RAG
Ground Dash’s SQL work in business definitions and query examples.
This walkthrough follows Demo OS at the reviewed revision, which pins Agno 2.8.1. Use the application’s checked-in requirements.
Dash combines database tools with searchable table descriptions, business rules, validated query patterns, and saved lessons. Its instructions ask the model to consult that context; the resulting answer still needs validation against the data.
Dash: SQL grounded in known patterns
The Analyst and Engineer receive knowledge=dash_knowledge and search_knowledge=True. The model can retrieve evidence through a tool before constructing a query. The shipped agents do not enable automatic add_knowledge_to_context retrieval.
Loading knowledge
After the sample database setup, load the corpus explicitly:
docker exec -it demo-os-api python -m agents.dash.scripts.load_knowledgeApp startup registers runtime services and schedules; it does not load this knowledge. Files live under agents/dash/knowledge.
How retrieval gets injected
In Dash, the model sees the result of its selected knowledge-search tool calls. If you extend an Agent with add_knowledge_to_context=True, the pinned framework instead retrieves references during message preparation and adds them to the user message. That is Agent behavior, not a special AgentOS system-prompt step.
See it in action
Ask Dash for current MRR or churn by plan. Inspect retrieval and SQL tool calls, then check the answer against the sample dataset and business definitions. Record useful corrections through the demo’s learning tools and inspect their stored form.
When hybrid search isn't enough
The configured PgVector path combines normalized PostgreSQL full-text and vector scores with a weighted sum. It is not BM25 or RRF. Dash does not ship a reranker; adding one requires choosing candidate and output limits explicitly. Measure retrieval quality on representative questions before changing the pipeline.
Next
agents/dash/agents/analyst.py · agents/dash/agents/engineer.py · db/session.py