How to Give Claude Long-Term Memory

The Direct Answer: Claude Forgets by Default, and How to Fix It

Each new Claude chat starts blank. By default the model is stateless: it does not recall what you said in a previous conversation, and through the API every call is independent unless you pass the history yourself. That is the gap behind the phrase "claude long term memory."

There are four practical ways to add persistence. Turn on Claude's built-in memory feature in the consumer app. Use Projects to keep a workspace of context that loads into every chat inside it. For developers, add the API memory tool so an agent can write and read its own memory files. And for recall you want to keep across any assistant, store facts in an external memory layer.

Pick the method that matches where you use Claude: the app, a focused project, or your own code.

• Default behavior: no memory across separate chats; API calls are stateless.
• Built-in memory feature: automatic summaries of your chats, opt-in.
• Projects: persistent context and documents scoped to one workspace.
• API memory tool: client-side files an agent reads and writes across sessions.

Option 1: Turn On Claude's Built-In Memory Feature

Anthropic shipped a memory feature for the Claude app that summarizes your conversations and carries that context forward. It launched on September 11, 2025 for Team and Enterprise plans, expanded to Pro and Max on October 23, 2025, and became available to all users, including the free tier, on March 2, 2026.

Memory is opt-in. You enable it in settings, and Claude builds a summary of details like your role, preferences, and recurring topics, then references that summary in later chats. You can view and edit what Claude has stored. Incognito chats let you run a conversation that is not saved and does not feed memory.

This is the lowest-effort path for everyday use. It covers the Claude app and web, not the API or Claude Code.

• Open Settings, find the memory control, and opt in.
• Review and edit the stored summary at any time.
• Use Incognito chat for conversations you do not want remembered.
• Applies to claude.ai and the apps, not API access.

Option 2: Use Projects for Persistent Workspace Context

Projects give Claude a self-contained workspace with its own instructions, document library, and chat history. Context you set in a Project loads at the start of every conversation inside that Project, so you stop re-explaining the same background.

A useful safety detail: Claude creates a separate memory for each Project. That keeps a client engagement isolated from an internal launch, so confidential threads do not leak into the wrong context.

Projects suit ongoing, bounded work: a specific client, a codebase, a research topic. The tradeoff is scope. Memory and instructions you set there do not follow you into unrelated chats.

• Create a Project and add background documents plus custom instructions.
• Each Project keeps its own isolated memory and history.
• Best for recurring work on one client, product, or topic.

Option 3: The API Memory Tool for Developers

If you build on Claude through the API, the model is stateless by default. The memory tool closes that gap. It lets Claude create, read, update, and delete files in a memory directory that persists between sessions, so an agent can build knowledge over time instead of holding everything in the context window.

The tool is client-side: you decide where the data lives, whether that is a file store, a database, or cloud storage. When enabled, Claude checks its memory directory before starting a task and pulls back only what is relevant. The current tool version is identified as memory_20250818 in the request.

This path needs engineering work, but it gives full control over storage, retention, and what gets written. Anthropic recommends restricting all operations to the memory directory and validating paths to block directory traversal.

• Add the memory tool to the request; implement client-side handlers.
• Claude views the memory directory before each task and reads relevant files.
• You own storage, retention, and security; lock operations to /memories.

Option 4: An External Memory Layer You Control

Claude's built-in memory and Projects live inside Anthropic's product. If you want a single recall layer that is not tied to one assistant, an external AI memory app fills that role. You save facts, notes, and context once, then bring them into whatever chat you are working in.

MemX, built by Neural Forge Technologies, is an external memory layer of this kind. It is meant for personal recall and saved context, not as a replacement for Claude as your chat assistant. The angle is simple: keep the things you want remembered in one place you control, instead of rebuilding context in every tool.

MemX is private by architecture: per-user isolation, encryption at rest, key management through Google Cloud KMS, and on-device handling. Use it when your memory needs to outlive a single assistant or span more than one tool.

• Store facts and context once; reuse them across chats.
• Positioned for personal recall, not as a Claude replacement.
• Private by architecture: per-user isolation, encryption at rest, Google Cloud KMS, on-device.

Which Method Should You Choose

Match the method to your setup. For casual app use, turn on the built-in memory feature; it is automatic and opt-in. For repeated work on a defined topic, use a Project so context loads every time and stays isolated. For software you are building, the API memory tool gives an agent durable, controllable memory. For recall that should follow you beyond any single assistant, add an external memory layer.

These options stack. You can run Projects for client work, keep the global memory feature on for personal preferences, and still hold reference facts in an external layer. The goal is the same: stop starting from zero in every conversation.

• Casual chats: built-in memory feature.
• Focused, recurring work: Projects.
• Custom apps and agents: API memory tool.
• Cross-tool personal recall: external memory layer.