Engineers

One memory across every tool. Readable, auditable, exportable.

Penny is a memory layer exposed over MCP: one endpoint, any MCP-capable client, no per-app glue. You can read the audit trail instead of the marketing.

What we won't tell you

It is not end-to-end encrypted. AI tools need plaintext to use memory, so E2EE isn't compatible with a memory layer that works. We lock down storage and transport instead, hold the keys, and say so. We don't claim we can't read your memory; we commit that we won't.

It is not open source. The engine is proprietary. What you can check instead is the access record for your own data, plus full read, edit, delete, and one-click export as JSON, Markdown, or CSV.

Embeddings are a disclosed limitation. Vectors are stored readable because search requires it, and published research shows embeddings can be partially inverted. We disclose it rather than imply otherwise. Full privacy and security detail.

The system, plainly

Transport

Standard MCP. OAuth authorization-code flow with PKCE; tokens scoped per account; revoke any client from the dashboard, effective immediately.

Storage

AES-256-GCM per record under a per-user key, envelope-wrapped by keys in a separate AWS KMS account from the database. A leaked database is ciphertext.

Access accountability

Every decryption by the service writes an audit record before the key is used, fail-closed. Out-of-the-ordinary access (support, incident, legal, compliance, key rotation) is published to an append-only, tamper-evident Merkle log on Cloudflare as one entry per event. Routine access (your own reads and exports, and our automated processing) is published as a daily per-user commitment: a Merkle root over the day's records, with a pseudonymous id, the day and the count. We operate the log's integrity checks, including a job that samples recent entries every ten minutes. Two kinds of key use sit outside this record: per-user key unwraps are logged in AWS CloudTrail, and restores from offline backup archives happen outside the running service.

Retrieval

Hybrid semantic + lexical search fused per query, a typed memory graph over your tags and note links, and a transparent scoring formula: relevance, graph proximity, recency, signaled importance.

Hygiene

Secrets are stripped before storage. Raw transcripts are discarded after distillation. Deletion is real, with a 30-day trash.

Coding subconscious

The Claude Code plugin injects your preferences and the current project's accumulated memory at session start. Cursor and Codex plugins are available today.

Questions engineers ask

Why not build it yourself?

You could. Most of us have, at least twice. The product parts are the consolidation pipeline, tamper-evident logging, per-user key isolation, and maintained plugins across clients. $99/year; first 250 accounts are $59/year, locked for life.

Which clients work today?

Any tool that speaks the Model Context Protocol. Today that includes Claude on desktop and web, ChatGPT, Perplexity, Cursor, Codex, and Claude Code.

Can I leave?

Yes. From your dashboard you can read, edit, and delete your memories and notes, and export everything anytime: a complete JSON file you can move to another tool, your notes as Markdown files, and your trackers as CSV. Your memory is yours; we just hold it.

Pricing

One price. Everything included.

Unlimited memories and notes, within normal use. Every MCP-capable AI tool. Per-user encryption with a public access log. Search by meaning and by keyword. Edit, delete, and export anytime. Email support. Connecting Claude or ChatGPT requires a paid plan with them.

Feature
Most popular
Founding member
$59/ year
Standard
$99/ year
Everything included
Included
Included
MCP-capable AI tools
All supported tools
All supported tools
Public access log
Included
Included
Export
Anytime, from your dashboard
Anytime, from your dashboard
Free trial
14 days, no card
14 days, no card
Rate lock
Locked for life
Standard annual rate

Ready when you are.

Connect me to your first AI in under a minute. Private, portable, protected.