Memory
Also written: Agentic memory, Copilot Memory, ChatGPT Memory, Memory management, Retrieves memory, Memory synthesis, Durable project memory, State management
AI's ability to recall past interactions and learned context to inform future responses.
AI memory is what the AI remembers about you and your conversations. This allows the AI to build on previous interactions, making conversations feel more continuous and personalized.
For example, if you tell an AI assistant your name and ask it to remember it, a good memory system would allow the AI to recall your name in a later conversation without you having to repeat it.
AI memory refers to the mechanism by which a model retains and accesses information from prior inputs, conversations, or training data to influence its current output. This can range from short-term context windows that hold recent dialogue to more persistent storage mechanisms that allow for long-term personalization and learned behaviors. The effectiveness of memory is crucial for maintaining conversational coherence, user personalization, and task completion over extended interactions.
Recent developments show AI systems are being designed to retain information across sessions and interactions. This allows for more sophisticated agentic behavior, like learning from user corrections or maintaining context in long-running tasks. Companies are exploring ways to manage and transfer this learned information, making AI more persistent and collaborative.
- psychology Gemini 3.8 Live Extended ThinkingGoogle · Gemini Models ·
- ChatGPT Space: sharing, data, and controlsOpenAI · OpenAI Help Center ·
- Agentic autofix now uses Copilot MemoryMicrosoft · GitHub Changelog: Copilot ·
- When Hiring Becomes Agent-Mediated: Evaluating Access and Recurrence in Two-Agent Résumé ScreeningAnthropic · arXiv cs.AI ·
- Run long horizon tasks with CodexOpenAI · OpenAI Developers Blog ·
- Publish and share artifactsAnthropic · Claude Help Center ·
- Import and export your memory from ClaudeAnthropic · Claude Help Center ·
- Lindy Teammate: Flo Crivello on Multiplayer Agents, Memory & Why He'd Ban the Chinese Models He UsesAnthropic · The Cognitive Revolution ·
- OpenAI Cookbook: Move Oracle agent memory cookbook to vector databases (#2873)OpenAI · OpenAI Cookbook ·
- OpenAI Cookbook: Unarchive Agents SDK memory cookbooks (#2868)OpenAI · OpenAI Cookbook ·
- Dreaming: Better memory for a more helpful ChatGPTOpenAI · OpenAI News ·
- Ep 87: Gemini Co-Lead on World Models, RL's Next Domains & Continual LearningGoogle · Unsupervised Learning ·
- Personalizing ChatGPTOpenAI · OpenAI News ·
- Durable Execution and the Infrastructure Powering AI AgentsOpenAI · AI + a16z ·
- Inside Praktika's conversational approach to language learningOpenAI · OpenAI News ·
- Samsung and SK join OpenAI’s Stargate initiative to advance global AI infrastructureOpenAI · OpenAI News ·
- Anthropic Claude
- Anthropic is enabling users to import and export their memory from Claude, and exploring agent-mediated processes that rely on recalling past information.
- Google Gemini
- Google Gemini's models are being developed to handle complex tasks that likely require retaining context and information over time.
- Microsoft Copilot
- Microsoft Copilot is directly integrating memory features, allowing the system to learn from user corrections to improve its performance.
- OpenAI ChatGPT
- OpenAI is enabling features like ChatGPT Space for sharing and collaboration, which implies a need for persistent data and context, and is working on tools for long-horizon tasks that require sustained memory.
- Context Window
- The context window is the amount of recent text the AI can consi
- Training Data
- Training data is the vast dataset used to initially teach the AI,
- Fine-tuning
- Fine-tuning adapts a model to a specific task or style using new da
The concept of memory in AI is fundamental to artificial intelligence research, evolving from early work on state representation and knowledge bases. In the context of large language models, memory is often implemented through attention mechanisms within the model's architecture and explicit storage of past interactions.
In ChatGPT or Claude, start a new conversation and tell the AI your favorite color. Then, start a new chat session and ask the AI what your favorite color is. See if it remembers from the previous conversation.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- PromptExport Claude memory for backup or migration to another AIExport all stored memories, context, and instructions from Claude in a single code block for backup or migration. · 5 min
- Setting tipEnable Copilot Memory for Agentic AutofixActivate Copilot Memory to provide context and learning for automated security fixes. · 1 min
- ToolAgentic autofix with Copilot MemoryAutomated security fix generation that uses past fixes for context and learns new patterns. · 5 min
- HeuristicDurable project memory prevents AI drift in long tasksUse markdown files for spec, plan, constraints, and status to keep AI on track. · 5 min
- ChecklistReview Claude memory import and export settingsChecklist for navigating and understanding Claude's memory import and export features, distinguishing between new and legacy experiences. · 10 min
Questions people ask
- How does AI memory work?
- AI memory can work in several ways, including using a limited window of recent conversation text or storing key details from past interactions in a separate database.
- Is AI memory the same as my chat history?
- AI memory often uses chat history as a basis, but it can also involve specific features designed to retain and recall information more effectively for personalization.
First seen on Subvolts Oct 1, 2025. Updated Oct 12, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.