Agno v3.1.11

Agno v3.1.11 introduces significant multi-agent orchestration enhancements along with critical bug fixes. Key features include: - Harness Agents in Teams and Workflows : ClaudeAgent and CodexAgent can now operate as Team members or Workflow steps (including inside control blocks like Parallel, Loop, Condition, and Router). Harness agents manage independent models, tools, workspaces, and conversation contexts, with containerized execution supported via sandbox= . - Session Search : New endpoints GET /sessions/filter-schema and POST /sessions/search support nested AND/OR/NOT filtering on session properties, IDs, types, and timestamps, matching the Traces API with optimized SQL-level pushdowns for PostgreSQL and SQLite. - Company Tasks & Routines : Added lifecycle status support for company tasks and administrative endpoints ( /cancel , /archive , /restore , DELETE ) for company routines. Notable bug fixes: - StepOutput.to dict() retains primitive types (ints, floats, bools) instead of casting to strings across HITL resumes and session reloads. - ReaderFactory.create reader() now generates fresh reader instances per call instead of reusing singletons, preventing shared mutable state across Knowledge instances. - AgentOSClient.continue agent run removes the non-existent stream parameter; developers must use continue agent run stream . - Fixes to Google Calendar timezone handling, RedisDb query limit pass-throughs, MMRReranker zero limits, and Hugging Face SDK parameter mapping.

• Switch from passing stream to AgentOSClient.continue agent run to calling continue agent run stream instead.
• Verify workflow deserialization handling if relying on stringified primitives from StepOutput.to dict() , as they now preserve native ints, floats, and booleans.
• Adopt the new POST /sessions/search endpoint and filter schema if custom session querying was previously performed on the client side.