- Honest comparison
- Claims dated July 2026
- No account needed to verify
MemexLab vs cloud second brains
Two kinds of tools share the “memex” ancestry (Vannevar Bush, 1945): cloud apps that capture and summarize your reading, and local infrastructure that gives AI agents governed memory. We build the second. This page is for deciding which you actually need — including when the answer isn’t us.
Pick by the job, not the name
Use a cloud second brain if…
…you want one-click capture of everything you browse — web pages, videos with transcripts, social posts, feeds — summarized for you, synced through their account, on their servers. WorldBrain’s Memex (memex.garden) is a mature example and capture is genuinely its superpower. If that is your job-to-be-done, use it.
Use MemexLab if…
…you want memory that stays yours and that agents can use without being
able to corrupt it: a plain-markdown vault on your disk, deterministic
retrieval your agent cites by [[slug]], writes that can only
land in inbox/ with provenance and an audit log — and any
model you like, at zero markup, or no model at all.
The honest scorecard
As of July 2026. We keep this table honest — including the rows we lose. If a claim is stale, tell us.
| Job | Cloud second brains | MemexLab |
|---|---|---|
| One-click / mobile capture breadth | ✅ Their superpower | 🚧 Inbox path on the roadmap (CLI, bookmarklet, share-sheet) |
| Video transcripts & timestamped notes | ✅ Yes | 🚧 Roadmap — timestamp-citable transcripts |
| Answers with checkable citations | Reference links | ✅ Deterministic [[slug]] citations to your actual notes |
| Where your data lives | Their cloud | ✅ Your disk, plain markdown, opens in any editor |
| Agent write governance | Agents write freely | ✅ inbox/-only writes + provenance + JSONL audit log, enforced in code |
| Model economics | Usage credits at a markup (less with your own key) | ✅ Your key or a local model, zero markup — deterministic modes need no model |
| Works with no account | Account required | ✅ uvx memexlab-mcp --vault ~/vault |
| Open source | Legacy extension is open; the live service is cloud | ✅ MIT engine, self-hostable end-to-end |
The name question, answered straight
“How is this different from WorldBrain’s Memex?”
Different product, shared ancestor. Theirs is a cloud capture/summarize
tool; ours is a local, governed memory layer for agents — markdown you own,
deterministic citations, writes only to inbox/ with a log. No
account, no cloud, MIT.
“And memex.tech?”
No relation — that’s an AI coding agent. We’re the memory harness underneath agents.
“Why not just use their MCP server?”
Theirs requires their cloud and account, and lets the agent write freely. Ours runs on your filesystem and cannot touch canonical notes by construction.