Agno 3.0.1 stops response time from scaling with chat length

Agno fixed a bug that made response time grow with every added conversation turn. Latency creep like that is easy to miss in testing. It shows up only in production, once a long-running agent session has racked up enough turns for replies to visibly slow down.

Before the fix, each reply rebuilt every tool's schema from scratch and reloaded the whole session history, so more turns meant more work per reply. The 3.0.1 release changes both. Tool schemas are now derived once and cached across runs, and session history loads incrementally per turn instead of reloading the full conversation each time.

Most of the rest of the changelog is MCP and Gemini correctness work. Function.process_schema_for_strict no longer raises a KeyError on schemas that omit properties, a pattern common in MCP server schemas, per Agno's 3.0.1 release notes.

A crash is hard to miss. A chat that gradually gets slower turn by turn is easy to ship past.