What survives when Anthropic deprecates a Claude model?

Your Projects, files, and custom instructions survive. Your conversations survive (as text). What doesn't: any model-specific behavior you got used to. A Project tuned for Claude 3 Opus will feel different the day it routes through Claude 4. Anthropic keeps the container; the personality inside changes.

How do I back up a Claude Project?

There's no one-click export for a whole Project. You can: download each conversation individually as text from the share menu, save the Project's uploaded files locally, and copy the custom instructions into a text file you keep outside Anthropic. It's manual, but it's the only complete backup.

Will Anthropic notify me before deprecating a model?

Usually yes — they've given multi-month notice for past sunsets. But notification doesn't mean preservation. The old model goes away on the date they pick, and any behavior that depended on it goes with it. Your Project shell remains; the engine doesn't.

Can I move a Claude Project to ChatGPT or Gemini?

Not directly. The format is Anthropic-specific. The portable parts — your uploaded files, your instructions, the conversation text — can be re-uploaded elsewhere manually. The Project's accumulated context (what Claude has learned about how you use it) doesn't transfer.

What's the difference between a Claude Project and Claude's memory?

Projects are scoped containers — files, instructions, conversations grouped around a single workstream. Memory (when Anthropic enables it for your account) is account-wide preferences. Projects are stronger for focused work; memory is broader but shallower. Both are reset-prone in different ways.

Does Konshus work with Claude Projects?

Yes. The Konshus Claude importer reads your downloaded conversations and Project files. We extract atoms across the whole thing — decisions, recurring people, preferences, themes — and assemble a persona that's not locked to any one Anthropic model or Project. When Claude 4 ships, your persona is still there.

A guide · ~6 min read

Claude Projects: what survives when Anthropic deprecates a model

Claude Projects are the cleanest workspace any AI lab currently offers. They're also tied to a model that Anthropic gets to retire on its own schedule. When the sunset email arrives, here's what your Projects keep, what they quietly lose, and how to protect the parts that matter.

What a Project actually contains

Files you've uploaded — PDFs, code, transcripts. These persist.
Custom instructions — the prompt at the top of every conversation. Persists.
Conversation history — every chat, threaded. Persists as text.
Model-specific behavior — the rhythm, voice, and judgment of whichever Claude version you've been using. Does not persist.

The deprecation pattern

Anthropic has sunsetted Claude 1, Claude 2, Claude 2.1, and older 3-series snapshots. The pattern is consistent: a few months notice, a graceful redirect to the newer model, and then the old version is gone. The Project shell stays. The intelligence inside it gets swapped without your input.

For most workflows this is fine. For Projects you've carefully tuned — a writing voice, a coding partner, a therapy-adjacent journaling setup — the swap can feel like a colleague being replaced overnight by someone who has read all the notes but doesn't quite get it.

The backup nobody tells you to do

Download each conversation from the share / export menu. There's no bulk export — it's per-conversation. Tedious; do it anyway for the ones you care about.
Save your uploaded files locally. If they only exist inside the Project, they only exist as long as the Project does.
Copy custom instructions into a text file you keep outside Anthropic. Treat them like config — version them.
Note the model version you've been using at the top of that file. When the next sunset comes, you'll want to know what you're replacing.

Short answer

Projects, files, custom instructions, and conversation text all survive a Claude deprecation. Model-specific behavior doesn't. There's no one-click backup — download conversations individually, save files locally, copy instructions to a text file outside Anthropic. Or load the lot into a vault that holds your context independent of whichever Claude version is current.

A version of you that outlasts the model

Konshus has a Claude importer. Drop in your downloaded conversations and we distill the cross-cutting atoms — the way you think about decisions, the people in your life, the themes you keep returning to. The persona lives in your vault, model-agnostic. The next time Anthropic ships something new, you carry your context across the upgrade instead of starting over.

Frequently Asked Questions

Survive the next Claude sunset.

A vault holds the version of you that's worth keeping — independent of whichever Claude model Anthropic is currently shipping.

Meet Konshus