AgentData is a Data Fabric / Semantic Layer that auto-discovers business entities across heterogeneous databases, maintains a unified semantic model, and serves analytical queries through a REST API and MCP — so people and LLMs can ask questions of your data in plain language.
This is the public repository for AgentData: the on-prem connectors you install to connect private databases, the connector protocol, and getting-started docs. The product itself (backend, UI, semantic engine) is hosted — you don't run it from here.
| Path | What it is |
|---|---|
connectors/ |
On-prem connectors you install on your own network |
connectors/relational/ |
The relational connector — PostgreSQL, MySQL, SQL Server, SQLite (read sources and/or write staging DBs) |
connectors/PROTOCOL.md |
The job contract between a connector and the backend (build your own in any language) |
QUICKSTART.md |
Connect a database and run your first flow |
A connector lets the hosted AgentData backend reach a database on your network without exposing it to the internet and without copying its data out. It runs inside your network, holds the database credentials locally, makes only outbound HTTPS calls (it polls — no inbound port), and returns only metadata and query/flow results. Raw rows and credentials never leave your machine.
AgentData (hosted) ◄── outbound HTTPS poll ── connector (your network) ──► your database
│ │
└──────────── metadata / results only ─────────┘
See QUICKSTART.md, then the connector you need — e.g. connectors/relational/ for SQL databases.
MIT — you're free to read, run, and adapt the connectors for connecting your own systems to AgentData.