Instant, private semantic search for your markdown — in your terminal and inside your AI agents. Describe what you're looking for; MegaMem finds the note.
Be first to hear when MegaMem launches. A quick confirm email, then the launch note — no spam, ever.
By joining, you agree to get the launch announcement and product updates from We Build Stuff (MegaMem) at this address. Questions: support@mega.ec. You can withdraw consent at any time — every subscriber email carries a one-click unsubscribe.
We'll email you the moment MegaMem launches.
We emailed you a confirm link. Click it within 72 hours to lock your spot — you're not on the list until you do.
Ask for “that note about why we dropped SQLite” and keyword search finds nothing — the note never contains the word SQLite. Ask a pure vector index for an exact error string and it shrugs. MegaMem runs both, every time.
| MegaMem | Keyword search | Pure vector search | |
|---|---|---|---|
| Finds paraphrases (“why we dropped SQLite”) | ● Yes — embeddings | ● Only the words you type | ● Yes |
| Exact anchors — error strings, names, tickers | ● Yes — BM25 | ● Yes | ● Often missed |
| Both signals, fused per query | ● BM25 + embeddings, RRF | ● One signal | ● One signal |
| Your documents leave your machine? | ● Never — 100% local | ● Depends on the tool | ● Depends on the tool |
| Warm query latency | ● ~0.4 ms | ● Varies | ● Varies |
One command sets up a repo — config, MCP registration and the first index. After that, search however you work.
megamem initScaffolds the config, registers MegaMem with your AI tools over MCP, and builds the first index of your markdown.
Every repo gets its own .megamem/ index — auto-gitignored — and edits reindex in about 2 seconds, so results are never stale.
The CLI in your terminal, an HTTP service on :7823, or MCP tools — search, read, grep — inside your agent.
Docs, notes, research archives, journals, tweet corpora — if it's text on your disk, MegaMem can answer questions about it.
Your documents and indexes never leave your machine. No cloud, no account, no API keys — a single 26 MB binary does everything.
Answers land in under a millisecond once the index is warm — fast enough that searching feels like remembering.
BM25 keyword search and neural embeddings run together on every query, fused with Reciprocal Rank Fusion. Neither signal is a fallback.
Cursor, Claude Code, Codex, Windsurf and VS Code query your notes through MCP tools — search, read, grep and more. One megamem init wires it up.
Each repo keeps its own index in a .megamem/ folder (auto-gitignored), and saving a file reindexes it in about 2 seconds.
Frontmatter — dates, tags, authors — powers recency boosting and filters. Result formats like graduated and scan keep agent context lean, with confidence labels on results.
MegaMem is open source (MIT) — install it now, star the repo, kick the tires. The waitlist above is for launch announcements and what's next.
What MegaMem is, how the search works and what the waitlist is for.
MegaMem is instant, private semantic search for your markdown and text. Point it at any folder — docs, notes, research archives, journals — and ask natural-language questions. It answers in about 0.4 ms from the terminal, over HTTP, or inside your AI agents via MCP.
Yes. MegaMem is a single 26 MB binary that runs entirely on your machine. Your documents and indexes never leave it — there is no cloud, no account and no API key.
Keyword search only finds the words you type, and pure vector search can miss exact anchors like error strings and function names. MegaMem runs BM25 keyword search and neural embeddings on every query and fuses the two rankings with RRF, so paraphrases and exact strings both land.
Anything that speaks MCP — Cursor, Claude Code, Codex, Windsurf and VS Code among them. Agents get search, read and grep tools (plus status, ingest, index and kg_query), and one command — megamem init — registers everything for you.
Launch announcements and product updates by email — and a say in what ships next. Joining is free; we email a confirm link first (double opt-in), then the launch note when it lands. No spam, one-click unsubscribe in every email.
The engine is open source under the MIT license — free to use, straight from GitHub. The waitlist is about launch news and what's next, not a paywall.
Join the waitlist and hear the moment MegaMem launches.
Join the waitlist