Halo learns what you teach it and keeps it. It wakes up knowing your day, drafts replies that already sound like you, and puts agents to work on the jobs you'd rather not do. Nothing is sent, booked or scheduled without your OK.
Every screenshot below is taken from Halo itself, with personal details replaced.
Open Halo and your brief is already written: what you're focused on, what's next and who's waiting, and what it learned since yesterday. Meeting prep is ready before you ask, with chat one line below.
Built around your calendar, your location and the time of day: what's free, and what's nearby worth doing.
"Organise my trip to Milan next week." Halo puts a crew of agents on it, using your calendar and what it knows about you, and reports back with the outcome, ready for your approval.
What's on and what's worth doing around you, on a list or a map. Plan a whole day in one sentence, like "what can I do in London 18–20 Sept?", and fit events straight into your week.
Most assistants start from zero every conversation, or paste old notes back into the prompt. Halo is conditioned on what it has actually learned about you. The gap between the two answers is the product.
Add documents, spreadsheets, PDFs, code or a folder. Halo reads each one passage by passage and replays earlier knowledge so it never forgets.
Tell it something new and it's written into the model, not a sticky note. You can watch how much of the new fact it picked up and how much it kept.
A shared home for one piece of work. Inside a project, Halo answers only from what the project has been told, never from anyone's private knowledge.
Halo runs on Voltaic, Verkko's spiking neural network built for continual learning. Ordinary networks overwrite old knowledge when they learn something new. Voltaic was designed from our lab's research so that it doesn't.
It learns at inference time, from your conversations and documents, without a retraining cycle.
Connections that matter become harder to overwrite. Teach a sequence of facts and the earlier ones stay recallable.
Event-driven neurons fire only when there's something to say, which keeps learning efficient enough to run per person.
Open a card to see the trigger, a worked example and every step.
Sign in with Google or Microsoft. Connect your calendar and it's useful by tomorrow morning.