Turn your memories into a private local
Turn your memories into a private local model, trained on-device on Apple Silicon
Vault
Train a Personal Memory Model
Fine-tune a compact AI model on your own vault memories, entirely on your Mac, then load it in LM Studio and ask your knowledge questions offline. You choose the memories, secrets are redacted before training, and the finished model is yours to keep, export, or share.
Overview
What Train a Personal Memory Model does
Fine-tune a compact AI model on your own vault memories, entirely on your Mac, then load it in LM Studio and ask your knowledge questions offline. You choose the memories, secrets are redacted before training, and the finished model is yours to keep, export, or share.
Retrieval finds the right memory and pastes it into context, but it is still fetch-then-answer and needs your vault reachable and online. Training turns the knowledge itself into a model you hold — one that runs on your laptop, works with no connection, and never sends your memories to the cloud. It is the difference between searching your knowledge and simply asking it, the way you would ask a colleague who has been in every meeting with you.
Vaultv1.8.0
Why it matters
The payoff for your AI memory
Scope training to your whole vault, chosen
Scope training to your whole vault, chosen topics, or a recent time window
Redact secrets and custom terms before any
Redact secrets and custom terms before any training example is written
Import the trained model into LM Studio
Import the trained model into LM Studio with one click
How it works
From first launch to reusable memory
- 1
Open 1AIVault and go to the relevant workspace.
- 2
Use Train a Personal Memory Model as part of your capture, classify, recall, and connect workflow.
- 3
Reuse the resulting memory across your connected AI tools.
FAQ
Common questions
Does my data leave my machine?
No. On an Apple Silicon Mac, training runs fully on-device — only a base model is downloaded once from Hugging Face. Your memories are never uploaded, and API keys, tokens, and private keys are redacted before any training example is written.
What do I need to run it?
A Mac on Apple Silicon with the uv tool runner installed. LM Studio is optional but enables one-click import when a run finishes; without it you can still export and share the model folder.
Which base model does it train?
You choose Qwen 2.5 in 0.5B, 1.5B, or 3B. 1.5B is recommended as the balance of recall quality and training speed; 0.5B is fastest for quick experiments and 3B is highest quality.
How do I use the model after training?
Import it into LM Studio with one click, then pick it from the model list under the 1aivault namespace and ask questions about your memories offline. You can also export it as a ZIP and share it with a teammate.
How many memories do I need?
At least four to start. The wizard previews how many memories are in scope and roughly how many training examples they produce before you commit, so you know what is going in.
Related
More Vault features
Your Portable AI Memory Vault
Save memories, decisions, preferences, facts, and skills once — every connected AI client recalls them through a single MCP server. Local-first by default; nothing leaves your machine unless you export it.
Learn moreStructured conversation entry records
Entry detail keeps user, assistant, system, tool, and terminal-style messages when the source provides full conversation shape.
Learn moreFree and Pro vault limits
Free mode supports 2,000 entries and 50 topics. Pro removes those caps and unlocks encrypted vault transfer.
Learn moreLocal memory, shared everywhere
Give every AI tool the same memory.
Start free, import real conversations, and reuse your memory across every AI agent you already use.