connect and get the route, table by table, with the evidence behind every step – tracing paths that span source systems with no declared foreign keys. Working that out by hand grows combinatorially with the size of the estate, and quickly becomes effectively intractable.
Honest no-path answers
When nothing meets the reliability bar, Reeve says so and shows what it tried. No invented links.
Match strength on every hop
Path reliability is the weakest hop, never a multiplied product. You set the bar per question.
Auto-generated data models
Ask for a picture and get one drawn from the live graph – solid for documented links, dashed for discovered.
Not a text-to-SQL tool
Text-to-SQL hands you a query you still have to trust. Reeve hands you the path and the evidence behind every hop, and refuses to guess.
Reeve does not write queries for you, does not read a row, and does not answer questions outside data navigation. Those are not policies. They are the boundary the tools enforce.
On rails, in your environment
Reeve operates on column-level aggregates only. It never joins back to individual source rows, and the tools it can call return schema and match rates and nothing else. Ask it for a person's records and it declines, and explains why.
You: Show me Jon Smith's claims
Reeve: I can't do that. I work on column-level aggregates and never read rows, so I can show you how the claims data connects to the patient master, but not what any individual record says.
Your data stays in your environment. Reeve reaches its language model through one narrow boundary: a fixed set of read-only tools over the graph. What crosses that boundary is the question you asked and the graph metadata the tools return – table names, path structure, match strength. Never a row, and nothing about your estate is trained into the model.
Reeve itself ships as a container into your own infrastructure, and where the model runs is part of the deployment conversation – a hosted frontier model to start quickly, or a model inside your own perimeter where policy requires it.
Powered by Relait
Relait powers it. Reeve makes it askable.
Relait is the deterministic discovery engine that maps how the data actually connects and keeps score over time. Reeve is the surface that makes the result answerable in business language.
See it in full
The full walkthrough, worked example and answer gallery live at askreeve.ai.