Technical support: t.me/hintsage_support
First run
On first run you get free time for the multistream and multilingual versions so you can try them all.
Fast model that recognizes and displays speech in real time and supports foreign languages.
First it records the audio, then it recognizes. For interviews and jobs where foreign terms are common. Shows recognized text with a short delay.
You can then choose which version suits you and use only that one.
Language model switching is in the options (gear icon top left). See the Settings section below for details.
Interface overview
All buttons show their current state.
Start button. Press it to begin a session and start voice recognition.
Before starting, choose or create a template: describe the interview topic and any other useful info. You can also set options for foreign-language interviews (main language, whether to translate and into which language). The template is used by the AI when generating answers and can be in any form. You can add instructions like answers must be in English. Choose whether you want short or detailed answers.
You can also set options for foreign-language interviews here: main language of the interview and whether to translate text and into which language.
The template is used by the AI when generating answers and can be in any form. You can add instructions like "answers must be in English". Also choose whether you want short or detailed answers.
Example: Junior analyst interview or Project manager interview at a foreign company. Answers in English. "Junior analyst interview" or "Project manager interview at a foreign company. Answers in English."
Example:
Start the app, create a template with the interview topic, set audio source to Microphone, press the green button, then ask a question. With the multistream version, when you see the recognized text on screen, press Get answer to see the reply. With the multilingual version, say the question and then press Recognize now and get answer. You can also find an interview video on YouTube, go to a question, start the app, set source to Speakers, press green and play the video — same flow. The app works the same during a real online interview.
Answer request mode
Manual mode. Voice is recognized automatically and shown in the chat, but an answer is requested only when you press Get answer at the bottom. You fully control when to get an answer.
Auto + manual (hands-free). The app detects when a question is asked and answers. Multistream: there can be a delay while it waits for a pause to detect end of speech; use Get answer for an immediate reply from already recognized text. Multilingual version: question detection and answer happen at the end of the recognition period, or immediately if you press Recognize.
Audio source
Speech can be taken from different sources. Speaker — recognize what is played to the speakers (not the microphone). Microphone — recognize what is said into the mic (not from speakers). For testing use the microphone; during a real interview use the speaker, because the interviewer's voice comes from the speakers.
Speaker. Main mode for interviews: voice is recognized from what is sent to the speakers.
Microphone. Use for testing when you ask questions as the interviewer.
Shots
When you press the first button (camera icon) a screenshot is taken and the task text is found. This operation is called a "Shot".
By default the whole screen is captured.
Second button (magnifier): code review — shows issues and suggestions.
You can copy the task to the clipboard (e.g. if it does not fit on screen) and press the third button (notepad) to use clipboard content instead of a screenshot.
Hotkeys can be enabled in settings.
Hotkeys:
- Alt + S — full screen shot
- Ctrl + S — select area for shot
- Alt + C — take task from clipboard
- Ctrl + Alt + P — toggle shot model
Tasks that don't fit on screen
There is no built-in simple solution. A workaround is a Chrome plugin that sends page text to the app on demand.
The installer includes a chrome_plugin folder with the plugin and setup instructions. After installing it, run the app, open a site with a task or code in an online editor. Hold Shift (or Ctrl if the text is in an online editor); a frame will appear around the shot panel. Green means connected to the plugin, yellow means not yet (wait a few seconds). The camera icon becomes T and the magnifier is mirrored so you can tell the mode. When the frame is green, use the T icon for on-screen tasks or the magnifier for code review.
The installer includes a chrome_plugin folder (in the directory where you installed the app).
It contains the plugin and setup instructions.
After installing, run the app and open any site with a task or code in an online editor.
Hold Shift (or Ctrl if the text is in an online editor) and a frame will appear around the shot panel.
Green means connected to the plugin; yellow means not yet (wait a few seconds).
The icons will change — camera to T, magnifier mirrored.
Once the frame is green, press the button you need.
Attention! May not work on some sites. The solution is not perfect but better than nothing.
We recommend disabling the plugin after the session. It tries to connect to the app every 10 seconds for requests, though this is not visible.
