tl;dr There is no open standard for cloud data backend for use in AI tool calls. This is a request for collaboration on an open reference standard for such data systems.
Figure 4.1. The whole design: one account key, one kind of token, a database and a blob store
PLAINTEXT EXISTS ONLY ON THE DESKTOP AND INSIDE AN APPROVED REQUEST
One account key. The desktop encrypts messages before upload; each reader or writer opens its own copy.
+-------------------+ +------------------------+
| Sources on the | | Readers and writers |
| desktop: mail, | | AI assistants, web app |
| chats, calendars | | 2nd devices, scripts |
+-------------------+ | each has its own token |
| +-----+-----------+------+
| reads | ^
| token + request | | results
v v |
+-------------------+ +------------------------+
| Desktop app | ciphertext | Server |
| makes, keeps key | + tags | finds token's copy |
| encrypts messages |------------>| opens key per request |
| computes tags | attachments | searches the tags |
+-------------------+ as they are | decrypts the hits |
| drops key afterwards |
+---+---------------+----+
^ ^
store, look up | | store, read
v v
+----------------+ +---------------------+
| Database | | Blob store |
| ciphertext | | attachments, |
| search tags | | encrypted under the |
| token hashes | | storage key, which |
| locked copies | | is sent with each |
| of the key | | request, never kept |
+----------------+ +---------------------+
The server never stores the account key, a token or plaintext. The blob store sees each attachment in the clear while it handles a request, and keeps only ciphertext.
I am providing a reviewable document hosted in GitHub that is open to comment, a review schedule and an announcement system to facilitate the document development. I have a lot of open questions about the right way to do this, but have provided a straw man version of the Security Architecture that is based on our own private code, and learnings applying AI to real customers.
The effort opened yesterday Nov 11, 2026 and the deadline for the first open comment period is Dec 11, 2026.
I am looking for:
- Other subject-matter experts to focus in their areas of expertise
- Other editors and those willing to help incorporate feedback
- what is the right way to do encryption and transport?
- what is the right way to achieve transparency and attestation of such a system?
Background: I am a 15+ year software developer, who like many others, is engaged in delivering powerful AI capabilities to domains besides coding. For me, industry and critical infrastructure via my brands PRCVD.ai and MatterBoard. This is a proposed design for MatterBoard, and is not yet implemented.
tl;dr There is no open standard for cloud data backend for use in AI tool calls. This is a request for collaboration on an open reference standard for such data systems.
Figure 4.1. The whole design: one account key, one kind of token, a database and a blob store
PLAINTEXT EXISTS ONLY ON THE DESKTOP AND INSIDE AN APPROVED REQUEST
One account key. The desktop encrypts messages before upload; each reader or writer opens its own copy.
The server never stores the account key, a token or plaintext. The blob store sees each attachment in the clear while it handles a request, and keeps only ciphertext.I am providing a reviewable document hosted in GitHub that is open to comment, a review schedule and an announcement system to facilitate the document development. I have a lot of open questions about the right way to do this, but have provided a straw man version of the Security Architecture that is based on our own private code, and learnings applying AI to real customers.
The effort opened yesterday Nov 11, 2026 and the deadline for the first open comment period is Dec 11, 2026.
I am looking for:
- Other subject-matter experts to focus in their areas of expertise
- Other editors and those willing to help incorporate feedback
The first thing to align on are the requirements of such a system which are clearly defined here: https://ajs-consult-llc.github.io/mirror-encryption-rfc/chap...
Big questions:
- what is the right way to do encryption and transport?
- what is the right way to achieve transparency and attestation of such a system?
Background: I am a 15+ year software developer, who like many others, is engaged in delivering powerful AI capabilities to domains besides coding. For me, industry and critical infrastructure via my brands PRCVD.ai and MatterBoard. This is a proposed design for MatterBoard, and is not yet implemented.
Here's the link to the straw man in full: https://ajs-consult-llc.github.io/mirror-encryption-rfc/