Sign InOpen Brain
arXivPaperNeeds Review

PsychoAgent: An Affect-Sensitive Cognitive Architecture for Conflict-Aware Memory in LLM Agents

PsychoAgent separates factual and affective memory, then reranks relevant memories by salience. It retrieved more conflict-critical context, but output-quality differences were not significant.

arXiv · Aug 7, 2026
Open Source Open MarkdownOpen JSON
Source Summary

PsychoAgent filters affective memories by semantic relevance, reranks them by salience, and combines them with factual memory through an executive controller. In **three conflict scenarios**, retrieval scored **0.933**, versus 0.500 and 0.667 for two baselines.

Practical Implication

For memory systems where importance is not captured by similarity alone, test a second ranking signal after relevance filtering. Keeping factual and affective stores separate also makes the retrieval policy easier to inspect.

Agent-Ready Context
PsychoAgent filters affective memories by semantic relevance, reranks them by salience, and combines them with factual memory through an executive controller. In **three conflict scenarios**, retrieval scored **0.933**, versus 0.500 and 0.667 for two baselines.

For memory systems where importance is not captured by similarity alone, test a second ranking signal after relevance filtering. Keeping factual and affective stores separate also makes the retrieval policy easier to inspect.

Five blinded raters assessed **27 outputs**. The full system led by **+0.22 SD** after within-rater standardization, but corrected pairwise differences were not statistically significant, and retrieval incurred a small semantic-similarity cost.
Connected Context · Feed7 Judgment

This adds affective salience as an inspectable second-stage ranking signal after semantic relevance and supports separating memory types before an executive controller combines them. It extends value-aware retrieval beyond generic utility scores, but the small, statistically inconclusive output study narrows the claim to a promising architecture and retrieval result rather than demonstrated behavioral reliability.

MemLens: A Value-Aware Memory Management System with Interactive Analytics for LLM-based AgentsMemLens makes memory value observable at record level; PsychoAgent provides a more specific implementation consequence by applying salience only after relevance filtering and keeping affective and factual stores separate.Citation Needed: Provenance for LLM-Built Knowledge Graphs — Daniel Chalef, Zep AISeparate memory stores and reranking improve policy inspectability, but provenance remains necessary to explain which source memories produced a combined response and to support correction or deletion.Wearing the Agent: From Group Chats to Glasses — Sai Krishna RallabandiAffective salience could increase retrieval priority without establishing disclosure permission; the shared-agent candidate therefore adds a governance prerequisite of per-user privacy, audience filtering, and silence policies.A Taxonomy of Cognitive Capability Gaps in Generative and Agentic AIPsychoAgent is a concrete memory and control architecture within the taxonomy’s broader durable-state and adaptation gaps, while its limited evaluation confirms that addressing one gap does not establish overall agent reliability.
Context Map
agentresearch#agent-memory#retrieval#agent-reliability
Uncertainty
Five blinded raters assessed **27 outputs**. The full system led by **+0.22 SD** after within-rater standardization, but corrected pairwise differences were not statistically significant, and retrieval incurred a small semantic-similarity cost.