DOCSEARCHbyAlgolia
Submit Docs

Page Assist Documentation

repository·main·Indexed 27 days ago

https://github.com/n4ze3m/page-assist

An open-source browser extension that provides a sidebar and Web UI to interact with local AI models (such as Ollama) or OpenAI-compatible endpoints from any webpage. It includes support for Model Context Protocol (MCP) tools via the Page Action extension to perform browser actions like clicking, typing, and navigating. Compatible with Chrome, Firefox, Edge, and other Chromium-based browsers.

Tokens
18.4K
Snippets
35
Records
139
Agent score
91%

What's inside Page Assist

  1. Understand Page Assist Data Privacy and Storage

    main
    Page Assist is designed for local-first privacy. It does not collect, store, or transmit user data to external servers. All operations, chat histories, and settings are stored locally within your browser. No third-party analytics or tracking services are integrated into the application.

  2. Understand Page Share Feature Data Handling

    main

    When using the Page Share feature, data handling follows these principles:

    • Data sharing only occurs when you explicitly connect to external sources.
    • A self-hosting option is available for complete data control.
    • Shared chats can be permanently deleted from the server at any time.
    • No data is retained by the service after deletion.

  3. Manage System Settings and Data

    main

    Handle core application behavior, data synchronization, and backups:

    • Data Management:
      • Export All Data: Exports chat history, knowledge base, prompts, MCP servers, model configs, and settings to a JSON file. You can select specific sections.
      • Import All Data: Imports data from a Page Assist export file (Open WebUI exports are also supported).
      • System Reset: Clears all data (irreversible).
    • Sync & Integration:
      • Enable Browser Storage Sync (Sync settings across devices): Uses browser storage to sync settings.
      • Sync Custom Models, Prompts for Firefox Private Windows 🧪: (Firefox only) Syncs settings into private windows.
    • Interface & Shortcuts:
      • Set Default Action for Extension Icon Clicks 🧪: Choose between opening the Web UI or SidePanel.
      • Set Default action for Context Menu 🧪: Choose between opening the Web UI or SidePanel.
      • Show Web UI Button in Side Panel: Adds a button to open the Web UI from the SidePanel.

  5. Authenticate MCP servers with Bearer Tokens

    main

    If your MCP server requires an API key or bearer token, follow these steps:

    1. Go to MCP Settings.
    2. Click Add Custom Server.
    3. Select Bearer Token as the auth type.
    4. Enter your token.
    5. Click Save.

    The token will be sent as Authorization: Bearer <token> with every request.

  6. Implement screenshot support in page-assist-2 consumer

    main

    To support screenshots, the page-assist-2 consumer must be updated to handle image content returned by MCP tools. Instead of collapsing image blocks into text strings, the consumer should:

    1. Detect image content in the tool result.
    2. Convert the image block into a HumanMessage containing an image_url part (matching the shape used by humanMessageFormatter).
    3. Push this message into the lcConversation to allow vision models to process the pixels.
    4. Ensure a capability gate is in place so screenshots are only processed if the selected model is vision-capable.

  7. Change keyboard shortcuts in Firefox or Firefox-based browsers

    main

    To customize Page Assist shortcuts in Firefox, Zen, Librewolf, or other Firefox-based browsers:

    1. Navigate to about:addons.
    2. Click on the Settings icon.
    3. Click the Manage Extension Shortcuts button.
    4. Locate the Page Assist extension.
    5. Update the shortcut keys for your preferred actions.

  8. Use the Sidebar and Web UI

    main

    Page Assist provides two primary ways to interact with your local AI model:

    • Sidebar: Opens on any webpage to allow interaction with your model and view results. Access it via the context menu or the default keyboard shortcut Ctrl+Shift+Y.
    • Web UI: A full-page interface similar to ChatGPT. Access it by clicking the extension icon or using the default keyboard shortcut Ctrl+Shift+L.

  10. Resume the last chat in the Web UI

    main

    To continue a conversation exactly where you left off, enable the option to automatically resume the last active chat whenever you open the Web UI.

    1. Go to settings
2. Under the `General Settings` section
3. Enable `Resume the last chat when opening the Web UI` option

  12. Configure embedding models for Knowledge Base

    main

    To use the Knowledge Base feature, you must configure an embedding model in the RAG Settings.

    Requirements:

    • Use an embedding model that supports text (e.g., nomic-embed-text).
    • Do not use a text-to-text model for this purpose.

    Warning: Data is processed and embeddings are stored in browser storage because there is no server-side storage. This may lead to performance issues depending on the volume of data.