Free beta · Works with Claude and ChatGPT via open MCP · Gemini via the extension

Your memory,
on every AI.

AI Pen Drive carries your context between AIs the way a pendrive carries files between computers. And unlike every other memory product, nothing is remembered without your yes.

a brand-new ChatGPT session

Continue my Lumen research where I left off.

retrieved 3 memories + 1 artifact from your pendrive

Picking up from yesterday: the color palette is final and the App Store description is drafted. Gemini stored the release flowchart. The open question was screenshots — here is a shot list that matches the palette…

Consent

Nothing is saved without your yes

Connected AIs propose memories, and every proposal waits as pending. You approve, edit or reject in the Review Inbox — or approve a single everyday memory by telling the AI in the conversation. Anything confidential, any batch, and anything that would replace an existing memory is confirmed by you inside AI Pen Drive. Near-duplicates arrive grouped as one decision, and keyboard triage clears a busy week in seconds.

Every candidate carries its origin: which AI, which conversation, which day, with a content hash you can verify later.

the live demo vault
The Review Inbox with two pending memories awaiting approval, each showing its source AI and provenance hashThe Review Inbox with two pending memories awaiting approval, each showing its source AI and provenance hash

Receipts

Every access, on the record

Each time a connected AI reads from your vault, the Activity page records which client asked, what it asked for, and which memories it received. Consent granted from inside a chat is receipted the same way.

No other memory product shows you this. We think that should be embarrassing for them, not a feature for us.

the live demo vault
The Activity page listing access receipts: which AI client read the vault, the query it sent, and whenThe Activity page listing access receipts: which AI client read the vault, the query it sent, and when

Your vault

Readable, searchable, deletable

Every memory is plain text you can open, edit, re-scope, or restrict. Search is semantic, so “what did we decide about pricing” finds the decision, not the keyword.

Forget stops a memory being served to any AI the moment you click — every connector, instantly — with 30 days to change your mind in Trash before the encrypted residue is purged for good. What an AI already read and remembered on its own side is beyond any vault's reach, which is why "ask" and "blocked" memories are never sent at all. Export hands you the whole vault as JSON. Both are free on every plan, forever.

the live demo vault
The memory browser with per-memory consent controls, source attribution, and Forget buttonsThe memory browser with per-memory consent controls, source attribution, and Forget buttons

Cross-AI

One connector. Every AI in sync.

The vault speaks MCP, the open protocol Claude and ChatGPT already understand. Add it once per AI and your context is simply there — and the browser extension captures Gemini for review.

Your connector URL

https://ai-pen-drive-api.onrender.com/mcp

  1. 1In Claude: Settings → Connectors → Add custom connector, paste the URL.
  2. 2In ChatGPT (on plans that allow custom connectors): Settings → Connectors → Add, paste the same URL.
  3. 3Sign in once per AI. Each connection is scoped to your vault and shows up in your receipts by name.

Capture from wherever you think

Browser extension

Right-click any selection on ChatGPT, Claude, or Gemini and file it as a pending memory. A side-panel inbox lets you approve without leaving the page.

ChatGPT import

Upload a ChatGPT data export and the pipeline extracts candidate memories, de-duplicates them, and queues every one for your review. Nothing lands active.

Security

Built like it holds your memory

Because it does. The full threat model is public — the short version:

OAuth 2.1 with PKCE

Every AI connects through the standard authorization flow and gets a token scoped to your vault alone. Revoke a connector and its refresh tokens die with it.

Encrypted at rest

Memory text is encrypted before it reaches the database. The threat model, including what we can and cannot see, is published on the Trust page.

Content-blind operations

Application logs never contain memory text. The operator console shows aggregates and metadata only, and every administrative action writes an audit record.

Real erasure

Account deletion erases memories, artifacts, receipts, and events in one cascade — after a 30-day grace window you can cancel. Forget removes a memory from retrieval instantly; the encrypted residue is purged for good 30 days later, restorable until then.

Others remember automatically. We ask first.

CapabilityBuilt-in AI memoryAuto-save memory appsAI Pen Drive
Approval before anything is saved✕✕✓
Receipts for every AI access✕✕✓
Works across all AIs (live sync)✕partly✓
Carries your work, not just facts✕✕✓
Export & forget — free foreverpartlyvaries✓

Their strength is automatic. Ours is accountable. As of September 2026, Claude's memory is on by default and adds topics as you chat (Anthropic, 25 August 2026).

Simple pricing. Trust is free forever.

Free

$0during the beta

  • 100 memories with full provenance
  • Claude + ChatGPT connections
  • Review Inbox and semantic search
  • Forget, receipts, and full export
Start free
Coming later

Pro

Laterafter the beta

  • Unlimited memories and artifacts
  • Project continuity across every AI
  • Priority extraction quality
  • Priority support
Start free

Believer

Laterafter the beta

  • Everything in Pro, forever
  • No subscription, ever
  • Founding-supporter badge
  • A direct line to shape the roadmap
Start free

AI Pen Drive is free while it is in beta. Paid plans come later; Forget and vault export will never be paywalled.

Stop re-introducing yourself to your own tools

Create a vault, connect an AI, and approve your first memory in under five minutes.

Open your vault