Your AI chat history is the new lock-in
by•
Everyone talks about model lock-in, but the real switching cost is your conversation history. Months of context, decisions, and half-finished projects living inside one app.
We just shipped chat import in Osaurus (free, open source Mac app) so you can export your conversations from ChatGPT, Claude, Grok, Gemini, or Open WebUI and continue them with any model. Building it made us realize how differently each app treats your data on the way out. Some exports are clean JSON, some are a mess clearly not designed to be read by anything else.
Curious how others here think about this. Would you switch AI apps more often if your history came with you? Or does starting fresh not bother you?
52 views
Replies
Worth separating two things: switching cost from re-explaining context vs. from habit and ui muscle memory. import solves the first, but people underestimate how much the second keeps them stuck."
Osaurus
@valerie_june Which one is harder to break in your opinion?
" I switched from gpt to claude last year and just gave up trying to bring history over, felt like starting from zero and honestly it wasn't that bad".
Osaurus
@hudson_reid How long did you use your gpt?
I run a multi-model setup — different models for different jobs — so I'm moving context between them constantly, and the thing I've learned is that the transcript is the wrong unit to port.
Facts and decisions travel fine. Conversations don't. A thread that made sense inside one model's context — its system prompt, the tool outputs, the files you'd attached — is often just noise dropped into another. You're moving the log, but not the environment that made the log make sense.
Which is maybe why Hudson's "started from zero and it wasn't that bad" rings true: the valuable part had already been distilled into decisions, and the raw history wasn't carrying much. What I'd actually want to import isn't my chat history — it's the extracted decisions and open threads, stripped of the conversational scaffolding. Closer to notes than to a transcript.
Clean export is the right first step, though — you can't distill what you can't get out.