Give your local AI model a memory that belongs to you

You downloaded the weights. You run an open-weights model under Ollama, LM Studio, or llama.cpp because you want control — over your data, your costs, your stack. And then you notice the gap: the model is brilliant, but it has no idea who you are. Every session starts from zero.

Cloud assistants at least remember you within their walls. A local model has no memory feature at all. That’s the single biggest experience gap between self-hosted and frontier AI — and it’s the easiest one to close, because the fix is just a file.

Why a memory layer is the right shape for local models

If you self-host a model for sovereignty, putting your memory on someone else’s servers defeats the entire purpose. So the memory has to be local-first by default. That’s exactly what an AI memory layer is:

• Your persona is a signed .person.json in a git repo on your disk.
• Your corpus is a local SQLite database at ~/.dotperson/corpus.db.
• Nothing is hosted unless you explicitly turn on mirroring for a specific data kind.

The model reads the persona; the loop stays on your hardware.

What your local model gains

Once your persona projects into a local model, it gains:

1. A persistent character. The voice, conventions and refusals you’ve shaped — applied from the first token, every session.
2. Memory of your work. Your corpus can answer “what was I working on last week?” without sending anything to a cloud.
3. Continuity with your other AIs. The same persona your Claude or ChatGPT reads is the one your local model reads. One AI, present everywhere — including offline.

The demo worth doing yourself

The most convincing thing you can do is move an AI between substrates:

1. Cultivate an assistant under Claude for a week — let it learn how you write and what you refuse.
2. Point a local Llama at the same persona file.
3. Notice that it greets you the same way, holds the same refusals, and remembers your projects.

That continuity — cloud to local, with the persona intact — is the proof that the memory is genuinely yours and not a vendor’s feature. It’s also, not coincidentally, the most mission-aligned thing aiperson does.

Get started

aiperson treats local-model users as a first-class audience, not an afterthought. Read the local-first guide, check the self-hosting docs, or get started free and give the model you run a memory that belongs to you.