Skip to content

Why TDB

The problem

Your data already exists — scattered across PostgreSQL, MySQL, SQL Server, Snowflake, and CSV files. Now every team wants AI agents and assistants to query it.

The hard part is doing that safely. Each source exposes data differently, and wiring an AI tool straight into a production database — with no access control and no record of what it asked — is how data leaks. Teams need governed, auditable, read-only access that works across the databases they already run, without handing everything to a third party.

And once it works, it has to stay affordable. An agent pays for every row and every schema it reads in its context window — in money, in latency, and in room left to think. A data layer that answers in bulky, repetitive JSON makes every question cost more than it should.

The common approaches today — and where they leave a gap

Most teams reach for one of three existing options. Each is good at what it does. None gives you governed, self-hosted, multi-source access to the data you already have.

Warehouse-native AI assistants — the AI layer built into a single cloud data warehouse. Excellent for data that already lives in that one warehouse. But your data has to be in that warehouse first; it doesn't govern your other databases; it isn't self-hosted on your own infrastructure; and the audit trail belongs to the vendor, not to you.

SaaS-embedded copilots — the assistant built into a productivity or SaaS suite. Great for documents and email. But it's SaaS: your data and query context flow through the vendor's cloud, it isn't a governed gateway over your production databases, and embedded copilots have been shown to be coaxable — via prompt injection — into exfiltrating data they could reach.

Do-it-yourself MCP servers — quick to stand up, but they typically ship with no authentication, no audit log, and no access control. (The widely-used reference open-source Postgres MCP server was archived after a critical security flaw — and is still downloaded tens of thousands of times a week because nothing better exists at this tier.)

What TDB does differently

TDB is a lightweight, self-hosted middleware that sits in front of your existing data sources — all of them — on your own infrastructure:

  • Self-hosted — data never leaves your perimeter; nothing is routed through a SaaS vendor.
  • Multi-source — connects to the databases you already run; no migration, no consolidation.
  • REST + SQL + MCP, simultaneously — one governed surface for dashboards, scripts, and any MCP-compatible AI tool.
  • Your existing SQL tools, unchanged — Metabase, DBeaver, psql and other PostgreSQL clients connect through the wire gateway with only a new host, port and password, and every statement is governed and audited like any other (enterprise).
  • Current with MCP — one endpoint speaks the 2026-07-28 MCP specification as well as the handshake-era revisions from 2024-11-05 to 2025-11-25, so new and older AI clients both connect without configuration. Claude Code uses the newest revision, and a free VS Code extension sets TDB up in one command.
  • Audited by default — every query, result, and tool call is logged to a tamper-evident audit log that you own, including which AI client made the call.
  • Governed — OAuth 2.1, single sign-on through your identity provider, role-based access control, column and row policies, a person's approval before sensitive columns are read, tenants that keep several teams or customers apart on one deployment, per-key rate limiting, and prompt-injection filtering.
  • Lean by design — results are sized for an agent's context window. Lean Mode sends rows as arrays and schemas as {column: type} — about a third fewer tokens on a 100-row answer, with the same answers — and can trim long text with a visible marker. Aggregates and filters run in the database, so an agent fetches the answer rather than the rows behind it.
  • Open-core — start free on the community edition; upgrade for production and multi-source.

In short: the others answer "AI on the data that already lives inside our platform." TDB answers "governed AI access to the data you already have, wherever it already is — without moving it, without surrendering the audit trail, and without wasting the agent's tokens."

Who it's for

Compliance-conscious mid-market teams — particularly in healthcare-tech and fintech — that need to make existing data accessible to AI agents without sending it to a SaaS vendor or consolidating everything into a single cloud warehouse.

Ready to try it? → Quickstart · Community Edition (free) · Compare editions