Agent memory: changing facts and reusable procedures

An agent may need to resolve what is true now, or reuse how it solved a task before. These are related but distinct memory problems.

For changing conversational facts

APEX-MEM: Agentic Semi-Structured Memory with Temporal Reasoning for Long-Term Conversational AI

An append-only property graph represents dated, entity-centric events and a multi-tool retrieval agent resolves conflicting or evolving information at query time.

Reported results include 88.88% on LOCOMO and 86.2% on LongMemEval.

Pipeline: raw conversation becomes typed entities, dated facts and events in a graph a SQL agent queries to answer. — Figure 1: End-to-end pipeline for constructing and querying APEX-MEM Graph, showing data flow from unstructured conversation to GraphSQL QA Agent. Taken from the paper.

Read the paper page PDF Citation + BibTeX

For repeated procedures

APEX-EM: Non-Parametric Online Learning for Autonomous Agents via Structured Procedural-Episodic Experience Replay

A typed Procedural Knowledge Graph stores complete successful and failed traces, retrieved through semantic search, structural signatures, and graph traversal—without changing model weights during deployment.

On a blind KGQAGen-10k test split with frozen memory, the paper reports 73.7% versus 42.0% without memory.

PRGII loop: plan, retrieve past successes and failures, generate, iterate, then quality-gate commit into the memory graph. — Figure 1(a): The PRGII workflow. A task flows Plan to Retrieve to Generate and Iterate to Ingest; retrieved successes and failures condition Generate, and the quality gate commits to memory M. Taken from the paper.

Read the paper page PDF Citation + BibTeX

Opportunities for future work

A natural next step is to test how temporal event memory and procedural experience replay can work together across shared settings and model backbones.