A real AI assistant, in one file you can send.
Meet Atom, a generative assistant packed into a single SHA-256-verified AT-1 container of about 33 MB. Point it at a knowledge base, and it answers questions grounded in your documents, entirely offline, on any device.
The only AI we know of that runs entirely inside a compressed, verifiable archive.
Verified, not vibes
The model weights, the embedder and your knowledge index live in one AT-1 container, each tensor SHA-256-verified on load. It's an auditable artifact you can sign, ship and store, not a black box behind an API.
Runs offline, on-device
Generative answers are produced on the user's own machine, phone, laptop, air-gapped workstation. No cloud round-trip, no data leaving the device. The whole assistant is the file.
Knows what it doesn't know
A relevance gate means off-topic questions get an honest “I don't have that in this knowledge base” instead of a confident hallucination. It answers from what you packed in, and nothing else.
Drop in a knowledge base. Get a Living AI.
A tiny distilled reader does the talking; a verified retrieval index keeps the facts. You never train a model, you just point Atom at your content.
Point it at your docs
A folder of text, Markdown, HTML, your handbook, your docs, your notes.
Pack one verified file
AT-1 embeds the index and bundles it with the on-device reader into a single, SHA-256-verified .at1m.
Ask, anywhere
Open it on a phone, laptop or air-gapped box. Grounded answers, offline, with sources.
at1 living-ai build ./my-knowledge-base atom.at1m
at1 living-ai ask atom.at1m "what's our refund policy?"One artifact, many jobs
Website assistant
Pack your docs and marketing pages into a Living AI and let it answer visitor questions, grounded in your real content, with sources.
Onboarding on a phone
Drop your HR and onboarding handbook in, export one file, send it to a new hire's phone. A private assistant for policies and process that works on the train.
On-device customer brief
A salesperson packs an account's history and notes into a container that lives on their laptop, answers in the meeting, with nothing synced to a third party.
Try Atom, right here, in your browser
This assistant runs entirely on your device. Ask it about AT-1, compression, queries, weights, the diagnostic add-on, or pricing.
Run Atom entirely in your browser, nothing leaves your device. Downloads ~120 MB of model once, then it's cached.
What it is, plainly. Atom is a small, distilled reader paired with retrieval over a verified index, not a frontier chatbot. It answers questions that your knowledge base actually covers, cites where it found them, and declines the rest. That trade, tiny, private, verifiable and honest about its limits, is exactly what makes it shippable as a single offline file.
Turn your knowledge into a Living AI
Build one from the CLI, the desktop app, or talk to us about embedding Atom in your product.