What Claude's built-in memory does, and where it stops
Claude can retain context about you and reference prior conversations, managed inside Claude. That is genuinely useful if Claude is the only assistant you use. The limit is structural rather than a missing feature: those memories belong to Claude. ChatGPT cannot read them, Cursor cannot read them, and a model you switch to next year cannot read them. Anything you want to survive a tool change has to live somewhere else.
Anthropic's memory tool is a different thing
Worth separating, because the names collide. Anthropic also ships a memory tool for developers building on the API — it lets Claude store and retrieve information across conversations in a directory of memory files. That is a building block for agents you write yourself, not the consumer memory feature inside Claude.ai, and not a product you install. If you are building an agent, it is the right primitive. If you are a Claude user who wants your own library available, it is not what you are looking for.
1. claude-mem — session memory for Claude Code
What it is: A persistent memory system built specifically for Claude Code. It captures observations from your sessions, compresses them into semantic summaries, and makes them available to future sessions, so a new session does not start cold.
Install: available through the Claude Code plugin marketplace, or npx claude-mem install.
Best for: People who live in Claude Code and keep re-explaining the same repository.
Honest limits: Scoped to Claude Code sessions — this is not a general knowledge base, and it is not what you would use to ask Claude about a YouTube video you watched last month. the source is on GitHub.
2. Mem0 — memory layer with a Claude Code path
What it is: An open-source memory framework for AI applications with a hosted platform on top, Apache-2.0 licensed and roughly 63,000 GitHub stars as of 17 August 2026 — the most-starred project here. It publishes a specific guide for wiring persistent memory into Claude Code over MCP.
Best for: Engineering teams building agents that need long-term recall, and Claude Code users comfortable with an API key.
Honest limits: It is a developer memory layer first. There is no install-and-start-saving consumer flow.
3. Supermemory — MCP server over a memory store
What it is: A memory layer aimed at developers, with a personal app and an MCP server on top. Its core is open source under the MIT licence, roughly 28,900 GitHub stars as of 17 August 2026.
Best for: Developers embedding memory in their own product who also want to use it directly.
Honest limits: Developer-first in shape; auto-capturing what you actually read and watch is not the centre of the product.
4. Zep — long-term memory for agent applications
What it is: An open-source memory server focused on long-term memory of conversations and structured facts extracted from them. Apache-2.0 licensed, roughly 4,800 GitHub stars as of 17 August 2026.
Best for: Developers building conversational products that need durable per-user memory with an open-source core.
Honest limits: Infrastructure, not a personal app. If you want to use AI memory rather than build with it, this is the wrong shape.
5. BrainTube — your own library, queryable from Claude (this is us)
What it is: A memory layer for the things you consume. It captures YouTube videos, podcasts, articles and PDFs, transcribes them, builds a knowledge graph, and exposes the whole library over a native MCP server — so Claude.ai, Claude Desktop, Claude Code, ChatGPT and Cursor all query the same corpus. Answers cite back to the exact timestamp in the source.
Price: Free $0 (30 credits/mo) · Starter $9 (300) · Pro $19 (1,000) · Pro+ $49 (3,000).
Best for: People whose knowledge arrives as long-form video, audio and articles, and who want it usable from every assistant.
Honest limits: This is knowledge-base memory, not conversation memory — it does not record what you said to Claude yesterday, and it is not a substitute for claude-mem if your problem is specifically Claude Code session continuity.
6. Claude's own memory, used deliberately
Not every problem needs a new tool. If Claude is genuinely your only assistant and what you want is continuity between chats, the built-in memory already does that, costs nothing, and needs no setup. The case for anything on this list is portability — wanting that context to outlive one vendor's product decisions.
Which one for which job
Three questions settle it. Is your problem Claude Code specifically? Then claude-mem is the closest fit, with Mem0 as the API-key alternative. Are you building an agent rather than using one? Then Anthropic's memory tool, Mem0 or Zep are the primitives. Do you want your own saved sources — videos, podcasts, articles — answerable from Claude and everything after Claude? That is what BrainTube is for. For the cross-assistant picture beyond Claude, see the wider cross-assistant roundup.
How these figures were checked
Licences, GitHub star counts and every linked page on this list were read from the vendor's own site or repository on 17 August 2026. Star counts move; treat them as an order-of-magnitude signal. Where a figure could not be verified it has been left out rather than estimated — claude-mem's star count is omitted for that reason.