Convex Agents Debugging

Debug unexpected Convex agent behavior by hooking into logging hooks and context handlers to surface raw LLM requests, context messages, and database state for analysis. This reveals exactly what inputs the agent receives.

Trace LLM tool calls in a Convex database by querying internal tables to monitor external utilities invoked by the agent. This helps diagnose errors and verify data integrity across production-like setups.

Inspect LLM context messages by logging context handlers to see exactly what context the LLM receives and in what order. You can also fetch context manually to simulate and verify the agent's view.

Yes, you can diagnose Convex agent errors in a debugging playground by reproducing tooling patterns and logging internal state. This allows you to analyze message structure and trace tool calls safely before production.

LLM output mismatches database state when context or tool calls are misaligned. Resolve this by inspecting database tables to query messages, threads, and tool calls to verify data integrity and message ordering.

Yes, setting up logging hooks is required to analyze LLM message structure in Convex. Hooking into agent components captures raw requests, roles, and content metadata needed to assess ordering and diagnose issues.