If you want a dedicated desktop experience with additional workflows, see BasiliskLLM (standalone app plus a minimal NVDA add-on). AI-Hub remains a full-featured option inside NVDA.

AI-Hub

AI-Hub is an NVDA add-on that connects your screen reader to multiple large-language-model (LLM) APIs. Use it for writing, summarizing, translation help, vision (images and screenshots), voice questions, transcription, and optional tool dialogs (TTS, OCR, and more)—without leaving NVDA.

The add-on’s package name in NVDA is still openai (for compatibility with existing installs). The display name you see in menus and settings is AI-Hub.

Features at a glance

• Chat in a dedicated main dialog with history, system prompt, and model/account selection.
• Images and documents as attachments from files; URLs to remote files with type checks aligned to the selected provider.
• Smart paste in the prompt field: paste files from the clipboard, paths from text, or a single URL (also available from the prompt’s context menu). Ctrl+V uses the same logic when the prompt has focus.
• Conversation save and history with rename, delete, and reopen.
• Ask a question from anywhere (no default key): assign a gesture in Input Gestures → AI-Hub to record, send, and hear or read the answer.
• Global describe: screenshot (NVDA+E) or navigator object region (NVDA+O) sent into a chat session.
• Tools submenu (under NVDA → AI-Hub): provider-specific utilities such as TTS, OCR, speech-to-text, Lyria audio, and Ollama model management.
• Reasoning / web search options appear only when the current model supports them (varies by provider).

This add-on does not include its own update checker. Updates are handled through NVDA’s official Add-on Store when you install from there. If you install manually from the releases page, install newer .nvda-addon builds the same way.

Supported providers

Configure one or more providers in NVDA Preferences → Settings → AI-Hub. Each provider can hold multiple named accounts (API keys, optional organization or base URL where applicable).

The add-on can pick up API keys from environment variables when set (for example OPENAI_API_KEY, DEEPSEEK_API_KEY, MISTRAL_API_KEY, OPENROUTER_API_KEY, ANTHROPIC_API_KEY, XAI_API_KEY, GEMINI_API_KEY / GOOGLE_API_KEY, and others). The settings UI remains the main place to manage accounts.

Speech-to-text (transcription) backends

For microphone / file transcription inside the main flow (not the separate OpenAI transcription tool), you can choose among whisper_cpp (local), openai (Whisper API), and mistral, under the Audio section of AI-Hub settings.

Installation

1. Open the add-on releases page.
2. Download the latest .nvda-addon package.
3. Install it from NVDA’s Tools → View/manage add-ons (or open the file from Explorer and confirm installation).

First-time configuration

1. Open NVDA → Preferences → Settings.
2. Select the AI-Hub category.
3. In API Accounts, choose Add account....
4. In the account dialog, select a provider, enter an account name, and fill the required fields (API key for most providers; base URL for Custom OpenAI and Ollama, with Ollama defaulting to local http://127.0.0.1:11434/v1 when empty).
5. Save, then optionally add more accounts, edit existing ones, or remove unused ones from the accounts list.
6. Optionally adjust Audio, Chat feedback, Advanced / temperature, and Auto-save conversation (enabled by default).

Until at least one provider account is ready, opening the main dialog will prompt you to add keys in AI-Hub settings.

Upgrading from older “Open AI” builds

If you used an older version of this add-on:

• Settings are migrated from the legacy OpenAI config section to AIHub. You should not lose preferences.
• Data files (conversations, keys store, attachments) are migrated from the folder openai under your NVDA user configuration directory to aihub.

You do not need to move files manually unless you use a custom setup.

NVDA menu: AI-Hub

Under the NVDA menu you will find AI-Hub (with the installed version in the label). Entries include:

• Documentation — opens the user guide in your browser (doc\en\readme.html).
• Main dialog… — open the chat window (NVDA+G by default).
• Conversation history… — manage saved chats.
• Tools — submenu grouping OpenAI, Mistral, Google, and Ollama utilities (see below).
• GitHub repository / BasiliskLLM — quick links.

