Make Claude smarter
about you.

Your second brain, Claude's long-term memory.

memex gives Claude persistent memory across conversations — saving insights automatically and surfacing them semantically when you need them. All data stays on your machine. No cloud, no API keys.

npm install -g @evan-moon/memexView on GitHub →
Claude
What did we decide about the auth approach for the new API? I can't remember.
memex · search_notes · “auth approach API decision” ›
Search results3 notes found
Auth Architecture — JWT Decision
Apr 14
#auth#backend#architecture
Chose JWT + refresh tokens over session-based auth. Rationale: stateless design fits horizontal scaling plan. Refresh token rotation every 7d, stored in httpOnly cookie...
1-on-1 Tom — API Security Review
Mar 28
#meeting#auth#people
Tom flagged: keep auth module decoupled from payment logic. Agreed to treat them as separate bounded contexts with no shared DB tables...
Security Audit Findings — Q1
Mar 10
#security#auth
No critical findings. Recommend adding rate limiting on /auth/token endpoint. Token expiry of 15min flagged as best practice minimum...
C

Based on your notes from April 14th: you chose JWT + refresh tokens over session-based auth. The reasoning was that stateless tokens fit your horizontal scaling plan. Tom also emphasized keeping auth decoupled from payment logic — separate bounded contexts, no shared tables.

Good. Also save this: we've decided to drop the legacy PHP auth module entirely by Q3.
memex · save_note ›
✓
Legacy PHP Auth — Deprecation Decision
saved · May 3
#auth#decision#roadmap
C

Saved to your second brain. I'll surface this whenever you're working on the auth migration or discussing Q3 roadmap items.

Features

~/
Local-first, always private
All notes are stored as Markdown files in your vault and indexed in a local SQLite database at ~/.memex/memex.db. Nothing ever leaves your machine. Obsidian-compatible out of the box.
∿
Semantic search across languages
Finds notes by meaning, not just keywords. Powered by multilingual embeddings running fully offline — no API key needed. Search in English, Korean, or mix both freely.
⬡
MCP server for Claude
Works as an MCP server for Claude Desktop and Claude Code. Claude searches your notes before answering and saves insights proactively at the end of conversations — no extra setup needed.
$_
Standalone CLI
Add, search, tag, and browse notes from the terminal. Filter by date range with --from / --to. Index external directories like existing Obsidian vaults.
⇆
Backlinks
Link notes with [[Title]] syntax. get_note automatically surfaces which other notes reference it — building a graph of connected thinking without any manual wiring.
∑
Daily digest
memex digest prints a summary of everything saved in the last N days, grouped by folder. A quick way to review what Claude captured this week.

Get started in 2 minutes

1
Install
Install the CLI globally. On first run, the embedding model (~450MB) is downloaded once to ~/.memex/models/.
npm install -g @evan-moon/memex
2
Connect to Claude
Register memex as an MCP server. Works with both Claude Code and Claude Desktop.
memex mcp install
3
Claude remembers
Claude now searches your notes before answering questions and saves important context at the end of conversations — automatically, without being asked.

CLI

zsh
$memex search "react performance tips"
Found 3 notes
React Server Components — notesApr 29#react #perf #nextjs
Memoization: when NOT to use itApr 15#react #hooks
Core Web Vitals deep diveMar 22#perf #web

Commands

memex addAdd a note interactively
memex search <query>Semantic search — supports --from, --to, --tag
memex digestSummary of notes saved in the last N days
memex listBrowse recent notes
memex show <id>View full note content
memex edit <id>Edit a note in your editor
memex related <id>Find semantically related notes
memex tagsList all tags with note counts
memex source add <path>Index an external directory (e.g. Obsidian vault)
memex mcp installRegister with Claude Code

Privacy

memex is designed for developers who don't want their notes in someone else's database. All data is stored locally — notes as .md files in your vault, embeddings in a local SQLite database. The embedding model runs entirely offline using @huggingface/transformers.

No telemetry, no accounts, no API keys required. Works without an internet connection once the model is downloaded.

View source on GitHub↗