Chat with your own notes and every call you've had.
Ask the way you'd ask a colleague — "what did we agree about the pricing page?" — and get an answer assembled from your own material, with the note, thread or call it came from sitting underneath it.
Stilou Chat answers questions using your own material — notes, dictations, screenshots, mail, Slack and meeting transcripts indexed on your Mac. You ask in plain language and get an answer with its sources attached, so you can open the call or thread it came from. It is not a general chatbot; the corpus is your own history, and in Private mode searching it never leaves the machine.
What does it mean to chat with your own notes?
Chatting with your own notes means asking a question in ordinary language and getting an answer drawn from a body of material you already own, rather than from the open internet. The technique underneath is retrieval: the relevant passages are found first, then a language model writes an answer out of those passages and nothing else. The distinction matters, because it is what separates an answer you can check from an answer that sounds right.
Most people currently do this in one of two unsatisfying ways. The first is search — Spotlight, Cmd-F, a mail client's search box — which is fast and literal, and therefore useless when you remember the meaning but not the words. You know someone raised a concern about the refund policy; you do not know whether they wrote "refund", "returns" or "money back", and search will not bridge that for you. The second is pasting into a general assistant, which understands the question perfectly and knows absolutely nothing about your work, so you spend the first three messages explaining the context and it forgets all of it tomorrow.
Stilou Chat sits in the gap. The corpus is you: everything you dictated, wrote, screenshotted, discussed on a call, or received in a connected mailbox or Slack workspace, all of it indexed on your Mac as it happened. You ask a question the way you would ask a person who had been sitting next to you all quarter, and it answers from that.
And it shows its work. Every answer carries the material it was built from — "from your Aug 3 call with Maria", "from the thread in #pricing" — so a click takes you to the original. This is the non-negotiable part: an assistant that cannot tell you where something came from should not be telling you it is true. If Chat cannot find support for an answer in your own material, it says so rather than filling the gap.
How to ask your notes a question
Four steps, and the first one is mostly a matter of using the Mac you were going to use anyway.
- Let the Brain fill upChat can only answer from what has been captured. Dictations, notes, screenshots and meeting notes go in as you make them; Slack, Gmail and Read.ai go in if you connect them, one at a time. A week of ordinary use is enough to make the answers interesting.
- Open ChatFrom the Stilou window, or from the Bar with a keystroke while you are in another app. You do not have to pick a folder, a notebook or a project first — the whole index is in scope by default.
- Ask in plain words"What did I promise Maria on our last call?" or "what was the objection to the new onboarding?" Names, half-remembered phrases and rough dates all work; you do not have to guess the exact wording that was used at the time.
- Check the source under the answerEach answer lists what it drew on, with dates and origins. Click one to open the note, the thread or the transcript itself and read the passage in context — which takes about two seconds and is the difference between an answer and a rumour.
Tip. Follow-ups work the way conversation works — ask "and what did we decide instead?" and it stays on the same thread rather than starting again. If the answer looks thin, the usual cause is that the material was never captured, not that the question was wrong.
What makes it different from search and from a general chatbot
Ask the way you'd ask a person.
No operators, no filters, no remembering whether it was in Notes or Slack. Meaning is matched rather than strings, so "the refund thing" finds the thread that said "money back" and never used the word refund.
Every answer shows what it was built from.
Each response names its material — a call and its date, a channel, a note you dictated — and links straight to it. You are always one click from checking whether the answer survives contact with the original.
Notes, calls, mail, Slack and screenshots at once.
The question does not need to know where the answer lives. A decision made verbally on a call and confirmed later in mail is one answer here, not two searches in two apps that you have to reconcile yourself.
It says when it doesn't know.
If your own material does not support an answer, you get told that instead of a confident invention. A personal assistant that guesses about your own history is worse than no assistant, because you have no way to tell the difference.
The second question knows about the first.
Ask something, then narrow it, then ask who said it. Context carries across the conversation, so you can circle in on the thing you half-remember rather than composing the perfect question in one go.
Searching your own memory never touches a server.
