Train a Personal Memory Model — plus instant startup and large-vault speedups

Train a Personal Memory Model — fine-tune a compact local model on your memories, entirely on your machine (Apple Silicon Macs), and ask it questions about your own knowledge offline.

Pick what it learns: your whole vault, selected topics, or a recent time window — with a preview of how many memories are in scope and roughly how many training examples they produce.

Secrets stay out — API keys, tokens, and private keys are detected and redacted before a single training example is written, with a pre-scan that shows what will be hidden.

Three ways to draft the training questions — by local Ollama, by an installed AI CLI like Claude Code, or by any MCP-connected tool through the vault connection.

Watch the run live through building the dataset, training, and packaging — and cancel anytime; the run keeps going in the background if you close the dialog.

Use the result anywhere — one-click import into LM Studio, or export the model or its dataset to share and reuse.

A memory model card on the Dashboard and a full run history that keeps each run's model, dataset, and settings together.

A brand-new startup experience — the app opens instantly on an animated 1AIVault logo, then fades into a fully-loaded Dashboard with no white flash or half-drawn screens.

Large vaults stay smooth — heavy background work (statistics, activity heatmaps, memory linking, tier promotion, conversation capture) now runs off the app's hot path.

Faster Dashboard and sidebar on big vaults — counts and stats load in the background instead of competing with what you're doing.