Hints
How a hint appears
When the other side asks something, a short hint appears, written the way you would say it out loud, and fills in as the AI writes. A few self-hosted AI endpoints answer all at once instead. Under the hint are the one to three things they are really asking about, and code gets a box of its own with a Copy button.
- A hint still being written: its text fills in as the AI writes it
- The phrase the hint answers; click it to jump there
- The hint, written to be said out loud
- The one to three things they are really asking about
Several hints at once
A new question is answered at once, beside the hints still being written, up to three at a time. Past that, questions wait and are answered together.
Help me
Ctrl+Alt+S, or , takes a screenshot of the screen you chose and asks for help with it and the last minute of the call, in one press. The window itself is never in the picture.
- Your screenshot, in the chat with the words "Help me with what is on my screen"
- The hint that answers it
Capture
Ctrl+Alt+P, or , takes the same screenshot but only adds it to your next question, so you can collect several screens (up to 10) and send them together with a typed question.
- The screenshot you captured, sent with the question you typed
Ask by typing
Type a question in the input bar (the window tour shows it) and press Enter, or Ctrl+Alt+A to send without focusing the window.
Documents and images
Paste an image (Ctrl+V), drop files on the window, or choose them with . Documents can be .txt, .md, .pdf, .docx, or .doc where LibreOffice is installed, up to 20 MB each and up to 10 per question; they are read on your computer, and only their text goes with your question. A PDF that is only a scan has no text to read.
- What waits for your next question
- A document attached to your question
- Every queued file has on its corner (tooltip "Remove this document" or "Remove this image") that takes it off
- An image: click it to see it at full size
- closes the picture
A second language
With a language you pick, a Quick-switch language row shows in Settings, then Advanced, then Audio, then Language. Ctrl+Alt+K then switches speech recognition and the hints to it, and the next press switches back; the switch lasts until you quit. On a stream to our hosted service the switch takes effect in place, from the audio already sent on, so the phrase under way is kept and nothing is sent or charged twice. Under Automatic the row is hidden, and the shortcut changes nothing: it says the language is detected automatically, and that to switch it by hand you pick a language in Settings, then Advanced, then Audio.
Your prompts
Keep a prompt for each kind of call, with a few lines about you and your role, so the hints draw on your real experience. Switch the prompt in the header, even mid-call. On a plan with credits, every hint uses credits; see what a credit buys.
Settings, then Assistant has the Conversation section, with Language, and the Prompts section. Language is English by default; see Languages. Prompts lists your prompts, one per row, as the picture shows (the app starts with one, Default).
- The round button at the start of a row (tooltip "The model answers with the active prompt") makes that prompt the active one, the one the header's picker also switches
- The prompt's title; the app starts with one prompt, Default
- (tooltip "Edit") opens the prompt in the editor below the list
- (tooltip "Delete") deletes the prompt; the last prompt cannot be deleted
- The wide button under the rows, (tooltip "New prompt"), adds a prompt and opens it
An open prompt has three fields, saved with Save like any other setting:
- Title: its name in the header's picker; required, unique among your prompts, at most 60 characters
- About me: who you are, in a few lines. Hints bring in your experience only when a question is about you or asks for an example from your work; it is sent to the model with every question, so keep it a summary: at most 4,000 characters
- System prompt: the rules the model follows when it writes a hint; required
- Title: its name in the header's picker; unique, at most 60 characters.
- About me: who you are, in a few lines: your role and years, your stack, notable projects and what they achieved. Hints bring in your experience only when a question is about you or asks for an example from your work. It is sent to the model with every question, so keep it a summary: at most 4,000 characters. The button Fill from a document… (tooltip "Draft About me (and, if you like, the System prompt) from a CV or notes, for you to review") drafts it from a CV or notes, and you review the draft before it goes in.
- System prompt: the rules the model follows when it writes a hint. The default one asks for short answers you can say out loud, code only when someone asks for it, and silence for small talk. Change it only for a call that needs other rules.