Main dialog

Open with NVDA+G or Main dialog… from the AI-Hub menu.

What you can do

• Chat with the selected model; review Messages with keyboard navigation and context menus (e.g. j / k to move between messages when focus is in the messages area—see on-screen hints for the active field).
• Attach local images or documents and add file URLs where the provider supports them. Unsupported types for the current provider may be warned before send.
• Paste (file or text) from the prompt’s context menu, or Ctrl+V in the prompt: the add-on may attach files, insert text paths, or treat a single URL as an attachment when appropriate.
• Record audio snippets, attach audio files, and use TTS for prompt text where the model supports it.
• Escape closes the main dialog (when no blocking modal is open).
• Ctrl+R toggles microphone recording (when applicable).
• F2 renames the current saved conversation (after it exists in storage).
• Ctrl+N opens a new main dialog instance (session).

Options that depend on the model

Some controls only appear or apply for certain models:

• Reasoning (“thinking”) for models that expose it; streamed reasoning is kept separate from the visible answer when the API provides that distinction.
• Reasoning effort and related controls where the provider supports levels.
• Web search only for models that advertise web search support.

Exact availability changes as providers update their APIs; the UI reflects the currently selected model.

System prompt

The system prompt steers model behavior. A default suited to accessibility assistance is provided; you can edit it, reset it from the context menu, and optionally persist the last used prompt (configurable in settings). The system prompt consumes tokens like any other input.

Conversation history

Use Conversation history… from the AI-Hub menu, or assign a gesture under Input Gestures → AI-Hub.

You can list, open, rename, delete, and create conversations. From the main dialog, F2 and Ctrl+N help manage the current session.

Auto-save

If Auto-save conversation is enabled in settings (default), the add-on saves (or updates) the stored conversation after each completed assistant response, and may persist state when you close the dialog if there is something to save. You can also save from the Messages field context menu. If auto-save is off, use manual save when you want to persist.

Ask a question (voice)

This command has no default key. Assign one under Input Gestures → AI-Hub.

• First press: start recording.
• Second press while recording: stop and send.
• If the answer is played as audio, press again to stop playback.

Modes:

• Direct audio — if the selected model supports audio input, your recording can be sent as audio without a separate transcription step.
• Transcribe then chat — otherwise the configured transcription backend processes the recording, then the text is sent to the chat model.

If the main dialog is focused, its current model is used; otherwise the add-on chooses a suitable model among configured providers.

Tools submenu

The Tools entry under the AI-Hub menu opens provider-grouped dialogs (each may require the corresponding API account):

If no account is configured for a tool’s provider, the add-on will tell you to add one in AI-Hub settings.

Global commands

All default gestures can be changed in NVDA → Preferences → Input Gestures → AI-Hub.

Where data is stored

Working files, saved conversations index, unified accounts.json, and attachments live under your NVDA user configuration directory, in the aihub folder (after migration from openai). Temporary files use a tmp subfolder and are cleaned up when reasonable (e.g. on add-on termination or dialog close).

Required dependencies (auto-retrieved during build)

Builds use scons to populate runtime libs under:

addon/globalPlugins/AIHub/libs/

When a required lib is missing, scons downloads pinned wheels and extracts only what is needed into that folder. Current pinned dependencies are:

• markdown-it-py 3.0.0 + mdurl 0.1.2 — extracted into libs/ for chat Markdown rendering.
• Pillow 12.1.1 for:
• Python 3.11 win32 -> libs/lib_py3.11_win32/
• Python 3.13 win_amd64 -> libs/lib_py3.13/

The libs directory is intentionally git-ignored; contributors do not need to commit vendored artifacts.

Troubleshooting (short)

• “No account configured” — Add an API key for the provider you selected in AI-Hub settings.
• Provider rejects an attachment — Check file type and size; try another model or provider that supports the media you need.

For issues and contributions, use the GitHub repository link from the AI-Hub menu.