The assistant is the chat built into Samurai Scribe. Ask it about your meetings and notes, and it answers with the moments it drew on, each one a click away. It can also write notes, fix transcripts and draft emails. What it changes follows its mode: by default every change waits for you to keep or discard it (see Changes, modes and undo).

The assistant is free. Reading a meeting while it is being recorded needs Pro: see Live meeting help.

Opening the assistant

Beside a meeting. Press the round copper button at the right end of the meeting’s title bar (its tooltip reads “Ask the assistant”), or press Cmd+J on a Mac, Ctrl+J on Windows. The same keys hide it again. You can change them in Settings → Hotkeys, under Show or hide the assistant.
Full width, in the Projects tab of the sidebar, where chats are grouped into projects. Beside a meeting, the project name above the chat has Open in Projects view.
In the companion window, the small window that floats over your call: its Assistant tab. See the companion window.

Other places start a chat for you:

Ask about this notebook, in a notebook’s menu in the Files tree, opens a question with the notebook attached.
Find names…, in a transcript’s ⋮ menu when a meeting has unnamed speakers, asks the assistant who they are.
Polish transcript… and Translate transcript…, in the same menu, have the assistant correct the transcript or translate it into a language you name (see Translate a transcript).
Running a skill from the Skills page shows the run in the assistant (see Skills).

New to it? Settings → Help & Setup → Assistant & Projects walkthrough shows you around.

What it answers from

The assistant searches everything it may read in your vault, and quotes what it finds:

Meetings: transcripts, summaries and other documents, and files attached to a meeting.
Notes: Markdown notes anywhere in the vault, such as your notebooks and people notes.
Other files: PDFs, Word documents (.docx), PowerPoint decks (.pptx, speaker notes included) and plain-text files, read as text.
Your calendar, when one is connected, so “Prepare me for my next meeting with this team” works.
Its memory of you and of the project the chat is in (see Memory).

It never reads a folder set to Hidden in Settings → AI Apps → Folder Access, nor hidden folders such as .obsidian or .samurai. If your vault sits inside another notes folder, Include Parent Folder in the same place lets it read the notes around the vault too. Folder access explains the rules.

Limits you may run into:

PDFs and text files over 100 MB are not read.
Notes and text files over 2 MB are found by name, but what’s in them isn’t searched.
Only the first 200,000 characters of a file’s text are read.
A file whose contents are only in the cloud, such as a OneDrive file that’s online-only, isn’t downloaded for searching. It is read when you attach it or ask for it by name.

Name the time when it matters. “My open action items from this week” or “what came up about pricing last month” looks at the meetings from then.

Pointing it at a meeting, note or notebook

What’s on screen comes along. Beside a meeting, a chip with an eye icon above the message box names that meeting. In the Projects view, it names the note open in the viewer. The assistant reads it first and knows what “this meeting” means, and it still searches everything else. To ask about everything instead, remove the chip with its × (Leave it out).

Add more with @. Type @ in the box, or press + (“Add a meeting, notebook or note (@)”), then start typing a name:

meetings, by title;
notebooks;
notes and files, by name.

Use the arrow keys and Enter (or Tab) to pick one; it becomes a chip, read and searched first. Backspace in an empty box removes the last chip.

Images. Paste an image (Cmd+V on a Mac, Ctrl+V on Windows) or drop it onto the box, up to 20 MB each. Only images can be dropped there; add notes and meetings with @.

An image is read into text when you send the question. Hover its chip before sending: it says what will read it.
On this computer, images are read by the built-in model’s image reading: an optional download, Image reading, listed under the built-in model in Settings → AI Models once the model itself is downloaded. Without it, an Ollama or LM Studio model that can see images reads them, if it is the assistant’s model.
A hosted model is sent an image only if you allow it: Let cloud models read images? (Send images) allows it for every image, and Send this picture to be read? for one. Both are off until you say yes.
On a computer without a graphics chip the built-in model can use, an image sent while a meeting is recording isn’t read: the question goes without it, so reading can’t make the live transcript skip.
Images stay with the chat on this computer, not in your vault. One reaches the vault only if you ask the assistant to put it in a note.

Citations

Answers cite their sources inline, as small chips: a time such as 12:13 for a moment in a transcript, a document’s name such as Summary, or a note’s name. Hover a chip to see the passage it quotes. Click it to open the meeting at that moment, the document, or the note.

Under your question, the steps the assistant took, such as searching meetings and reading notes, fold into one line. Click it to see each step.

