Model Context Protocol
Any AI, grounded in your own data lake
datons Ground connects any AI model to your company's own data through the open Model Context Protocol, with a governed catalog of skills, SQL functions, and reports. Deployed in your tenant, identity-preserved, and maintained by datons.
Reference tenant live at test.ground.datons.com, OAuth-connected to claude.ai.
Why it's different
Most AI-over-data tools ask you to move your data, trust a black box, or lock into one vendor. Ground does none of that.
Governed as enablement, not a lock
A curated catalog of skills, SQL functions, and reports is the 'how' of your business — so the AI is reliable and useful, and each team gets exactly the tools its use case needs.
Identity preserved end to end
Entra OAuth On-Behalf-Of plus row-level security: every user sees only what they're allowed to. The agent inherits the person's permissions, never bypasses them.
Your cloud, no data migration
Postgres, ClickHouse, Fabric, and more connect as data sources — the engine governs queries on top. The only thing it deploys is its own governance catalog. Your data never leaves your tenant.
Interoperable, maintained by datons
Built on the open MCP standard — the same one Claude, ChatGPT, and Copilot speak — so you're not tied to any model or to us. Your catalog is yours; datons keeps the engine current.
In your AI client
Your team just asks — in the AI tool they already use
No new app to learn. A question in Claude (or any MCP client) runs a governed function under the person's own identity, and answers from real data — and can render the whole thing as an auditable report.
A real question runs a governed function — under the user's identity (OBO), RLS enforced, read-only.
Right in the conversation, the agent renders the answer as a deterministic report — KPIs, charts, and the exact SQL — ready to save.
That exact report is saved to your platform — same KPIs, charts, and SQL, now governed and shareable across the team.
Interoperable by design
One governed server. Every AI your company already uses.
Ground speaks the open Model Context Protocol, so any MCP-compatible client connects to the same governed catalog. Switch models or tools whenever you like — your skills, functions, and access rules stay yours. No lock-in to a vendor, or to us.
…and any other MCP client — Windsurf, JetBrains, Zed, and more.
The governance portal
A web app your team owns — not a config file IT has to babysit
The same governed surface the AI sees, managed by your own people: catalog, groups, roles, and your data-source connections.
Every skill, function, and report in one place — typed and scoped.
Governance you can read
Who can see, run, and edit what — on one page
The whole access model exports as readable YAML and re-imports declaratively — edit in bulk, preview the diff, apply idempotently. No hidden rules, no console spelunking.
See Ground on your own data
Book a working session: we connect a governed agent to a sample of your data lake and you watch your team's real questions get answered — securely, in your cloud.