Head to head
claude-mem vs codegraph
claude-mem and codegraph solve the same job (Memory & context), so this is a pick-one decision. claude-mem is our default here: Runs with no intervention after install, which is why it beats a manual notes file. Content wrapped in <private> tags is excluded from storage.
Quick read
- Editor's pick: claude-mem.
| Field | claude-memgithub.com Not yet swept | codegraphgithub.com Not yet swept |
|---|---|---|
| What it is | Captures what the agent does via hooks, compresses events into semantic summaries, stores them in SQLite with FTS5 plus a Chroma vector DB, and injects the relevant slice into future sessions. Progressive disclosure: search returns a ~50–100 token index before you fetch detail. ~91k stars. | Parses the repo with a Rust/tree-sitter kernel into a local SQLite graph of symbols, call edges and imports, served to agents over MCP — so the agent asks one structured question instead of grepping dozens of files. 100% local. Benchmarked at 88% fewer tool calls and 62% fewer tokens. ~67k stars. |
| Category | Memory & context | Memory & context |
| Cost tier | free | free |
| Pricing | Free · Apache-2.0 | Free · MIT |
| Why builders pick it | Runs with no intervention after install, which is why it beats a manual notes file. Content wrapped in <private> tags is excluded from storage. | Pays off once the codebase is big enough that Claude spends its first ten tool calls just finding things. Auto-syncs via OS file watchers, so no manual reindexing. |
| Watch out for | not logged | not logged |
| How to wire it up | npx claude-mem install | npm i -g @colbymchenry/codegraph |
| Editor's pick | Yes | No |
Other matchups worth running
Agent endpoint: GET /api/public/compare?slugs=claude-mem,codegraph