sqldbagent

Service-first database intelligence for safe inspection, profiling, snapshotting, guarded querying, retrieval preparation, and agent-facing execution over relational databases.

Note

The public docs focus on the product and operator surface. Working notes, experiments, and repo memory stay in docs/_internal/.

Safe by Default

Every agent-facing query goes through guardrails, row limits, and read-only enforcement before execution.

One Core, Many Surfaces

CLI, dashboard, MCP, LangChain, and LangGraph all sit on top of the same normalized metadata and service layer.

Durable Artifacts

Snapshots, docs, diagrams, prompts, and retrieval indexes are stored so context can be reloaded without re-introspection.

Built for Real Databases

The initial target is Postgres and MSSQL, with SQLite used as a lightweight smoke and local E2E target.

Architecture

        flowchart LR
    A["Datasources<br/>Postgres, MSSQL, SQLite"] --> B["Engines + Safety"]
    B --> C["Normalized Metadata Core"]
    C --> D["Snapshot + Profile + Diagram + Docs"]
    D --> E["CLI"]
    D --> F["Dashboard"]
    D --> G[MCP]
    D --> H["LangChain / LangGraph"]
    H --> I["LangSmith Tracing"]

Guide