How does dIKta.me work offline?

dIKta.me runs Whisper V3 Turbo and local LLMs (Gemma 3, Llama 3) directly on your GPU. No audio or text data ever leaves your machine. It is 100% air-gapped by default.

What operating systems does dIKta.me support?

dIKta.me is available for Windows 10+ (x64). macOS and Linux support are on the roadmap.

How much does dIKta.me cost?

Free trial with cloud credits included. Full Version: $20 one-time purchase for unlimited local dictation, all voice modes, and lifetime updates. No subscription required.

What languages does dIKta.me support for speech recognition?

Whisper V3 Turbo supports 90+ languages with automatic language detection. Bidirectional English-Spanish translation is built-in.

Do I need an NVIDIA GPU to use dIKta.me?

An NVIDIA GPU is recommended for the fastest local STT and LLM processing. However, dIKta.me also works on CPU (slower) and offers a cloud mode with wallet credits for users without a powerful GPU.

Vision (See)

The Vision feature lets you capture anything on your screen — a screenshot, a selected region, or a window — and ask an AI about it. The response can be injected at your cursor, copied to the clipboard, or sent to Quick Chat, without leaving your workflow.

TIP
Vision pairs naturally with your voice. After capturing, you can record a spoken question (e.g. *"What does this error mean?"* or *"Summarise the data in this table"*) before the AI processes the image.

Starting a Vision Capture

Press the Vision hotkey from any application:

Ctrl + Alt + S (Default)

The screen dims and a transparent snipping overlay appears with a hint bar at the bottom.

Capture Modes

Input	What happens
Click on a window	Captures that specific application window
Click and drag	Captures a custom rectangular region
Press `F`	Captures the full active monitor
Press `A`	Captures all monitors as one wide image
Press `Esc`	Cancels and returns to your work

After making a selection the overlay closes and the Vision Action panel appears.

The Vision Action Panel

This step lets you type (or record) an optional question and choose what the AI should do with the screenshot.

Optional Query

The text field accepts your question. You can type it, or click the microphone button to record a voice query (up to 30 seconds). dIKta.me transcribes your question and fills the field automatically.

If you leave the field empty, the default query is used: "Describe what you see and extract any visible text."

Local / Cloud Toggle

Switch between Local (your configured Ollama vision model, runs on-device) and Cloud (Gemini, Claude, or OpenAI) for each individual capture.

Action Buttons

Button	What it does
Save	Saves the screenshot to a file and copies it to the clipboard. No AI involved.
Clipboard	Sends the image + query to the AI and copies the response to your clipboard.
Chat	Attaches the screenshot to Quick Chat so you can have a multi-turn conversation about it.
Note	Runs the vision pipeline and records a voice note appended to your notes file.
OCR	Extracts all visible text from the screenshot exactly as it appears, and copies it to your clipboard.
Table	Extracts tabular data as tab-separated values (TSV), ready to paste into Excel or Sheets.
Color	Opens the Color Picker on the captured screenshot.
Record	Starts a video recording of the selected region.

NOTE
**Table** always uses the cloud provider regardless of your Local/Cloud toggle, because local models produce unreliable structured output for this task.

Color Picker

Ctrl + Alt + C also opens the Color Picker directly, without going through the Vision overlay.

Once the overlay is open:

Move the mouse to see a live magnifier showing the exact pixel colour under the cursor, with its hex and RGB values.
Click to pick a colour. Picked colours accumulate in a palette strip at the bottom.
Backspace to undo the last pick.
Enter to finish and copy all picked colours to the clipboard.
Tab to finish and send the palette to the AI for a colour analysis.
Esc to cancel (if you haven't picked any colours yet) or finish with the current palette.

Video Recording (Capture Moments)

Start a recording from the Vision Action panel by clicking Record, or use your configured video hotkey.

The snipping overlay appears so you can select a region or full screen. Once confirmed, a compact floating bar appears at the top of the screen showing:

A blinking red dot and elapsed timer
A Pause / Resume button
A Stop button

Recording captures your screen, microphone audio, and system audio simultaneously. An optional webcam bubble (picture-in-picture, bottom-right corner) can be enabled in Settings.

The default maximum recording duration is 120 seconds.

After Recording

When you click Stop, a post-capture panel appears:

Button	What it does
Save	Saves the MP4 file locally. No AI processing.
Describe	Uploads the clip to Gemini and returns a description of what happened.
Document	Asks Gemini to write step-by-step instructions for the actions shown.
Bug Report	Asks Gemini to produce a structured bug report based on what it sees.
Chat	Attaches the clip to Quick Chat for a multi-turn conversation.

NOTE
Video AI actions require a cloud connection. The **Save** action always works offline.

Output

By default, Vision responses are injected at your cursor, exactly like Dictation. You can change the default per-action in Settings → Vision:

Mode	Behaviour
Inject (default)	Response is typed into the active window at the cursor.
Clipboard	Response is copied to the clipboard. A toast notification confirms.
Toast Only	Response is shown in a Windows notification. Nothing is written or copied.

Local Vision Models

If you have Ollama installed, you can process images entirely on your own machine. Recommended models:

Model	Ollama command	VRAM	Best for
`minicpm-v` (default)	`ollama pull minicpm-v`	~2 GB	General use, OCR, description
`moondream`	`ollama pull moondream`	~1.2 GB	Fast descriptions on low-VRAM hardware
`llava-phi3`	`ollama pull llava-phi3`	~2.5 GB	Stronger reasoning

Configure your local vision model in Settings → Vision → Local Vision Model.