Save your AI chat's working state in one model, load it in any other
I kept hitting the same wall. Every time I switched from Claude to ChatGPT, or just started a fresh thread, I lost the working state. The decisions, the context, the stuff we'd already ruled out. Then I'd spend the first ten minutes re-explaining all of it.
Bigger context windows didn't fix it. Memory features didn't either, because they're tied to one provider and they remember you, not the work.
So I built Soil. You type "soil save" in the chat and it pulls out the working state (verified on save, secrets stripped). Then you "soil load" it in any other model, client or provider. Claude, ChatGPT, Cursor, same ground.
It plugs in as an MCP server, so it lives inside the tools you already use. There's a free tier at nativesoil.dev.
I'm building it in the open. Would love feedback, especially from people who switch models a lot. What breaks your flow when you move between AI tools?
The 'ruled out' category is the one I keep tripping over. Even with clean state transfer, a fresh model can treat ruled-out options as still-on-the-table — it has the record of the decision but seemingly no permanent instruction that it's closed. Does Soil's load schema do anything to signal that distinction to the receiving model, or is that left to the user?
The interesting question to me is whether users are trying to preserve context or preserve decisions.
Those sound similar, but they're not.
A lot of information can be transferred between models and still leave someone rethinking the same choices. The real value may end up being in preserving what was already concluded, not everything that was discussed.