personal AI operating layer
The context layer for everything that knows you.
You OS is the private model of a person: memory, relationships, voice, values, anxieties, goals, preferences, and the patterns that make their answers different from anyone else's.
Memory
What happened, what was said, what mattered, and what should not be lost.
0pages
Persona
How the user thinks, speaks, decides, worries, learns, and wants to be answered.
0signals
Relationships
The people, promises, tensions, care, history, and emotional context around the user.
0people
What the OS should learn
Diary layerRaw thoughts, voice notes, photos, videos, meetings, and imported context.
Second brain layerEntities, links, facts, decisions, tasks, goals, and long-term recall.
Reflection layerRecurring anxieties, moods, emotional patterns, beliefs, conflicts, and growth edges.
Personal AI APIA controlled context endpoint future apps can call to personalize answers for this user.
First 100-user principle
Do not start with surveillanceBegin with intentional capture and optional imports. Ask for confirmation before treating inferred emotions as truth.
Make learning visibleAfter every capture, show what You learned: keep, edit, delete, or mark private.
Make the first wow personalWithin 24 hours, the user should ask a question and get an answer that could only come from their own memory.
Keep hardware as one surfaceThe Pebble is a beautiful input/output ritual; the defensible product is the OS underneath.
Tap to begin. Talk for as long as you like.
What I'm hearing
Recent rambles
Import a piece of your life.
Text, voice, images, and context notes become proposed memories first. You decide what the OS is allowed to keep.
This is the product behavior we want for the first 100 users: never silently rewrite a person's identity. The model proposes memory, relationship, mood, goal, and value updates; the user keeps, edits, deletes, or marks them private.
ImagesDescribe visible people, places, objects, mood, and the context the user adds.
AudioTranscribe, extract themes, then route proposed memories to review.
VideoPrototype stores metadata and extracts from user-provided context; full frame/audio analysis comes next.
TextBest current input for dense diary, philosophy, relationship, and second-brain signals.
Review what You learned.
This is the trust layer. Nothing becomes long-term memory until you accept it.
Pick a page to read.
Your brain.
A living model of what you've told me. Growing with every ramble. Trained on a GPU when you're ready.
Pages absorbed
—
Connections
—
Pending since last train
—
Last trained
never
How it works. When you train, every wiki page is turned into 3–4 Q&A pairs in your AI's voice, sent to a GPU on RunPod, fine-tuned as a small LoRA adapter, and pushed to a private Hugging Face repo. ~5 minutes, ~$0.04. The brain pulses while it learns.
Chat with the version of you that's been baked.
checking…
What do you want to ask the wiki?
your wiki is thinking