Two-tier secret-scan
Hard-block patterns stop known token shapes cold; heuristic warnings flag secret-like assignments for a human to confirm. Nothing reaches a published wiki unseen.
Security & trust
Your memory never leaves your machine without your say-so.
Memophant earns trust the boring way: your knowledge is plain files in your own repo, secrets are gated before they're ever published, and the license guarantees the code outlives us.
Credentials in the Keychain
Every secret — your Anthropic API key and each Vendors credential — lives solely in your iCloud Keychain, synced across your own Macs by Apple and never leaving them. Agents read or write a vendor secret only through a one-click approval prompt; the repo holds just a reference.
Local by default
Memophant is non-sandboxed so it can run git and index your repos. Your memory never leaves your machine without your say-so — the only content sent anywhere goes to Anthropic, and only when you ask. (The app checks for updates and license status separately; neither sends your memory.)
Secret-scan
Two tiers between you and a leaked token.
Every commit and publish runs the scan. Hard hits block outright; soft hits surface the exact file and line so you can confirm a false positive — publishing is permanent, so it's deliberate.
- Hard-block on known token patterns
- Heuristic warnings on secret-like assignments
- Exact file:line shown for every soft hit
- Vendor notes are scanned on save — the credential itself lives in the Keychain, never the file
Vendor credentials
Agents can use your secrets — only with your say-so.
Memophant gives agents two MCP tools: get_vendor_credential to fetch a secret, and set_vendor_credential to store one they just minted (a fresh API key, a wrangler token). Every call surfaces a native approval prompt — nothing is read or written without your click — and a verification phrase you set marks the real dialog so a look-alike can't fool you.
- The secret lives only in your iCloud Keychain — synced across your Macs by Apple. Memophant runs no servers that store your data, so it never reaches the repo, a log, or anything we own
- Per-call consent: approve once, approve for 30 seconds, or deny — auto-times-out in 60
- Headless and CI sessions can't auto-approve — by policy they time out
- An audit log records every access — who, which vendor, and why — but never the secret itself
Data ownership
There's nothing to escape from.
No lock-in is the point. If you stopped using Memophant tomorrow, every byte of your memory would still be sitting in your repo.
Plain markdown
Every note, page, and task is a file you can read in any editor. No proprietary database, nothing to export.
Built-in memory engine
Search and indexing run on Memophant's own memory engine — patterned after the best open-source memory platforms, with no external CLI to install. The files still work in any editor without it.
Git is the truth
The app stores only a thin projection for the UI. Your memory is versioned, diffable, and reviewable in normal pull requests.
The Heirloom License
A commercial license with a promise built in.
While the developer is active, Memophant is a normal paid product. But if it's ever truly abandoned, the source is released under MPL-2.0 — automatically, irrevocably, for everyone, forever. The trigger is mechanical: a single commit, release, or signed heartbeat resets the clock.
Change LicenseMPL-2.0
Dormancy window12 months
TriggerNo maintenance signal + 90 days unanswered support
Heartbeat recordgithub.com/awizemann/Memophant-Public/blob/main/heartbeat.log
VariantHL-1.0-MPL2.0-12mo
The plain-English summary is an overview, not a contract — where it differs from LICENSE.md, the license wins.
Trust that's written down, not promised.
The calm, native memory layer your security team will actually approve.