by App Vendor · AI Agents · v3.0.4
A conversational analyst for your own data. Connect one or more datasets and ask questions in plain language: the agent finds the relevant tables, writes the SQL, runs it inside the confidential cluster space and answers with figures and charts. Raw records never leave the space and never reach the model — only aggregates do.
A weighted aggregate of four independent review lanes. No single source can inflate the score — community, machines, auditors and AI must agree.
What this version says it will deploy: its components and the exact images they run
A deployment of this version measures itself and must arrive at this. A data owner can decide to trust it before anyone has deployed it — and a deployment that stops matching stops being served.
Fields that legitimately differ between deployments — a namespace, a hostname. The list is measured along with everything else, so it cannot be widened without changing the digest.
Minimums. A target cluster space must have at least this much free before the deployment is accepted.
Datasets are selected at deploy time and stay inside the enclave
The data the agent answers questions about. Each dataset's published table and column descriptions become the agent's knowledge of it — there is nothing else to fill in.
Across community, agencies, scanners and AI
The knowledge graph and the data-source registrations are built at deploy time from the bound datasets' published schemas, so what the agent believes about the data cannot drift from what the data owner declared.
The knowledge graph and the data-source registrations are built at deploy time from the bound datasets' published schemas, so what the agent believes about the data cannot drift from what the data owner declared.
The knowledge graph and the data-source registrations are built at deploy time from the bound datasets' published schemas, so what the agent believes about the data cannot drift from what the data owner declared.