The index lives on your Mac and retrieval runs there. In Private mode the answer is written by a local model too, so the whole exchange happens on the machine — on a plane, with the network off, it still answers.
Stilou Chat vs Spotlight vs ChatGPT
These three are not competing for the same job, which is the most useful thing to know before choosing between them.
| Stilou Chat | Spotlight | ChatGPT | |
|---|---|---|---|
| Knows your notes, mail and calls | Yes — that is the corpus | Indexes files, matches text | No, unless you paste it in |
| Knows things about the world | Only in Connected mode | No | Yes — this is what it is for |
| Finding a file by its name | Not what it is for | Instant, and hard to beat | No |
| Understands a vague question | Yes | No — it matches strings | Yes |
| Shows where the answer came from | Yes, on every answer | The file is the answer | Only when it searched the web |
| Remembers last month's context | Yes, from your own history | n/a | Partly, and not from your work |
| Works with the network off | Yes, in Private mode | Yes | No |
| Your material used for training | Never, by us or the providers | n/a | Depends on your account settings |
Do not switch away from either of the other two. If you want a fact about the world, or code written from scratch, ChatGPT is the right tool and Chat will politely have nothing to offer. If you know the filename, Spotlight will find it faster than you can type a question. Chat earns its place on one specific kind of question — the one where you clearly remember the conversation but have no idea which app it happened in.
Where the question and the answer go
The index never leaves your Mac. The Brain that Chat searches is a set of files on your own disk, and retrieval — the part that decides which of your notes are relevant — runs locally in both modes. Searching your own memory is not a network operation.
In Private mode, the answer is written locally too, by a model on your machine, so a full question-and-answer round trip happens with the egress counter at zero. In Connected mode, the passages a specific question needs are sent to a hosted frontier model — Anthropic or OpenAI — to write the answer, and nothing beyond those passages goes with them. Your whole history is never uploaded. If you bring your own API key, that request travels from your Mac directly to the provider and never through Stilou's relay.
Nothing you ask and nothing it reads is used to train a model, by us or by the providers. Sources like Slack and Gmail are connected individually and revoked individually, and anything you delete from the Brain stops being answerable from immediately.
Frequently asked questions
Can I really chat with my own notes and documents on a Mac?
Yes — that is what this does. Anything captured in Stilou (dictations, notes, screenshots, meeting notes) plus any source you connected (Slack, Gmail, Read.ai) becomes answerable in plain language, with the origin shown on every answer. It is macOS only.
How is this different from ChatGPT?
The corpus. ChatGPT knows the world and nothing about your work; Chat knows your work and only comments on the world in Connected mode. Ask ChatGPT what a good refund policy looks like; ask Chat what your team actually decided about yours, and it will tell you which call that was on.
Does it make things up?
Answers are built from retrieved passages of your own material rather than from memory, and each one shows the sources it used so you can check it in a click. When nothing in your material supports an answer, it says so. That is a much narrower failure surface than a general chatbot, though the honest answer is that no language model is immune — which is precisely why the source is always on screen.
Is my data sent anywhere when I ask a question?
Not in Private mode — retrieval and the answer both run on your Mac and nothing is sent. In Connected mode, only the specific passages needed to answer are sent to a hosted model, never your whole history, and nothing is used for training. With your own API key the request goes from your Mac straight to the provider.
What can it search over?
Everything in your Brain: dictations, notes and ideas, screenshots and their contents, and meeting notes from Joey. Plus any connected source — Slack, Gmail and Read.ai today, with calendar coming. Each source is connected by you and can be revoked on its own.
Do I have to organise anything first?
No. There are no folders to build, tags to apply or notebooks to choose. Material is indexed as it is captured and linked to the people and projects it mentions, which is what lets a question find an answer that was filed nowhere in particular.
Can it answer about something from months ago?
Yes, as long as it was captured at the time. The index does not roll off after a period; a call from March stays answerable in September, with the date shown on the answer so you know how old the information is.
What do I need to run it?
A Mac on a current version of macOS. Private mode's local answering runs best on Apple silicon; Connected mode works on any supported Mac with a network connection. There is a free tier, and the app updates itself.