AI Context Portability: Why Vendor Lock-In Persists

AI context portability — carrying your conversation history and personalization data between ChatGPT, Claude, and Gemini — is still unsolved in 2026. Every major assistant now ships its own persistent memory, but none of them reads another vendor's memory format, and none offers a one-click transfer to a competitor. The closest thing to a portability layer today is exporting your own data and converting it yourself, which is exactly what tools like open-context automate.

What "AI context portability" actually means

Portability isn't just having a copy of your data — it's having a copy in a format the next tool can actually use. A ZIP file full of raw JSON logs technically contains everything you said to an assistant, but it doesn't tell a new assistant how you like to be talked to, what you're working on, or what it already knows about you. Real portability means converting that raw history into the specific fields a destination platform reads: a preferences field, a memory field, a project's context window.

That distinction matters because every major AI vendor has now solved memory for their own product, without solving portability between products.

Every vendor solved memory. None of them solved portability.

2026 was the year cross-conversation memory went from a novelty to a default feature across the three largest assistants — and the year it became clear that each version is a retention feature for its own ecosystem, not a shared standard.

ChatGPT

OpenAI's official export process is documented in its Help Center: go to Settings → Data Controls → Export data, and OpenAI emails you a ZIP containing your full conversation history as HTML and JSON. That export is a genuine backup of what you've said — but it does not include a separate file for ChatGPT's own saved memories. To keep those, you open Settings → Personalization → Manage memories and copy the list out by hand.

Claude

Anthropic's own announcement on chat memory describes memory as now available to every Claude user, including the free tier, with Pro and Max users additionally getting project-scoped memory, full visibility into what's remembered, and an incognito mode for chats that shouldn't be saved at all. That's a genuinely user-respecting design — you can see and edit exactly what Claude keeps — but it's still a record that lives inside your Claude account, not a file you can hand to a different assistant.

Gemini

Google's memory, now branded Personal Intelligence, works the same way: it's tied to your Google account and configured through Gemini's own settings and Gems. It's built to make Gemini better at being Gemini, not to read a memory file another vendor produced.

The pattern across all three is consistent: cross-session memory, yes; cross-vendor memory, no — by design, each is a retention feature for one product, not a portable record you own.

What the law does and doesn't guarantee

It's tempting to assume data protection law already solved this. Under GDPR Article 20, EU users have a right to receive personal data they've provided to a service "in a structured, commonly used and machine-readable format," and to transmit it to another controller. That's a real, enforceable right — and it's why every major AI vendor now offers some form of data export.

What it doesn't do is require ChatGPT to understand a Claude memory file, or Claude to understand Gemini's. The regulation guarantees you get your data out; it says nothing about the destination being able to read it. The technical translation between formats — the actual portability step — is left to the user, or to a third-party tool built to do that translation.

What real portability requires

Closing that gap takes three things none of the vendor-native memory features provide on their own:

1. A neutral schema — a common representation of a conversation (role, content, timestamp) that isn't tied to any one provider's export format.
2. A conversion step — logic that maps a source provider's format into that schema, and then into a destination provider's specific fields (a preferences field, a memory field, a markdown file).
3. Storage you control — somewhere the converted context lives that isn't locked inside a single vendor's account, so switching tools again later doesn't mean starting over.

How open-context approaches this

open-context is built around exactly that three-part gap. Its pipeline parses a ChatGPT export's conversations.json, normalizes it into a common schema shared across supported providers, and generates Claude-ready preferences.md and memory.md — the same manual migration path described in our ChatGPT-to-Claude migration guide, automated instead of copy-pasted conversation by conversation.

It also ships an MCP server so Claude can save and recall context going forward, without the churn of a one-time import each time you switch tools. And because the context store is a pluggable adapter rather than a fixed JSON file, you can keep it in any of 15 backends — from a local SQLite file to Postgres — instead of trusting it to live only inside one vendor's servers. That's the same self-hosting argument we make in more detail in why self-hosting your AI context matters for privacy: portability and ownership are the same problem, viewed from two angles.

FAQ

What does "AI context portability" actually mean?

It means being able to take the conversation history, communication preferences, and remembered facts you've built up with one AI assistant and carry them into a different one, in a format the new assistant can actually use — not just a ZIP file of raw logs.

Can I export my memory from ChatGPT, Claude, or Gemini to use somewhere else?

Only partially, and only manually. ChatGPT's data export gives you a ZIP of your conversation history, but saved memories aren't included in that file — you copy them by hand from Settings. Claude and Gemini let you view and edit what they remember, but neither offers a one-click export into a competitor's format.

Does GDPR guarantee a right to move my AI chat data between providers?

GDPR Article 20 gives EU users a right to receive their personal data in a structured, commonly used, machine-readable format. It requires a vendor to hand over your data — it does not require any AI vendor to read another vendor's memory schema, so the technical translation step is still left to you or a third-party tool.

Is open-context affiliated with Anthropic, OpenAI, or Google?

No. open-context is an independent, open source project. It is not built, endorsed, or operated by Anthropic, OpenAI, or Google.

How does open-context keep my context portable?

It parses a provider's export into a vendor-neutral schema, then generates Claude-ready preferences and memory files from it. The resulting store can live in a plain JSON file or any of 15 databases you choose, so your context stays under your control instead of locked inside one vendor's account.