Shamrock
☘ the private analystChat with your CSVs — without ever sending the raw data to an LLM. The model sees only column names, writes a query, and your own database runs it on the real values. Lower cost, lower latency, and cells that never leave the room.
The model never sees your data
Ask a question in plain language. The LLM is shown only column metadata — never a single cell — and returns a SQL query. Your own database executes it against the real data, in your environment.
Actual values never cross the boundary. The privacy guarantee is in the architecture, not a checkbox.
Runs entirely on your ground
The whole system — including the database — runs entirely inside your environment, on your own infrastructure. Nothing outbound is required.
Point it at a bundled local model and Shamrock is fully offline: air-gapped analysis, no exceptions.
- 01
Metadata-only prompting
Only column names and types reach the model. It reasons about structure and writes SQL; it never handles the contents of a single row.
- 02
Your database does the work
Queries run on a single bundled database that holds accounts, permissions, usage, audit, and your datasets — one fast store for everything.
- 03
Governed access
IAM, roles, quotas, usage metering, and a full audit trail — administered from an Admin panel, with the model provider switchable on the fly.
- 04
Optional local model
Turn on the bundled local model and nothing needs the internet at all — the whole system runs offline, air-gapped.
To explain three-in-one, Patrick reached for the commonest weed underfoot. Its leaflets fold shut at night to protect themselves.
Three leaflets on one stem — model, query, and your database — where the model only ever touches the leaf it is shown, never the root. The clover is Shamrock's alone.