The Mac AI assistant that already knows.
Neo is the part of Stilou that does things. It is reached by voice, briefed by everything you have captured, and it proposes before it acts — so the useful part arrives without the part where you explain yourself first.
Neo is the AI assistant inside Stilou for macOS. You reach it by voice with a single key and ask in plain language. Before answering it reads your own brain — your calls, threads, notes and decisions — so you never start by explaining the situation. It shows where each answer came from, and proposes anything consequential before doing it.
Like a chat assistant — except you never have to start by explaining the situation.
The wholepitch, once
Every chat starts from zero
You know the ritual. Open the assistant, and before you can ask the actual question you have to reconstruct the world it lives in: who the client is, what was agreed in March, what the current draft says, why the last approach was dropped.
By the time the context is loaded, the thought that prompted you has cooled, and half the time you paste in the wrong version of the document anyway.
The model was never the bottleneck. The briefing was.
So Neo starts on the other side of that. Everything you captured this week is already connected — which means the assistant opens with the context that the other ones make you type.
Four beats, every time.
The order matters more than any one step. In particular, the proposing happens before the doing — which is what makes it safe to give an assistant real reach.
You ask, out loud.
A key, then talk. No window to find, no tab to switch to, no paragraph of background to type out first — which is the part that usually stops people bothering.
The Brain briefs it.
Before answering, Neo pulls what is relevant from your own material: the thread, the call, the promise you made, the version of the doc that actually shipped.
It proposes.
You get a draft of what it intends to do, not a confirmation that it already did it. Anything with consequences waits for you to say go.
It shows its work.
Every answer carries where it came from. A line you can click back to the Aug 3 call or the thread it was pulled from — so you can trust it, or catch it.
It proposes. You decide.
The industry is racing towards assistants that act on your behalf and tell you afterwards. We think that trade is backwards — for anything that touches a client, a calendar or a colleague, the review step is the product, not the friction.
Everything reversible.
Reading, connecting, summarising, drafting, pulling the relevant history into view, preparing what you are about to need. All of it costs you nothing if it gets it wrong — so none of it needs your permission.
Everything with a consequence.
Anything sent, published, changed or promised on your behalf gets shown to you first, in full, as a proposal. You approve it, edit it, or throw it away — and it never learns to stop asking.
“Draft the recap for Alex.” — and under the draft: from your Aug 3 call, the pricing thread, and the promise you made on Tuesday.
Provenance,on every answer
Not a destination.
The shape of an assistant matters as much as its intelligence. A window you have to go to gets visited on purpose, occasionally. Something you can reach mid-sentence gets used all day.
The key you already reach for.
Neo is voice-first because speech is the fastest way to get a complicated request out of your head. You talk the way you would to a colleague who was in the room.
Wherever the work is.
The assistant belongs at the point the question comes up — in the doc, the thread, the call you just left — not in a separate destination you have to remember to visit.
Briefed, before you speak.
Because Capture and Connect ran all week, Neo starts from the whole picture instead of from nothing. That is the entire difference.
Neo vs a general AI assistant.
ChatGPT and Claude are the real competition here, and they are excellent. The difference is not the model — it is what the model knows when you start talking.
| Neo | ChatGPT / Claude desktop | |
|---|---|---|
| Knows your context at the start | Yes — reads your own brain first | No, you paste it in each time |
| Shows where an answer came from | Yes, every answer carries provenance | Sometimes, for web sources only |
| Reached by voice from any app | Yes, one key | App or browser window |
| Acts on your behalf | Proposes first, always | Varies by tool and setting |
| Quality of the underlying model | Frontier models, same class | Frontier models |
| Works with no network | Yes, in Private mode | No |
| Knows the public internet | Less — the corpus is you | Yes, broadly |
If what you need is a model that knows the world — research, code, general questions — keep using the one you like, and keep it open. Neo is for the questions where the answer depends on your own week, which is the category those tools structurally cannot serve without you doing the briefing.
Context is the whole advantage.
Everything else is a model anyone can rentHow is this different from ChatGPT or Claude?
Those are excellent models with no idea who you are. Every session starts by you explaining the situation — and you are the one paying that tax, every time. Neo starts from your own brain: your calls, your threads, your decisions. Same class of model, radically better starting position.
Will it do things without asking?
No. Neo proposes and waits. You see what it intends to send, write or change before it happens, and anything with a consequence needs your go-ahead. An assistant that acts first is a liability dressed as a feature.
Can I check where an answer came from?
Always — that is a hard rule rather than a setting. Answers carry their provenance, so you can follow a claim back to the meeting, message or note it came out of. If it cannot show you, it should not be telling you.
Does using Neo mean my data goes to the cloud?
It depends which mode you are in, and the mode is visible. Private mode runs locally against your local index. Connected mode sends only what a request needs. Either way, your library stays on your Mac.