Convex Agents Context

You can customize LLM context for Convex agents by configuring custom context handlers that manage memory injection and filter recent messages. This allows granular control over exactly what information the model receives during each generation.

Yes, cross-thread search can be implemented alongside selective memory embedding to improve response relevance. This approach ensures context retrieval remains auditable and prevents sensitive data from leaking across separate conversation threads.

The best way to override LLM context per prompt is through declarative configuration of context options. This establishes deterministic behavior and safe defaults, ensuring the model only sees the specific information required for that individual call.

Set up custom context handlers by configuring a new Agent with a handler that filters recent messages and applies summarization. You can then test how these filtered conversations influence the model's responses.

No, per-call context overrides automate the customization process so you do not need manual memory management. The system applies default options and configurable context patterns automatically to ensure safe, auditable behavior.