Context

A context tells the assistant who you are and how to answer: your role, resume, technologies, language, answer format and length. Only one context can be active. It applies to every answer in a meeting: chat messages, quick actions, screenshots, voice, auto-reply and System Design mode. Mock interviews and meeting summaries don't use the context.

Without an active context, the assistant answers without personalization and in the language of the question, and the overlay shows a reminder to pick a context.

Scenario

The scenario is chosen when you create a context and can't be changed later in the app. It decides which fields the editor has and what goes into the answers.

Job search is for interviews: role, resume, technologies, level and answer format.
Exam sets up answers the way you'd give them at an exam, without self-presentation or stories from experience.
Custom replaces all fields with a single text of your own instructions.

Creating a context

For the "Job search" scenario, you can build a context from resume: upload a PDF, TXT or MD file, and AI fills in the role, technologies, resume, additional data and level, while "Answer naturalness" sets the speaking style. The result is a draft that you save with "Save". The answer language in the draft follows the interface language; change it if the interview is in another language.

You can also take the resume from hh.ru or GetMatch if that account is connected for auto-responses. hh.ru only gives the latest jobs with short descriptions, so the draft comes out thinner than from a PDF.

A new context becomes active right away. A copy is created inactive and repeats all settings of the original. A meeting uses only the saved version: until you click "Save", the assistant won't see your edits. If the context was saved in another window or on another device while you were editing it, saving asks whether to overwrite it with your edits or load the fresh version.

Fields

Field	What it affects
Role ("Subject / role" in an exam)	The depth and vocabulary of answers. The assistant answers the question that was asked, even if it's outside the role, and doesn't invent experience. The role also helps auto-reply understand which of the interviewer's lines need an answer.
Response languages	Up to three languages. The assistant always answers in the first one, whatever language the interviewer speaks. If the first one is "Auto", the assistant answers in the language of the interviewer's latest line, or, without one, in the language of the question in your message or screenshot; your own speech doesn't affect the choice. The second and third add a translation under the answer. The speech transcription language is set separately, in the recording menu.
Resume (job search)	Used for questions about experience and self-presentation. The assistant doesn't make up achievements that aren't in the resume. From a PDF, AI takes only positions, achievements, companies and dates. If you need more, paste the text manually. Text longer than 5000 characters is cut off.
Additional data (job search)	Anything that didn't fit into the resume. Instructions here are weaker than the other settings: for strict rules, use a custom format.
Use technologies (job search)	The stack the assistant picks when the question leaves the choice of technology open. If the interviewer asks about a different technology, the answer is about that one.
Expertise level (job search)	"Junior" explains in simple words and spells out jargon, "Middle" answers in a standard way, "Senior" covers trade-offs, failure modes and alternatives.
Exam format, Education level, Rigor, Curriculum / course (exam)	How to answer: coherently for an oral exam, step by step for a written one, briefly for a test, with definitions and examples for an exam ticket. Education level and rigor change terminology and depth, and the curriculum keeps answers within the course.
Context (custom)	Your text becomes the main instruction for the assistant. Of the other settings, only the response language, the meeting's vacancy and the auto-answer instructions still apply.
First person	The answer comes as ready speech you can say out loud: no headings, lists only for enumerations. "I" is used only for what's in the resume and additional data; the rest sounds like "we" or like an approach. Turns off answer format and length and turns on speaking pace and conversational words.
Speaking pace	Sentence length: long, regular up to 15 words, or short and almost telegraphic.
Conversational words	How many connectors like "in practice" or "essentially" to add, from none to a natural conversational rhythm. The assistant doesn't insert fillers like "um".
Answer format	"Terms": a definition, then properties. "Terms and examples": plus a short example in words, code only on request. "STAR" and "Problem-Action-Result" are built only from your experience. "Custom" opens a field where you describe the format in your own words. That format outranks the other formatting rules and turns off answer length.
Answer length	"Auto" adapts to the question; "Short", "Medium" and "Long" set a target size. It's a target, not a minimum: there's no filler to pad the length, and length doesn't add examples.
Recognition terms	Words speech transcription should recognize: technologies, companies, products. Enabled technologies from the context are added automatically. "Generate" fills both recognition fields from the role, stack and resume and keeps the terms you entered; the result applies after you click "Save". Works with realtime Deepgram, Soniox and Inworld and with server models except Parakeet, Fish, MAI-Transcribe and Nemotron.
What the conversation is about	A couple of sentences about the meeting that help transcription understand speech. Used by Soniox, Whisper, gpt-4o-transcribe and your own models (BYOK).
Auto-answer instructions	How to answer when an answer starts automatically on the interviewer's question, for example "first one sentence to say out loud, then code". It works like a quick action message: it can ask for code or a different format. It doesn't change which lines trigger auto-answer and doesn't apply to chat messages, quick actions or screenshots. Available in every scenario, including custom.

During a meeting

You can switch the context right in the overlay, and it applies from the next answer: if the assistant is printing at that moment, the current answer finishes with the old one. Clicking the active context turns it off. The current context's name is shown in the overlay toolbar and in the HUD.

The vacancy isn't part of the context: it's set when you create a meeting and is taken into account with any context.

"Reset context" (Ctrl/Cmd + Shift + X) is unrelated to this page: it clears the history of the current subchat, and the profile stays as it was.

Testing a context

The "Test" button opens a test chat with the current fields, even unsaved ones. Write a question as the interviewer or click "Generate question". "As auto-answer" shows the answer the way auto-answer would give it, with the auto-answer instructions. Paste a job description into "Vacancy" and the model takes it into account as in a meeting with a vacancy. The test remembers only the last 10 questions. It has no web search or translations.

Built-in restrictions

These rules apply with any context, and a custom one can't override them either.

The assistant briefly declines medical and legal questions. If the role in the context is tied to medical or legal practice, it declines any question. Software for healthcare and 1C are not affected.
Code, commands, configs and diagrams appear only when explicitly requested in the question or in a custom format.
If there isn't enough information for a useful answer, the assistant asks one clarifying question.

Tips

Keep separate contexts for different interview types. "Create copy" makes a variant quickly.
Before a meeting, check the context with "Test".
Put strict rules like "no more than three sentences" into a custom format, not into additional data.