Portable AI memory, in one sentence
Portable AI memory means the record of what you have taught an AI belongs to you, not to whichever chat window you happened to be using. thinqOS keeps that record, your Mind, as a service your identity owns. Any tool that connects to it, whether that is Claude Code today or a chat client tomorrow, is reading and writing the same underlying memory rather than keeping its own separate copy.
What is live today
Claude Code and Codex connect through the thinqOS CLI, which installs capture and recall hooks: a fact you confirm in one gets recalled in the other, because both reach the same Mind over MCP. That link is real and runnable today, not a roadmap item; see the setup guide for the exact commands.
Reaching it from ChatGPT and Claude.ai chat
thinqOS connects to any MCP-compatible client, and both ChatGPT and Claude.ai support custom MCP connections. What we have not yet measured is how often a plain chat session, with no installed hooks, actually chooses to call thinqOS on its own; a model has to decide to reach for the tool, and nobody has published that number. Say connects to, not works automatically, until that is measured. If you want memory that recalls and captures without you prompting it, the coding-tool hook path is the one with hooks doing that work today.
What travels, and what does not
What is portable is the context: beliefs, preferences, lessons and their sources. What is not portable is the model itself, its weights or its provider-specific behavior. Switching from one AI tool to another still means the new tool has to reason its own way through a task; portability means it does not have to start that reasoning from zero background every time.
Take it with you, or check it before you commit
Mind exports support a signed archive and open representations, including JSON-LD, YAML-LD and a Markdown vault, so the portability claim is checkable rather than asserted. Verify which fields and layers your use actually needs before depending on any of them, and remember that portable data is not the same as a portable running system: tool connections and application behavior still need their own migration plan.
Why keep memory separate from any one vendor
Every AI tool wants to be the place your context lives, because whoever holds that context has the easier path to being asked next time. That is a reasonable business incentive for them and a real cost to you: rebuilding your preferences and decisions from scratch every time a better model or a cheaper tool shows up. Keeping the Mind separate, owned by your identity rather than by any one provider, is what makes trying a new tool a small decision instead of a migration project.
This does not make thinqOS neutral in the way a standards body is neutral; it is a vendor too, and evaluating it the same way you would evaluate the tools it connects to is the right instinct. What it does mean is that the record itself is exportable and inspectable, so that evaluation does not depend on trusting a claim you cannot check.