Switch models.
Keep the arc.
Memrith is BYOK — bring-your-own-AI-key — by design. The technical reason is the same as the philosophical one: your continuity layer shouldn't belong to whichever AI company you used last year.
The model can change. Your memory doesn't.
What "model-portable" actually means
The AI you use is already a moving target. ChatGPT in 2022 doesn't sound like Claude today. Provider pricing changes. Companies get acquired. APIs deprecate. Your usage shifts as new models get better at the specific things you care about. Local models — from Ollama to llama.cpp to LM Studio — are already here and improving fast.
That much is predictable. What isn't predictable is which provider you'll be using. And the standard tools for AI memory all assume the answer is "the same one forever."
ChatGPT's memory feature is locked to OpenAI accounts. Claude Projects are locked to Anthropic. Notion AI's memory layer is locked to Notion plus whichever model they've negotiated with this quarter. Most "second brain" AI products are locked to one cloud and one provider relationship — by architecture, not by accident. Switching providers means starting over.
Memrith works differently. Your memory database lives on your machine, in a documented format, that any of the supported AI providers can read. Switching from Claude to GPT to OpenRouter — or to a local model running on your own machine — is a Settings change. Your entries, your accumulated context, your edited corrections, your relationships between memories — all of it stays exactly where it was.
Memrith stays constant. The AI provider rotates. Your memory lives on your machine, in a documented format any provider can read.
What this changes about long-term AI tools
Most AI products today are designed under one assumption: "the user will keep using us forever." So the memory, the context, the personalization all live inside the product's cloud and only work with the product's model. The user becomes the moat.
The honest position is the opposite. The AI you use should be a replaceable component. The continuity — the slowly-accumulated understanding of your work, your goals, your projects, your phrasings, your past decisions — should belong to you. When you decide a different model is better for what you're doing this year, the model swap should be a half-minute Settings change, not a year of re-explaining yourself.
BYOK is how we make this honest. You pay your AI provider directly, on actual usage, with no Memrith markup. Memrith never sees your conversations. The provider becomes infrastructure: useful, paid-for, swappable. The continuity layer is the part you actually own.
How Memrith compares to the locked-in alternatives
The trade-off, stated honestly
BYOK means you have to set up an AI provider account before Memrith is useful. That's a two-minute step — we wrote a walkthrough for each provider. It's not nothing, and we wouldn't pretend it is. The reason we accept that trade-off is the rest of this page.
The alternative — bundling AI access into Memrith — would mean we'd have to mark up the cost (so we can pay our provider), store enough metadata to bill you (so we can measure usage), and lock you to one provider relationship (the one we negotiated). All three are structural reasons we don't do it.
If your priority is "I just want it to work without setup," Memrith is not the right tool today. ChatGPT and Claude's first-party memory features are simpler at the moment you start. The difference is ownership — theirs is a memory held in their cloud, locked to their model; yours is a local file you can open, edit, and point at a different model this afternoon. Rent the intelligence. Own the memory.
What this means practically
- You pick the provider. Anthropic (Claude), OpenAI (GPT), or OpenRouter (one key, many models). All three are supported equally — we don't get a kickback.
- You pay exactly what you use. No Memrith markup. No monthly minimum. Set a hard spend cap in your provider's dashboard. Most users land at $2-15/month for typical use.
- You can change your mind. Switching providers is a Settings change. Your memory, entries, and conversations are unchanged. The next chat opens with the same context, routed through the new model.
- Bring your own local model too. Since v1.3.0, point Memrith at a server you run yourself — Ollama, llama.cpp, or LM Studio — with no API key needed for most of them. Move between cloud and local and the memory you've accumulated carries across unchanged.
Why we wrote this page
The "your memory is portable" claim is the kind of thing that's easy to bury in a feature list and easy to lose in a marketing tagline. It's also the kind of structural advantage that compounds over years — quietly, until the moment you actually want to switch.
Memrith is built around long-term continuity. That commitment isn't credible if the continuity layer disappears the moment a better AI provider shows up. So this page exists to say the claim plainly, upfront, in writing: the model can change, your memory doesn't.