Proposal: Personal AI Context Standard (PACS) — Portable, User-Controlled AI Context
Concept Contributor: Bünyamin Toplugedik
AI personalization is becoming increasingly important, but today a user’s accumulated context is largely tied to individual AI platforms.
When users switch between AI systems, models, agents, or local AI infrastructure, they often have to reintroduce themselves and rebuild the way they prefer to work.
I believe this could be addressed through an open, vendor-neutral Personal AI Context Standard (PACS).
The idea
PACS would define a standardized format for a user’s personal AI context, allowing that context to be:
-
owned and controlled by the user
-
portable between compatible AI systems
-
versioned and auditable
-
selectively shared
-
usable across cloud and local AI
-
extended over time through user-approved AI observations
The profile could contain information such as communication preferences, working style, interests, long-term goals, projects, and other voluntarily provided personal context.
An important distinction
AI systems should be able to propose observations, but should not silently redefine the user’s identity.
For example:
“The user has repeatedly shown strong interest in networking and fiber-optic technology.”
That could become a proposed observation with a confidence level and source.
The user could then approve, reject, or edit it.
This creates a distinction between:
Confirmed information — provided or approved by the user
Proposed observations — inferred by AI and awaiting approval
Temporary context — relevant to the current task but not necessarily permanent
Privacy and security
PACS should contain personal context, but it should not be a secret store.
Passwords, API keys, private keys, recovery codes, banking credentials, and authentication secrets should never be part of the profile.
The user should also be able to control which parts of their profile an AI system can access.
Portability
The profile could be transferred through standard files, URLs, QR codes, NFC, local infrastructure, or other mechanisms.
A user could therefore maintain one personal AI context and use it with:
ChatGPT → Claude → Gemini → HomeLLM → future AI systems
without having to start from zero every time.
Why I think this matters
As AI systems become increasingly personal and persistent, the accumulated context itself becomes valuable.
That context should not necessarily be locked to a single provider.
The model may change.
The provider may change.
The user’s personal context should remain portable.
AI should learn how to work with a person without forcing that person to repeatedly reintroduce themselves.
I would be very interested in hearing whether something similar is already being considered, and whether OpenAI sees value in a standardized, user-controlled context layer for AI personalization and long-term continuity.
Concept Contributor:
Bünyamin Toplugedik