PRODUCTS / AI & MCP

Browser Companion

Browser extension for contextual AI assistance, structured page observation and constrained browser actions with checks on sensitive operations.

Project overview

Browser extension for contextual AI assistance, structured page observation and constrained browser actions with checks on sensitive operations.

For shipped features, installation requirements, version and license, consult the latest documentation in the public repository.

SECTION 01

Documented capabilities

Capabilities and tools described in the public project documentation.

What the software does

  • Chrome Manifest V3 extension with persistent side panel, page observation and constrained actions.
  • Background service worker mediates operations; local native host connects to CLI providers and optional HTTP AI services.
  • Deep Search collects public sources and produces reports, with follow-up questions and refinement.

Source and latest documentation ↗

SECTION 02

Architecture and integrations

Side panel → extension service worker → page probe or approved action executor → optional native host / AI provider.

Architecture overview

Side panel
extension service worker
page probe or approved action executor
optional native host / AI provider.

Source and latest documentation ↗

SECTION 03

Requirements and getting started

For development, load the JavaScript extension unpacked in Chrome and configure an available supported provider. Consult the repository for native-host installation and browser permissions.

Prerequisites

Chrome Manifest V3Unpacked extensionSupported AI provider

Quick start

  1. 1Download the repository and open Chrome Extensions.
  2. 2Enable Developer mode and load the unpacked extension.
  3. 3Configure a supported AI provider or native host.
SECTION 04

Limitations and operating boundaries

The model cannot execute arbitrary page code. Risky actions require confirmation; file selection is handed to the user. Deep Search and local-provider features have documented limits.

Key boundaries

Execution

The model cannot execute arbitrary page code.

Approval

Risky actions require confirmation; file selection is handed to the user.

File access

Deep Search and local-provider features have documented limits.

Source and latest documentation ↗

Need a custom integration?

This repository is a public software project, separate from paid professional AI consulting. For integration with existing systems, see the dedicated service.