Ask your whole estate anything. Safely.
Plain-English questions become governed SQL across every system — grounded in the graph, PII-masked, access-scoped, run in a sandbox, and fully audited.
Anyone can ask — in plain English
Grounded
Grounded in the graph, not guessing.
It doesn't hand your question to an LLM and hope. Davynci links your words to real tables, walks the join paths in the graph, then writes SQL that's actually correct.
- Entity linking fuses vector, keyword, and graph signals
- Join paths taken from the real knowledge graph
- SQL validated before it ever runs
Guardrails
Guardrails on every single query.
Every question passes the same gates: sensitive values masked, results scoped to the user's role, execution sandboxed, and the whole thing written to an audit trail.
- PII masked in responses automatically
- Access-scoped by role and domain
- Sandboxed execution, never an unsafe source query
- Immutable audit trail on every answer
For everyone
One layer for humans and agents.
The same governed layer serves a business user in the UI and an AI agent through one open interface — every answer carrying its meaning, lineage, and risk context.
- Plain-English UI for any business user
- One interface for Copilot, Claude, or your own agents
- Schema, lineage, PII and quality context per answer
Under the hood
Everything the query agent does.
Hybrid entity linking
Vector, keyword, trigram and PageRank fused (RRF).
Graph join paths
Real relationships from the graph, not guessed joins.
Sandboxed execution
Runs in DuckDB — never an unsafe live-source query.
PII masking
Sensitive values masked in the response, by role.
Access-scoped
Answers bounded to what the user is allowed to see.
Full audit trail
Every question and answer logged, immutably.
One brain, many modules
Ask Your Data is one module on the semantic layer.
Give everyone safe answers from all your data.
See a plain-English question become a governed, masked, audited answer.