MemCollab: Cross-Agent Memory Collaboration via Contrastive Trajectory Distillation
arXiv:2603.23234v1 Announce Type: new Abstract: Large language model (LLM)-based agents rely on memory mechanisms to reuse knowledge from past problem-solving experiences. Existing approaches typically construct memory in a per-agent manner, tightly coupling stored knowledge to a single model's reasoning style. In modern deployments with heterogeneous agents, a natural question arises: can a single memory system be shared across different models? We found that naively transferring memory between agents often degrades performance, as such memory entangles task-relevant knowledge with agent-specific biases. To address this challenge, we propose MemCollab, a collaborative memory framework that constructs agent-agnostic memory by contrasting reasoning trajectories generated by different agents on the same task. This contrastive process distills abstract reasoning constraints that capture shared task-level invariants while suppressing agent-specific artifacts. We further introduce a task-a
arXiv:2603.23234v1 Announce Type: new Abstract: Large language model (LLM)-based agents rely on memory mechanisms to reuse knowledge from past problem-solving experiences. Existing approaches typically construct memory in a per-agent manner, tightly coupling stored knowledge to a single model's reasoning style. In modern deployments with heterogeneous agents, a natural question arises: can a single memory system be shared across different models? We found that naively transferring memory between agents often degrades performance, as such memory entangles task-relevant knowledge with agent-specific biases. To address this challenge, we propose MemCollab, a collaborative memory framework that constructs agent-agnostic memory by contrasting reasoning trajectories generated by different agents on the same task. This contrastive process distills abstract reasoning constraints that capture shared task-level invariants while suppressing agent-specific artifacts. We further introduce a task-aware retrieval mechanism that conditions memory access on task category, ensuring that only relevant constraints are used at inference time. Experiments on mathematical reasoning and code generation benchmarks demonstrate that MemCollab consistently improves both accuracy and inference-time efficiency across diverse agents, including cross-modal-family settings. Our results show that the collaboratively constructed memory can function as a shared reasoning resource for diverse LLM-based agents.
Executive Summary
The article proposes MemCollab, a collaborative memory framework that enables cross-agent memory sharing for large language model-based agents. By contrasting reasoning trajectories from different agents, MemCollab constructs agent-agnostic memory that captures task-level invariants while suppressing agent-specific biases. This approach improves accuracy and inference-time efficiency across diverse agents, including cross-modal-family settings, demonstrating the potential for collaboratively constructed memory to function as a shared reasoning resource.
Key Points
- ▸ MemCollab enables cross-agent memory sharing for large language model-based agents
- ▸ The framework constructs agent-agnostic memory by contrasting reasoning trajectories
- ▸ MemCollab improves accuracy and inference-time efficiency across diverse agents
Merits
Improved Accuracy
MemCollab consistently improves accuracy across diverse agents, including cross-modal-family settings
Increased Efficiency
The framework enhances inference-time efficiency by conditioning memory access on task category
Demerits
Complexity
MemCollab requires contrasting reasoning trajectories from different agents, which may add computational complexity
Expert Commentary
The proposed MemCollab framework addresses a critical challenge in large language model-based agents, namely the inability to share memory across agents. By constructing agent-agnostic memory, MemCollab enables diverse agents to collaborate and improve their reasoning capabilities. The use of contrastive trajectory distillation to suppress agent-specific biases is a novel approach that demonstrates the potential for collaboratively constructed memory to function as a shared reasoning resource. However, further research is needed to fully explore the implications and applications of MemCollab.
Recommendations
- ✓ Further experimentation with MemCollab in real-world scenarios to demonstrate its practical applications
- ✓ Investigation into the potential of MemCollab to facilitate knowledge transfer between human and artificial intelligence systems
Sources
Original: arXiv - cs.AI
