Skip to content

Conversation

@syed-mohsin-s
Copy link

Summary

Implements the Orchestration Memory feature for cross-agent memory sharing in multi-agent workflows.

Related Issues

fix #1695

Changes

New Files

  • core/framework/runtime/orchestration_memory.py - Core module with:
    • OrchestrationScope enum (WORKFLOW, AGENT_GROUP, GLOBAL)
    • OrchestrationMemory class with async share/read operations
    • Aggregation strategies (list, dict, first, last)
    • Agent groups and subscriptions
  • core/tests/test_orchestration_memory.py - 23 comprehensive tests

Modified Files

  • core/framework/runtime/shared_state.py - Added StateScope.ORCHESTRATION
  • core/framework/runtime/agent_runtime.py - Integrated orchestration memory

Usage Example

# Share data from one agent
await runtime.orchestration_memory.share(
    key="results", value=data, from_agent="search_agent"
)
# Read in another agent
data = await runtime.orchestration_memory.read_shared(
    key="results", from_agent="search_agent"
)
Testing
All 23 tests pass ✅.

@syed-mohsin-s
Copy link
Author

The orchestration memory module is solid and ready for the next phase of integration hooking it into the GraphExecutor and the checkpoint system as outlined in the
RESUMABLE_SESSIONS_DESIGN.md
. Would you like to proceed with that, or test any specific flow in more depth?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

[Feature]: Orchestration Memory for Cross-Agent Coordination

1 participant