What else it can do

Ask in plain words. The assistant can:

write a new note, or edit, move or delete one;
make a notebook, or add a meeting to one;
fix a transcript: correct words, move a line to another speaker, name a speaker, split or merge lines, or remove lines you ask it to;
translate a transcript into another language, as one change, with every line keeping its time and speaker;
rename a person in every meeting they speak in;
draft an email;
remember something about you or the project;
run a skill, or draft a new one for you;
work in a connection such as Linear or Notion (beta).

It changes Markdown notes only, and only in folders set to Can edit. Every change follows the mode under the message box: Ask, Review (the default), Accept edits or Auto. Changes, modes and undo explains each, and how to undo.

Saving an answer

Hover an answer for its buttons (they stay faintly visible under the latest one): Copy copies it, and Save as note saves it as a note of your own. You choose the name; it starts as your question. New notes go to your Notes folder (Notebooks/Unfiled unless you moved it), or to the folder a chat is about, when the button reads Save to that folder. The note is written at once, without a review, since you asked for it. Its citations become footnotes that link to what they cite.

Email drafts

Ask for a follow-up or a recap and the assistant writes it in a card under its answer, marked Email draft · not sent. You can edit To, Subject and the text. Open in Mail opens it in your mail app, Open in Gmail in Gmail, and Copy copies it. Nothing is sent from Samurai Scribe: you send it from your own mail.

Skills in a chat

Type / at the start of a message to see your skills, each with its description. A skill that takes an argument shows what it takes after its name, such as /translate-transcript [language]. Pick one, type the argument or anything else it needs, and send: the skill runs in this chat. The assistant also uses a skill by itself when your request matches that skill’s description. See Skills.

Connections in a chat

Once a service is connected, a puzzle-piece button appears beside + (“Use a connection in this chat”). Tick the connections this chat may use; a connection marked (every chat) is offered in all of them. When the assistant wants to act in a service and that action is set to Ask, a card shows what it will send, with Approve and Deny. Connections are in beta.

Memory

The assistant keeps two memories, both plain text you can read and change:

About You is read in every chat. Open it in Settings → AI Apps → Memory → About You → Open.
Each project’s memory is read in that project’s chats. Open it from the project’s ⋮ menu in the Projects tab: Memory.

The assistant adds to them when you tell it something worth knowing next time, such as your role, the people you work with, or how you like things written. Each addition is a change like any other, so it follows the mode. In the memory window you can change or delete anything, then Save. If the assistant remembered something while you had it open, The memory changed offers Reload or Keep mine.

A chat starts with the first 2,000 characters of each memory. The rest is kept, and the assistant reads it when it needs to, so put what matters most at the top.
A skill running by itself never changes a memory without asking: its additions wait for you.
Both memories live in your vault’s hidden .samurai/ folder, so they reach your other computers if you sync hidden folders too (see Syncing and backing up).

Which model answers

The assistant uses the built-in model on your computer unless you choose another. It is the same model that writes summaries, so download it first (see Your first meeting).

The model’s name is under the message box, with a computer icon for a model on this computer or a cloud icon for a hosted one. Click it and choose Open Model Manager to change it. That opens Settings → AI Models at the Summarization card’s Assistant row. Pick the built-in model, Ollama, LM Studio, a hosted model from Anthropic, OpenAI, Google Gemini or OpenRouter, or a server at your own address. The link button beside it (“Use the meetings model for the Assistant, and change with it”) makes the assistant follow whichever model writes your summaries; the line under the picker says which it does: A model of its own or Same as meetings model.

A warning icon beside the name means “A small model: a larger one is recommended for the assistant”. A model that writes good summaries can still be too small for the assistant, which often takes several steps to answer. The first question after the app starts may show “Loading the model…” for a moment.

What leaves your computer

With a model on your computer (the built-in model, Ollama or LM Studio on this machine), nothing you ask, and nothing it reads, leaves your computer.
With a hosted model, your question, the chat so far, and what the assistant reads to answer it are sent to that provider. That includes passages from your meetings and notes, its memory, and the text read from your images. If you let cloud models read images, the images themselves go to that provider too.
A connection sends the service only what an action needs.

The network disclosure lists every request the app can make.

Your chats

Chats are kept on this computer, separately for each vault, and they don’t sync. Projects reach your other computers with the vault; their chats stay where they were held.

Beside a meeting, Conversations in the assistant’s header lists your past chats: those about the meeting on screen first, under About this meeting, then Everything else. New chat starts afresh.
In the Projects view, the sidebar lists the chats in each project and starts new ones.
The chat’s title menu has Rename…, Move to project… and Delete. Deleting a chat also removes images that only it used.
While the assistant is working, Stop ends its turn.
To take a chat back to an earlier question and undo what came after, hover the question and choose Rewind to here. See Rewind a chat.