Compare · checked against READMEs and docs, 16 October 2026

claude-mem alternatives for Claude Code (and Windows)

claude-mem records what Claude Code does and feeds it back next session. If you want fewer moving parts, a setup that behaves on Windows, or memory that other tools can read too, these are the options.

By Ilian Raykov, founder of BrainTube

The short answer

Before replacing claude-mem, check what Claude Code already does: CLAUDE.md holds the rules you write and auto memory stores what Claude learns, and both load every session with nothing installed. If you need more, a hosted MCP memory server (Mem0, Supermemory or BrainTube) avoids the local worker, vector database and hooks that cause most Windows trouble, and Claude Code connects to it with one claude mcp add command. For local files you own, use Basic Memory or the official Knowledge Graph Memory server.

ToolRuns whereWindows notesLicenceWrites from Claude Code
Claude Code built-in (CLAUDE.md + auto memory)local filesnothing to install—Yes (auto memory)
claude-memlocal worker + hooks (optional hosted observer)documented installer; history of Windows issuesApache-2.0Yes, automatic capture
Basic Memorylocal Markdown, or Cloud $15/moPython installAGPL-3.0Yes
Knowledge Graph Memory serverlocal JSONL filelaunch through cmd /cMITYes
Mem0hosted or self-hostedhosted: nothing localApache-2.0Yes
Supermemoryhosted or self-hostedhosted: nothing localMITYes
BrainTubehostednothing local; one claude mcp addElastic-2.0 (source-available)Yes (add_note, ingest_content, bulk_ingest)

What claude-mem does, and why people look elsewhere

claude-mem captures tool-use observations from your sessions, compresses them into summaries with an AI model and injects the relevant ones into later sessions. You install it with npx claude-mem install or from the Claude Code plugin marketplace. It needs Node.js 20 or newer and pulls in Bun and uv if they are missing; search runs on SQLite plus a Chroma vector store, driven by a background worker and Claude Code hooks. Source: claude-mem README.

That design is what makes it automatic, and also what people trade away when they switch: several local processes, an optional hosted observer that asks you to sign in after install, and on Windows a run of platform-specific bug reports on its issue tracker (most recent ones closed within days). If you want memory you can read and edit yourself, or memory your other tools can use, a different shape fits better.

Claude Code memory on Windows, without an add-on

Claude Code’s built-in memory works the same on Windows. Your personal rules go in %USERPROFILE%\.claude\CLAUDE.md; project rules in CLAUDE.md or .claude\CLAUDE.md at the repository root, which you can commit so they travel with the code. Auto memory is on by default and writes to %USERPROFILE%\.claude\projects\<project>\memory\ with a MEMORY.md index. Type /memory in a session to open and edit these files.

The limit is reach: auto memory is machine-local, and none of it is visible to Claude.ai, Cursor or another machine. That is the gap the tools below fill.

1. Basic Memory: local Markdown you own

Notes are plain Markdown files that form a knowledge graph through observations and wikilinks; Claude Code reads and writes them over MCP (claude mcp add). Good if you already live in Obsidian-style files. Limits: AGPL-3.0, Python install, and syncing between machines is your job unless you pay for its Cloud plan.

2. Knowledge Graph Memory server: the minimal option

The reference memory server from the Model Context Protocol project stores entities, relations and observations in one local JSONL file. One command: npx -y @modelcontextprotocol/server-memory (on Windows, launch it through cmd /c). Limits: no semantic search and no capture; Claude saves only what it decides to save.

3. Mem0 and Supermemory: hosted memory for agents you build

Both are memory platforms with hosted and self-hosted options and MCP servers, built mainly for developers adding memory to their own apps and agents. With the hosted versions nothing runs on your Windows machine. Limits: they store short memories about users and tasks, not a library of documents, videos and podcasts.

4. BrainTube: one library for Claude Code and every other client

BrainTube is a hosted knowledge base of what you save (YouTube videos, podcasts, articles, PDFs, notes) that Claude Code can search and write to over MCP. Its write tools include add_note, ingest_content and bulk_ingest, so Claude Code can save what it finds during a session. The same library answers in Claude.ai, ChatGPT, Cursor and other MCP clients. On Windows, connect it with one command and an API key from your account:

claude mcp add --transport http braintube https://mcp.brain-tube.com/mcp --header "X-BrainTube-Token: bt_your_key_here"

Then type /mcp to confirm the connection. Limits: it is a library of what you choose to save, not automatic session capture like claude-mem; AI actions use credits, though MCP searches have a free monthly allowance; the server code is source-available, not open source. The Free plan costs $0 and includes MCP.

How to choose

  • You want every session recorded automatically: keep claude-mem, or try it again with the current installer.
  • You want rules and project facts to persist: CLAUDE.md and auto memory are enough.
  • You want files on disk you can edit: Basic Memory.
  • You want nothing running locally on Windows: a hosted MCP server.
  • You want Claude Code and your other AI tools to share what you have read and watched: BrainTube.

Deeper reading: best memory tools for Claude Code (eight tools in detail), Claude memory alternatives, give Claude long-term memory with MCP and let Claude Code save what it learns with bulk_ingest.

Frequently asked

What is the best claude-mem alternative?
Start with Claude Code's own memory: CLAUDE.md files for rules you write and auto memory for what Claude learns, both loaded every session with nothing to install. If you need more, pick by job: Basic Memory for local Markdown you own, the official Knowledge Graph Memory server for a minimal local store, Mem0 or Supermemory for hosted memory inside apps you build, BrainTube for a hosted library of docs, videos and notes that Claude Code can search and save into.
Does claude-mem work on Windows?
Yes. Its README documents Windows: npx claude-mem install puts a launcher under %LOCALAPPDATA%\claude-mem\bin and adds it to your user Path, and npx claude-mem doctor checks hooks that cannot find the executable. Its GitHub tracker has a long run of Windows-specific issues, most of them fixed; check the open ones before you install.
Where does Claude Code keep memory on Windows?
Your user CLAUDE.md is in %USERPROFILE%\.claude\CLAUDE.md, project memory is ./CLAUDE.md or ./.claude/CLAUDE.md in the repository, and auto memory is under %USERPROFILE%\.claude\projects\<project>\memory with a MEMORY.md index. Auto memory is machine-local.
Which claude-mem alternatives need nothing installed locally?
Hosted MCP servers: Mem0's and Supermemory's hosted services and BrainTube. You add a URL to Claude Code and sign in or pass a key; there is no background worker, vector database or hook on your machine.
Can Claude Code save to the memory, not just read it?
With claude-mem it is automatic: hooks capture each session. With an MCP server it depends on the server's tools. Basic Memory, Mem0, Supermemory and BrainTube all expose write tools; BrainTube's include add_note, ingest_content and bulk_ingest.
Is claude-mem free?
claude-mem is open source under Apache-2.0. After install it offers to sign you in to a hosted observer with a free trial (30 days per its README on 16 October 2026); you can skip that and use your own Anthropic plan or an OpenRouter or Gemini key.

Save once. Query anywhere.

Your YouTube library, in the AI tools you already use.

Start free

Keep reading