camel-ai
[Feature Request] Implement Shared Memory Mechanism for Workforce Handoff
Required prerequisites
- [x] I have searched the Issue Tracker and Discussions that this hasn't already been reported. (+1 or comment there if it has.)
- [x] Consider asking first in a Discussion.
Motivation
Currently, the workforce system does not store the complete function-calling trajectory of each worker in memory. This creates challenges in task handoff, as subsequent agents do not have sufficient context to continue processing efficiently.
Proposed Solution:
Develop a shared memory mechanism to retain and share the memory context across all agents.
Modes of Memory Sharing:
All (High Priority): Share complete memory across all agents, ensuring the reasoning agent and other critical agents receive the full context.
Memory Pool: Create a configurable memory pool where relevant context can be shared between selected agents.
No Sharing (The current workforce): Disable memory sharing for specific tasks that do not require continuity.
Key Tasks:
Implement memory initialization configuration to define the desired memory-sharing mode.
Develop a memory pool manager to dynamically allocate and update shared context.
Enable seamless integration with the workforce’s function-calling trajectories.
Test and validate the performance of different memory-sharing modes.
Reference
- https://github.com/openai/openai-agents-python
Solution
No response
Alternatives
No response
Additional context
No response
Related memory features in CAMEL:
https://github.com/camel-ai/camel/blob/58a4f6ec0879040fbd1b77840cbd936160f80481/camel/agents/chat_agent.py#L358
https://github.com/camel-ai/camel/blob/58a4f6ec0879040fbd1b77840cbd936160f80481/camel/agents/chat_agent.py#L422
@GitHoobar im going to take a bit of time to implement this properly, if you want this done quickly, feel free to re-assign someone else.
hey @JINO-ROHIT , @sunchengxuanivy already did some pre study on this and I plan to schedule a joint discussion in the middle of this week, feel free to join if you have time!
