Agent Memory

This survey separates verbal feedback by when it acts and what it changes, helping agent builders distinguish task specification, test-time guidance, and parameter updates.

S3Gym tests whether agents can explore, judge outcomes, and reuse experience. Raw history, compressed memory, and training each help in some tasks, but none is a reliable default.

Exa’s internal agent clone combines writing examples, decision evals, and permission-scoped tools—a concrete pattern for encoding judgment without sharing full access.

Reusable agent skills transferred better when induced from subtasks and stored as text. Whole-task skills often hurt versus no memory, so retrieval alone is not evidence that a memory helps.

Compound engineering treats each agent mistake and product judgment as reusable context. The useful discipline is to invest in memory and review so the next feature becomes easier, not merely faster.

BATON explores robot subtasks independently, stores their solutions, and manages handoffs between them, replacing multiplicative whole-task search with additive composition.

Agent handoffs should preserve task-relevant predictive state, not blindly compress the transcript. The paper offers a three-part record for carrying decisions, evidence, and irreducible observations across sessions.

SheetCompass keeps spreadsheet structure as a hierarchical relation graph instead of flattening cells into text. Its agent memory retains task-relevant context across sheets, tables, and columns.

Persistent memory is a compute and product tradeoff, not just retrieval. Profiles need conflict detection, visibility, editing, and deliberate update cadence before agents can rely on them.

A useful agent knowledge base starts with abundant raw Markdown, then adds controlled tags, backlinks, and generated indexes. Scheduled agents can maintain it, but sync and taxonomy need explicit rules.

Stronger models can still behave like smart novices inside company-specific workflows. Agents need learning loops that compress experience into reusable context, procedures, and judgment without losing reliability.

Training directly on a private corpus can memorize it without producing useful behavior. Durable context learning likely needs synthetic study tasks and recursive self-improvement, not a single fine-tuning pass.

Always-on production agents can turn releases and operational events into tailored monitoring tasks. The useful pattern is persistent context plus scoped triggers, checks, and escalation.

SkillProx evolves reusable agent instructions through measured retries, regression rollback, and utility-based pruning, reporting a 3-point accuracy gain over its strongest baseline.

Blast Radius predicts which code and context a prompt will reach, then reversibly archives unused history. Across seven OpenAI models, it cut token consumption by 17–26%.

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.

Chained RLM repeatedly gives the same model a fresh context while preserving summaries, a blackboard, and artifacts—a concrete harness pattern for limiting error buildup.

PRECOG precomputes SSM corpus states and injects them at query time, avoiding context re-ingestion. On a 1.2B edge model, reported prefill fell from about 27 seconds to under 6 ms.

This survey organizes long-horizon agent weaknesses into five capability gaps, offering a useful checklist for harness design and evaluation rather than a new implementation.

Shared agents need a different harness: action-boundary security, selective group memory, per-user privacy, and routing that decides both who receives information and when the agent should speak.

UniMem routes novel experience into retrieval memory and consolidates recurring patterns into expandable parameters, offering a design for agents that learn across unlabelled task streams.

MemLens scores individual agent memories, stores them by value, and exposes quality, latency, and token tradeoffs—an inspectable alternative to retaining every interaction equally.

Netflix’s performance-agent playbook feeds production profiles and exact deployed code to agents, then relies on canaries and engineers to validate proposed optimizations.

A delayed KV-eviction policy helps when memory reuse is sharp and observable, but loses its edge on natural-text benchmarks—a useful warning for agent-memory claims.

LLM synthesis breaks simple source pointers when facts merge, change, or outlive one input. Model provenance inside the context graph so agents can filter trust, debug derivation, and apply deletion policies.

Atlan’s agent experiments argue for shared, versioned context instead of per-agent memory: a portable layer for business facts, skills, norms, retrieval, and feedback across changing harnesses.

holaOS is a local-first workspace where Claude Code, Codex, and a built-in agent share memory, tools, skills, and apps. It reduces setup duplication, but its modified Apache license adds distribution conditions.

Claude-Mem records agent activity, compresses it into persistent project memory, and retrieves selected history later, trading extra local services and stored data for cross-session continuity.

LoopX gives long-running coding-agent work durable goals, gates, ownership, evidence and quotas across bounded turns. It is useful when chat history and a scheduler no longer provide enough control.

An open-source memory hub turns agent conversations, workflows, docs, and code into governed assets that can be reused across sessions and roles, reducing repeated project setup.

TradingAgents is an open-source LangGraph reference for role-based agent debates, durable memory, checkpoint recovery, and provider portability, with trading as its test domain.

ai-memory gives coding CLIs a shared, Git-backed memory and bounded handoffs, so work can move between agents without treating stale recollections as current code truth.

A 100-person factory built a multi-agent sales system around organized company memory, specialist roles, and human approval. The useful pattern is retrieval and governance, not custom model training.

eve’s Chat SDK channel lets one agent span messaging adapters while retaining threads, approval cards, proactive sends, webhook handling, and overridable defaults.

Anthropic's harness pattern for multi-session agents: an initializer sets up the env, a JSON feature list, and progress files; each session then ships one feature, verified end-to-end and committed to git.

Compound Engineering packages a six-stage agent workflow into 32 skills for Codex, Claude Code, Cursor, and others. Its core idea is to persist plans, reviews, and solved-problem notes as future context.

Atlas links agent sessions to commits and shares local project memory across Claude Code, Codex, and ACP agents. It makes agent provenance queryable, but macOS is the only supported platform.

ECC packages skills, hooks, memory, orchestration, and security controls for multiple coding-agent harnesses, but its breadth makes selective installation and verification essential.