Getting Started¶
Install¶
With PDM:
pdm install -G :all
With pip:
pip install "sqldbagent[cli,postgres,langchain,langgraph,mcp,dashboard,docs,test]"
Start Local Infra¶
make up
make demo-migrate
That raises the local integration stack declared in infra/compose.integration.yaml, including the demo Postgres database and Qdrant.
Common First Commands¶
pdm run sqldbagent config validate
pdm run sqldbagent inspect tables postgres_demo --schema public
pdm run sqldbagent snapshot create postgres_demo public
pdm run sqldbagent profile unique-values postgres_demo customers segment --schema public
pdm run sqldbagent prompt export postgres_demo public
make dashboard-demo
The demo dashboard prefers durable Postgres checkpointing and durable Postgres-backed long-term memory when local persistence configuration is present, and it will tell you when it has to fall back to session-only memory.
The current dashboard flow gives you:
chat over the persisted LangGraph-backed agent
schema inspection through an interactive graph plus generated PNG/SVG fallback output when Mermaid rendering is unavailable
prompt review with token-budget breakdowns for the base prompt, final system prompt, saved enhancement, and live explored context
live prompt exploration that saves additional read-only database context back into the prompt artifact and can sync concise schema notes into long-term memory
retrieval-index management for the active saved snapshot, even when the current thread state has not populated snapshot identifiers yet
guarded query execution with read-only default behavior
What to Expect¶
Snapshots are stored under
var/sqldbagent/snapshots/<datasource>/<schema>/.Document bundles are stored under
var/sqldbagent/documents/<datasource>/<schema>/.Prompt bundles are stored under
var/sqldbagent/prompts/<datasource>/<schema>/.Retrieval manifests are stored under
var/sqldbagent/vectorstores/<datasource>/<schema>/.
Recommended Local Flow¶
Inspect and snapshot a schema.
Export docs, diagrams, and prompt context from the stored snapshot.
Optionally build retrieval indexes for snapshot documents.
Use the dashboard, MCP server, or LangGraph runtime against those stored artifacts before hitting the live database again.
Use live prompt exploration only when you want to enrich the stored prompt with current high-signal categorical values, index hints, or join paths from read-only profiling.