Program settings (⚙️)
₽Balance
Balance and subscriptions — top up the Russian-language version or buy a subscription for the multilingual version.
⚙️General
Here you find the main behavior options: get answer on hover, cursor protection, and other options affecting the app.
- Get answer on hover — click the answer button by hovering the cursor over it, so you can use the app in silent mode.
- Cursor protection mode — when on, the cursor does not change over the app window; when off, it changes to an arrow.
🎙Voice source and model
Here you set audio sources and choose the voice recognition language model.
Voice model (multistream, multilingual)
Multistream recognizes and displays speech in real time and supports many foreign languages.
Multilingual model First it records the audio, then it recognizes. It does not run in real time, so there are three modes:
1. By period. After session start the app buffers audio every 10 seconds and then recognizes it into the chat. Some overlap at period boundaries is normal. Period length can be changed in config: MULTILANG_MODEL_PERIOD_SEC, range 10–30 sec.
2. By silence. After session start the app buffers audio and recognizes when there is silence (pause). Pause length can be set in settings.
3. Manual start/stop. You decide when to start and stop recording. Use when you start before the question and stop after. Audio is transcribed and an answer is generated.
To start/stop recording use ALT-E or the green/red button. Max continuous recording is 2 minutes; recognition works best with segments up to 30 seconds.
💬Chat type
Here you choose the chat type that best suits you functionally and visually.
By default Avalon is used.
If you need more themes for code, you can choose RichText.
If you need support for formulas, UML diagrams, and system design diagrams, choose Internal Browser.
🗨️Chat
Here you can choose the chat style: two ways to display the conversation with the AI.
A continuous thread you can scroll up and down. New messages appear at the bottom and push older ones up.
After each AI reply the chat clears and the new answer is shown at the top. Use the History button to open the full chat and scroll back through earlier messages.
📸Shots (on-screen tasks)
Here you can choose the AI model for shots. For algorithm and math tasks we recommend Claude 4. You can also set the monitor for shots if you have multiple.
🔊Audio
Audio settings — if recognition does not work, select the correct speaker and microphone devices here.
⌨️Hotkeys
Here you can view the current hotkeys and change key combinations. Cursor protection and phantom window modes are described separately.
- Cursor protection mode. In this mode the cursor does not change over the window and follows the elements behind it. You can still click main window elements. When opening options or choosing a template, the cursor will switch to an arrow.
- Phantom window mode. The window is fully transparent to mouse moves and clicks. It stays on screen but all mouse actions pass through it. You control it with hotkeys.
- Virtual cursor. When enabled, a second yellow cursor appears. This cursor and the main window are invisible; you control the virtual cursor until you turn the mode off. The real cursor stays in place. Only works with cursor protection mode on.
✈️Telegram
Telegram — you can configure your Telegram account to receive answers there.
Knowledge base
The knowledge base lets you upload documents the AI should rely on when answering.
If you already have reference answers and theory, upload them here.
Answers that use the knowledge base take longer than a normal answer. You can enable a separate button to request an AI answer with the knowledge base.
How the knowledge base differs from a template
Knowledge base
The answer is retrieved based on the question — whatever was asked is what is looked up in the knowledge base.
Template
The neural network always applies the template; it does not depend on the question and may contain instructions and commands for the model.
Detailed videos and walkthroughs are in the Video section, and on the home page in the Knowledge base block.
🕒Sessions
View session history in the browser
Foreign language interviews
In the template create/edit form, expand the Foreign language interviews section to set languages and options.
In the field for languages in the audio, enter each language, then click Add, then the next language, etc.
If you need the answer in a specific language, set "Answer language" to that language.
If you need the question translated before the answer — set "Also in the answer give translation of the question into" to the desired language.
For live translation of phrases, enter the language in "Translate recognized text into". Translation modes: automatic, on button press, or both.
Example
Simple case: interview in German. Add German to languages in audio. To get answers in German so you can read them out, set answer language to German, save, and test (e.g. via YouTube or microphone).
For voice recognition, start with the Multistream model — it supports many languages and gives real-time translation.
Contact us
User group: t.me/hintsage_news
Technical support: t.me/hintsage_support