A GitHub guide compares ten ways to give Claude memory across sessions

Claude forgets what you told it last time. A new GitHub repository lists ten tools that fix that, side by side. It reached 125 stars in five days.

The repository is claudememory/claude-memory, under the MIT licence. It was created on 8 November. The README compares ten ways to give Claude memory that lasts between chats. It covers tools for Claude Code, Claude Desktop, and the Claude API.

What the ten options are

The ten tools the README names: Glen for shared team memory. Claude Code's built-in CLAUDE.md file and auto memory. The Claude app's own memory on claude.ai and Desktop. The claude-mem plugin that captures session summaries. Mem0 for memory in agents you build. Supermemory's hosted MCP server. Basic Memory, which stores memory as plain Markdown on disk. The official MCP Knowledge Graph Memory Server. Cognee, which builds a knowledge graph from documents. Graphiti and Zep, which build a time-aware knowledge graph.

For each tool, the README gives a short setup note. It says what the memory looks like. It says who can read it. It lists pros and cons. The format is the same for every entry. A reader can scan the ten and compare.

What the list says about trade-offs

The README calls Basic Memory "human-readable, local-first, you own the files." It calls claude-mem a single-user tool. Its summary step "runs on your Anthropic plan," which counts against the user's own tokens. It calls Glen "free for up to 3 people and 15,000 memories, no credit card."

The free Claude Code approach of a CLAUDE.md file is the shortest entry. The README calls its strengths free, no setup and plain files. It calls its limit: goes stale. That stale file is the reason the other nine tools exist at all.

Who wrote it and what to watch for

The README says plainly that "this list is maintained by the team behind Glen," the first tool on it. That is useful honesty. A reader should still read the Glen entry as a vendor's own description. It is not an independent review.

The repository quotes one Glen number. Codex with Glen merged 21% faster and used 29% fewer tokens per task. Glen ran that test on 100 merged pull requests. The source of the number is Glen, not the repository itself.

For a team picking a memory tool, the value is the shape of the list. It is not the ranking at the top. The nine other entries are described in short, factual terms. They match what each project's own documentation says.

A team deciding what to adopt this week can read the list in ten minutes. They can then remove the five options that do not match their shape: single user versus team, local versus hosted, plain text versus graph. Only then do they open any one project's own docs.

Source

claudememory/claude-memory on GitHub.