Autonomous AI, calibrated to you
Work done the way you would.
Phaneron learns how you work from your own files, then carries the overnight load the way you would: reading the material, planning the work, asking what it needs up front, and holding anything irreversible for your approval. You wake up to a clear account of what it did, and what's waiting for you.
An interactive illustration of the Phaneron Command surface. Scripted — no live data.
How it works
Not a chatbot. A system that does the work.
Phaneron runs a task end to end, the way you would — and stops exactly where it should.
It learns how you think.
Point it at your mail, calendar, and files. Phaneron reads your real material and learns how you decide — so the work comes back in your judgment and your voice. It's designed to model one person deeply and keep learning from how you work: thin at first, and sharper the more it sees of you.
It plans, and asks first.
Give it a goal and it breaks the work into a plan before touching anything. If something is genuinely ambiguous, it asks — once, up front — then runs the night uninterrupted, working against that plan.
It does the work.
It moves across your connected tools — gathering, reasoning, and checking its work the way you would — and carries each task through to a finished result. Not a suggestion. The work itself, done to your standard.
It reports back — and holds the rest.
You wake to a clear account of every step it took and every call it made. Anything that would send, pay, or post is prepared and held for your approval. Nothing irreversible happens without you.
Security
Secure with your most private files.
Privacy isn't a policy here — it's how the product is built. The trust model is structural.
Your files stay on your machine.
Phaneron is local-first. Your corpus lives encrypted on your own device, under a key only you hold — not in escrow, not for recovery. Phaneron, the company, never holds your key and structurally cannot read your files or the model built from them.
Only the minimal slice ever leaves.
When a task genuinely needs frontier reasoning, only the minimal slice it needs is sent — encrypted in transit — to a model contractually bound to keep nothing and learn nothing. Never your whole corpus, and what's shared stays under your control.
Nothing irreversible happens without you.
Reversible work — reading, drafting, preparing — runs freely overnight. Anything that would send, pay, or post is prepared and held for your approval. The agent can't fire it on its own.
Never used to train models.
Your files and your work are never used to train any model, and never shared, sold, or repurposed. The slice a task needs is sent only under a zero-retention agreement: kept nowhere, and gone the moment the work is done.
Phaneron is in active development. The full security and privacy details ship with the product.