Query routing for AI database agents: send the question to the right source before SQL runs
Not every database question should hit the same source.
“What did revenue do last quarter?” can use a warehouse. “Is this customer currently blocked?” may need live operational data. “Export every user who churned” may need approval before it runs at all.
AI database agents need query routing before SQL generation becomes execution.
Routing is part of safety
Many demos treat the database as one box. Production systems rarely look like that.
There are primaries, read replicas, reporting warehouses, materialized views, caches, audit tables, and third-party APIs. They have different freshness, permissions, latency, and risk profiles.
A good MCP database layer should choose the safest valid path for the question.
Related: AI database connector architecture.
Route by intent, not just by table name
Routing starts with the intent of the question.
- Operational status may require live data.
- Trend analysis may belong on a warehouse.
- Customer support answers may need tenant-scoped approved views.
- Large exports may require dry-run and approval.
- Unclear write requests should fail closed.
The agent should not decide this from table names alone. The tool layer needs metadata about freshness, sensitivity, allowed operations, and expected result shape.
Related: Schema context for AI database agents.
Freshness changes the route
A cached or replicated answer can be correct for one question and unsafe for another.
For example, a weekly usage summary can tolerate a delayed warehouse table. A payment status answer probably cannot.
That is why routing decisions should include freshness windows and source metadata.
Related: Freshness windows for AI database answers.
Return the route as evidence
Do not hide routing from the user.
A useful result can say:
Answered from the analytics warehouse, refreshed at 01:05 UTC, using approved view
monthly_revenue_summary. Live payment status was not queried.
That makes the answer easier to trust, debug, and audit.
Related: Tool result contracts for AI database agents.
Where Conexor fits
Conexor connects databases and APIs to MCP-compatible AI clients through controlled infrastructure.
The point is not just to make data reachable. It is to make the route explicit: which source, which scope, which freshness window, which policy, and which evidence produced the answer.