AI memory: questions and answers
How AI memory works: types of memory, long-term memory and shared memory.
- Is RAG the same as AI memory?
No. RAG (retrieval-augmented generation) is a retrieval technique: it finds matching passages in a source, usually existing documents, and puts them into the prompt. Memory is keeping, updating and later recalling what the AI learns, such as preferences, decisions and corrections, and many memory systems use RAG to find those memories again, but RAG alone does not decide what to keep or which fact is out of date.
- What are the types of memory in AI agents?
AI agents have short-term memory, the context window and session state of one task, and long-term memory outside the model. Long-term memory is sorted by content (episodic, semantic, procedural), by storage form (instruction files, agent notes, a retrieval index, a memory service) and by scope. For a team, scope matters most: does the memory belong to one user, one project, one agent or everyone?
- What is A-Mem (Agentic Memory for LLM Agents)?
A-Mem, short for Agentic Memory for LLM Agents, is a 2025 research paper by Wujiang Xu and colleagues, listed for NeurIPS 2025. It stores each memory as a structured note with keywords, tags and a context description, links it to related notes, and lets new notes update older ones, based on the Zettelkasten method. The code is open source on GitHub.
- Why does AI forget what you told it earlier?
A language model only knows what you told it if that text is in its context window, the text it sees while answering. A new chat gets your instructions, project context and a selection from memory, but not your earlier conversations in full. In a long chat, accuracy also drops as the window fills, and older messages can be dropped or summarized.
- Does OpenClaw have memory, and what does the Mem0 plugin add?
Yes. OpenClaw keeps memory as plain Markdown files in its workspace, such as MEMORY.md and daily notes, and searches them with memory_search. The Mem0 plugin (@mem0/openclaw-mem0) takes over OpenClaw's memory slot: it extracts durable facts after each turn, recalls relevant ones before the next, and stores them in Mem0's cloud or in a self-hosted open source setup.
- Mem0 vs Zep vs Supermemory vs Letta: which agent memory tool should I use?
Pick Mem0 for a memory API that extracts facts per user or agent, Zep when facts change over time and you need to know what was true when, and Supermemory for a usage-priced memory and retrieval API with plugins for coding tools. Pick Letta if you are building the agent itself. All four are developer platforms first, though each now also connects to some existing AI clients.
- How does Anthropic's memory tool for Claude agents work?
The memory tool in the Claude API lets an agent keep files in a /memories directory across conversations. Claude only asks for operations such as view, create, str_replace, insert, delete and rename; your application runs them on storage you control. It pairs with context editing and compaction, so an agent can save what matters before old context is cleared or summarized.
- How do you evaluate the memory of an LLM agent?
LLM agent memory is usually evaluated with benchmarks that feed an agent long, multi-session conversations and then ask questions about them. The best known are LoCoMo (2024, about 300 turns over up to 35 sessions), LongMemEval (ICLR 2025, 500 questions on five abilities) and MemoryAgentBench (2025, incremental multi-turn input). For a team, the most useful test is a small set of its own questions with known answers.
- How can a team share one AI memory across ChatGPT, Claude and Cursor?
Put one memory store outside the tools, connect ChatGPT, Claude, Cursor and your own agents to it, and give each person a private scope next to a shared one with clear read and write rights. The built-in memory of ChatGPT and Claude belongs to one person, and shared projects or instruction files such as CLAUDE.md and AGENTS.md only share context inside the tools that load them.
- What is SimpleMem (Efficient Lifelong Memory for LLM Agents)?
SimpleMem is a memory framework for LLM agents from a 2026 paper by Jiaqi Liu and colleagues, mainly at UNC-Chapel Hill, identified by its repository as an ICML 2026 paper. It filters out low-value dialogue, rewrites the rest into short self-contained facts with resolved names and absolute dates, merges related facts at write time, and retrieves them through semantic, keyword and metadata indexes. The code is open source on GitHub.
- How should an AI agent manage its memory: what to store, update and forget?
AI agent memory management is the set of rules for what an agent writes to memory, how it updates or removes memories when facts change, and when old memories expire. The 2025 Mem0 paper lets an LLM choose add, update, delete or no change for each new fact, and Anthropic's docs advise capping memory size and deleting files that have not been used for a long time.
- How do you give an LLM long-term memory?
A language model forgets everything between calls, so long-term memory always lives outside the model. The application saves what matters to a store (memory files, a database, a vector index or a graph) and puts the relevant part back into the prompt next time. You can build that with a memory tool such as Anthropic's, a framework store such as LangGraph's, or a memory service such as Mem0.
- What is memory poisoning in AI agents, and how do you defend against it?
Memory poisoning means getting false or malicious records into an AI agent's memory, so the agent repeats them in later sessions, often for other users. The MINJA paper (Dong et al., 2025) showed that a regular user can do this through normal queries alone, with an average injection success of 98.2% across its tests. OWASP lists Memory & Context Poisoning as ASI06 in its Top 10 for Agentic Applications.
- What is a memory OS for AI agents (MemoryOS, MemOS, MIRIX)?
A memory OS is a design in which an AI agent's memory is managed like an operating system manages computer memory: in tiers, with rules for moving, updating and retrieving what is stored. The idea started with MemGPT (2023). Best known since are MemoryOS ("Memory OS of AI Agent", 2025), MemOS from a 39-author team, and MIRIX with six memory types. All are research projects with open code.
- What are the privacy risks of AI agent memory?
AI agent memory can leak what earlier users asked, because stored records are pulled back into the prompt and a crafted question can make the agent repeat them. The 2025 paper "Unveiling Privacy Risks in LLM Agent Memory" showed this with an attack called MEXTRA on a healthcare agent and a shopping agent. Leakage grew with bigger memories, more retrieved records and more attack attempts.
- What is Hindsight agent memory and how does it work?
Hindsight is an open-source, MIT-licensed long-term memory system for AI agents, built by Vectorize. Instead of storing raw conversations, it extracts facts, links them to entities and time, and exposes three operations: retain, recall and reflect. Recall runs semantic, keyword, graph and temporal search in parallel. You can self-host it on Postgres or use Hindsight Cloud.
- How does memory work in an n8n AI agent?
In n8n, memory is a sub-node you attach to an AI Agent node. It keeps the recent chat history for one session, identified by a session ID, and sends the last few interactions back to the model. Simple Memory keeps it in the workflow data and is not safe in queue mode; Postgres or Redis Chat Memory store it in a database. None is long-term knowledge on its own.
- What is ReasoningBank, the memory framework that learns from failures?
ReasoningBank is a memory framework from a 2025 paper by Siru Ouyang and colleagues, mostly at Google Cloud AI Research, accepted at ICLR 2026. Instead of storing raw past attempts, the agent judges its own attempts as success or failure and distills each into a few short strategy notes with a title, description and content. On new tasks it retrieves the most similar note. Lessons from failures are kept too.
- What is Vertex AI Memory Bank and how does it work?
Memory Bank is Google Cloud's managed long-term memory service for AI agents, launched as part of Vertex AI Agent Engine and now documented as Agent Platform Memory Bank in Gemini Enterprise Agent Platform. It uses Gemini models to extract facts and preferences from conversations, merges them with existing memories per scope, such as a user ID, and lets an agent retrieve them later by similarity search.
What does your AI forget?
- 01You tell us which AI you use and where it keeps starting over.
- 02We show what a shared memory changes there, on your own examples.
- 03You keep the plan, with or without Bombos.
