Purpose: Sarvam AI assistant lets you use the features of the Sarvam AI platform in an easy, accessible manner, directly from NVDA. It brings Sarvam's speech, language and document services to blind and low-vision users through a fully keyboard-driven, screen-reader-friendly interface.
Where to find it: Open the NVDA menu (NVDA+N) → Tools → Sarvam AI. Or press the shortcut NVDA+Alt+S at any time to open the Sarvam AI menu. Settings are also available under NVDA Preferences → Settings → Sarvam AI.
About OCR: The primary engine is Sarvam's Document Digitization API, which recognises 23 languages from a PDF or image. Images and screen/clipboard captures are automatically wrapped into a one-page PDF before being sent. A second, fully offline Windows OCR engine is also available for instant screen and object recognition.
https://dashboard.sarvam.ai).All features are under Tools → Sarvam AI: Text to Speech, Speech to Text, Translate, Transliterate, Detect language, OCR, AI Assistant / Summarise, Settings, Logs, Check for updates, Help and About.
The main shortcut NVDA+Alt+S opens the Sarvam AI menu, from which you can reach every feature with the arrow keys. All other commands (open Text to Speech, speak selection, Speech to Text, Translate, OCR, recognise screen, AI assistant, Settings) are unbound by default; assign your own keys under NVDA → Preferences → Input Gestures → Sarvam AI. You can also rebind NVDA+Alt+S there if it conflicts with another add-on.
Note: NVDA keyboard commands are single chords, so a combination such as "NVDA+S+A" (two letter keys together) is not possible; NVDA+Alt+S is used instead.
Type or paste text into the edit box, or press Upload file to load a plain text, Microsoft Word (.docx) or EPUB e-book (.epub) file. Choose the model (bulbul:v3 is newest), a conversion style (neutral, news, audiobook, entertainment, conversational, storytelling), the gender and voice, and the language (leave it on Auto detect, the default, to handle mixed-language text automatically). Adjust the Speed and Pitch sliders. Then press Convert. After conversion, press Listen to play it, or Save as MP3 / Save as WAV to export the audio. Choosing a style automatically sets suitable speed, pitch, loudness and expressiveness.
Choose the audio language (or leave it on automatic). Select a source with Choose audio file… or Start recording (your microphone), then press Transcribe to send it to Sarvam. Tick Translate to English for an English transcript. Save the result as plain text (.txt) or a Word document (.docx), or copy it.
Pick the operation, enter or load text, choose source and target languages, then press Go.
Choose a task – chat, summarise or understand a document – enter your text or question and press Send.
Choose an engine: Sarvam Document OCR (online, 23 languages, best for PDFs and photos) or Windows OCR (offline, instant, best for reading the screen). Choose a source — Choose PDF / image file, Use clipboard image, or Use current screen / object — then press Perform OCR. The recognised text is editable; copy it, save as .txt or Word (.docx), or translate / summarise it with Sarvam.
| Symptom | Cause / fix |
|---|---|
| "No Sarvam API key set" | Enter your key in Settings. |
| "Authentication failed" | The key is wrong; re-check it. |
| "No credits available" | Your account needs credits – add them in the Sarvam dashboard. |
| "Windows OCR is not available" | Install a Windows OCR language pack in Windows Settings. |
| Nothing happens on the network | Check your connection, proxy and network timeout in Settings; open Logs for details. |
Your API key is stored obfuscated in NVDA's configuration. Text, audio and images you process are sent to Sarvam for processing; OCR runs locally in Windows.
Created by Rakesh.
Sarvam AI and its API are the property of Sarvam AI. This is an independent community add-on and is not affiliated with or endorsed by Sarvam AI or NV Access. You need your own Sarvam API key (with credits) to use the online features.
Version 1.0.0. See the CHANGELOG for the list of changes.
Sarvam AI assistant is free software, released under the GNU General Public License, version 2.