CELESTOS
CELESTOS
Every thought you've had, every document you've fed it, every session you've run — organised into concept planets in a personal knowledge galaxy. The more you use Celestos, the richer it gets. Ask it what you know, fly through what you've built, let it connect what you never realised was related. All of it private. All of it yours.
Celestos groups everything you've given it by theme and renders it as a 3D galaxy you can fly through. Each planet is a topic cluster — Personal, Work, Hardware, Science, Code, Learning, and more. Click in and you see every memory, session, and document connected to that world. Zoom out and you see how they all relate to each other.
Goals, journals, voice memos, life decisions, relationships, recurring thoughts.
Projects, client notes, decisions, meeting minutes, research, strategies.
Repos, technical decisions, debugging sessions, architecture notes, snippets.
Study notes, flashcards, papers, lectures, ideas you're still figuring out.
Specs, experiments, build logs, research threads, anything technical.
Celestos creates new planets automatically as your knowledge grows into new territory.
Each concept planet has moons — your individual sessions, orbiting the topic they belong to. Zoom into a planet and its moons appear. Click a moon and the camera flies to it. Double-click and Celestos reanimates that entire conversation — dropping you straight back into where you were, on the same model, picking up exactly where you left off.
The top bar in the Graph lets you switch between nine visual modes — each one draws the same memory in a completely different way. Not themes. Different ways of understanding your own knowledge.
The default orbital view. Concept planets float in space, session moons in orbit around them.
Every session laid out chronologically. See how your thinking has evolved over time.
Connections between concepts are the focus. See what's wired to what — the map of how your knowledge is linked.
Nodes float freely, gravity-free. A fluid, ambient view of your knowledge — less structured, more atmospheric.
Concepts expand outward from a centre like a flower. Good for exploring a single topic and everything it touches.
Deep space scale. Everything visible at once — the full scope of what Celestos knows about you.
A more psychological layout — nodes weighted by how often you return to them. Your most-visited thinking, front and centre.
Uncertainty-inspired layout. Nodes probabilistically positioned — the graph as it might be rather than what it is.
A noise-grain aesthetic. The same data through a different visual filter — striking for screenshots and overview glances.
Toggle Hands-Free in the Graph, say the wake word, then speak. Celestos scores every session and concept against what you said and flies to the best match. The more specific your phrase, the faster the lock. And if two memories are equally likely, it reads them back and waits for you to say "one" or "two."
Matching is word-boundary and meaning-aware. Filler words are ignored. If two memories match equally well, Celestos reads both back: "I found 2: one, Tesla coil resonance. Two, Tesla Model 3 road trip. Say one or two." It never guesses wrong silently.
Toggle Hand Mode and a webcam picture-in-picture opens with a live hand skeleton overlay. You fly the entire galaxy with gestures. Point to aim. Pinch to select. Open palm to orbit. Fist to reset. It runs entirely offline — the hand-tracking model is bundled with the app. Your webcam feed never leaves your device.
Celestos doesn't just store your knowledge — it maps the semantic distance between every memory. A voice memo from three months ago surfaces as related to a document you added yesterday. A research session connects to a client conversation you had in a completely different project. The graph finds the edges your conscious mind skipped.
Celestos uses embeddings to find related memories even when the words don't overlap. "financial model" connects to "spreadsheet from the board meeting" connects to "pricing decision" — because they're about the same thing, not because they share words.
Create a Project for a client, a book, a startup idea, or anything ongoing. Celestos keeps a compounding wiki for each one — updated automatically as you add material. Every time you open that project, Celestos knows what's already been decided, what's still open, and what connects to what.
Each project gets a wiki that Celestos updates as you add material. It's not a summary you write — it's an evolving knowledge base Celestos builds automatically from everything you've fed into that project. The longer you use it, the better it gets.
When a session grows long, most AI tools just lose the beginning. Celestos does something different: it summarises the oldest turns into a Briefing Note and archives the raw content. The active window stays focused. You keep the full history. Nothing is deleted — it's compressed.
Oldest turns are distilled into a compact Briefing Note. The reasoning, the decisions, the key facts — preserved in one dense block.
The full raw conversation is kept in your memory graph. You can recall any part of it later — it's not gone, just out of the active window.
The active context carries the Briefing Note plus the newest turns. You never lose the thread — even in sessions that run for hours.
Your memory graph lives in plain files on your own machine under data/memory/. You can browse it, back it up, or delete any part of it — it's just files. If you run a local model, the recall and reasoning over your memories both happen on-device. Nothing leaves your computer unless you choose a cloud model.
Pick a project or a topic. Celestos synthesises everything it knows about it into a two-host dialogue, renders it with local TTS voices, and saves it as audio you can listen to anywhere. Your second brain — narrated back to you on the commute, before a meeting, or whenever you're away from a screen.