🚀 Latest: v1.0.0 By Abdul karim mia

Your Web Browser.
Driven by AI.

BrowseCortex is an open-source Chrome extension giving you a persistent AI companion with autonomous browser control. Bring your own AI provider and watch it navigate, complete tasks, and manage files.

No Vendor Lock-in
Local & Private
MCP Connected
cortex-agent-loop
Groq: llama3-70b
U
Find the best tech events in San Francisco happening this week and compile them into a text file.
Skills 📎 Attach
Analyzing web sources...
https://www.google.com
🧠

Designed for Power. Built for Privacy.

A feature set engineered for complete independence and absolute browser control.

Bring Your Own Provider & Model Tuning

Connect to Groq, OpenAI, Mistral, Together AI, or local engines like Ollama. Tune parameters directly: choose your preferred agent mode (Full Auto, Notify Only, Confirm Destructive), adjust reasoning effort (Low/Medium/High), and set custom safety loop caps.

Advanced Browser Tools

Equipped with 55+ browser tools including tab group management, session controls, cookies reading/deletion, clipboard read/write, history searches, and bookmarks manipulation.

Virtual Sandboxed Filesystem

A secure workspace built on top of IndexedDB. The agent can write files, create folders, pack data into zip structures locally, and export directly to your downloads directory.

Context Compaction & Annotation

Features automatic context compaction that tokenizes and summarizes older turns to save context window space. Complemented by DOM Annotation overlays that assign numbered visual identifiers to interactable elements, avoiding expensive raw HTML payloads.

Repository Health

BrowseCortex is actively developed and updated by the open-source community. Key repository stats:

-- Stars
-- Forks
-- Open Issues

Creator & Maintainer

BrowseCortex was designed and built by developer Abdul karim mia. You can check his full work, developer portfolio, and other active projects at:

abdulkarimmia.in →

Advanced Core Capabilities

A closer look at the advanced extension mechanics running under the hood. (Click to expand details)

+

Model Context Protocol (MCP)

Integrate local and remote MCP servers directly. BrowseCortex dynamically consumes external server schemas, injecting custom tool capabilities directly into the active agent loop.

  • Connects local stdio/SSE servers via the Node.js WebSocket relay
  • Supports remote HTTPS/SSE MCP tool integrations
  • Autoloads schemas and injects tools dynamically into LLM prompts
+

Smart Cooldowns & Failovers

Automatic rate-limiting failover routing. If a provider returns a 429 backoff error, the agent places that endpoint on cooldown and automatically routes execution to alternative APIs.

  • Detects HTTP 429 and rate-limiting responses immediately
  • Applies exponential cooldown backoff times (up to 5 minutes)
  • Intelligently fails over to secondary user-configured models
+

Intelligent Vision Fallbacks

If a configured text model lacks native vision (like Llama-70b), BrowseCortex intelligently routes page screenshot analysis tasks to a secondary designated vision model (like GPT-4o).

  • Checks model compatibility tags from the LiteLLM database
  • Captures tab viewports natively using Chrome extension APIs
  • Transports base64 frames to secondary vision models automatically
+

Subtask & Checklist Manager

Organizes agent commands into checklists. The AI creates, updates, and completes subtasks dynamically, displaying visual progress bars in the extension's dedicated Tasks tab.

  • AI self-manages tasks using task creation and updating tools
  • Stores status configurations in a local IndexedDB store
  • Features real-time checklist renders directly in the Side Panel UI
+

AES-256-GCM Secure Backups

All extension configurations, conversations, files, and histories can be encrypted using local Web Crypto APIs and backed up/restored in a single click.

  • Uses Web Crypto API to generate local cryptographic keys
  • Encrypted ZIP files stored locally; no cloud upload required
  • Features daily auto-backup snapshot alarms
+

Personalized Prompt Additions

Inject customized rules and guidelines globally. Guide the agent's tone, focus area, coding standards, or behavior boundaries on every conversation loop.

  • Injects custom directives directly into system prompt layers
  • Applies preferences across all configurations and providers
  • Provides tone matching, coding styles, or output guidelines

Why Choose BrowseCortex?

A direct comparison of features and models against standard closed-source web agents.

Feature BrowseCortex Closed-Source Competitors
Pricing Structure 100% Free & Open Source (Direct pay-as-you-go keys) Expensive Monthly Subscriptions (with markups)
AI Provider Flexibility Bring Your Own Key (Groq, OpenAI, Mistral, Ollama) Locked to single pre-selected providers
Data & Credentials Privacy 100% Local (Stored directly in Chrome storage) Proxied through external cloud environments
Local Offline Support Supported (Ollama/LM Studio on localhost) No offline capabilities (Cloud reliant)
Model Context Protocol (MCP) Supported (Connect local and remote tools) No user extensions or MCP adapters
Licensing MIT License (Open to customize, host & fork) Proprietary closed systems

Get Started in Minutes

Setting up BrowseCortex is straightforward and requires no complex setups.

01

Install the Extension

Load the packed extension or build it from source and load unpacked into Chrome Developer Mode.

02

Configure Provider

Enter your favorite provider's API key. You can also connect to local models (e.g. Ollama) with zero keys required.

03

Prompt and Automate

Open the Side Panel, type a command, and watch the agent navigate, read, and execute steps on your behalf.

Drive the Extension from Local CLIs & MCP Agents

BrowseCortex features a built-in Model Context Protocol (MCP) bridging subsystem. By running our lightweight Node.js relay, external tools and CLI agents (like Claude Code) can utilize the extension's browser tools over a secure local SSE websocket.

Bidirectional SSE WebSocket translation
Token-based secure authentication
Proxy initialization & parallel tool calls
Terminal
$ npm install -g browsecortex-relay
$ browsecortex-relay --token my-secret-auth-token --port 7822
BrowseCortex relay listening on http://localhost:7822
MCP SSE endpoint: http://localhost:7822/sse
Extension WS: ws://localhost:7822/ws?token=my-secret-auth-token
Extension connected. Active & listening...

Open Source Contributors

Thanks to the developers who have contributed to the BrowseCortex core and tools.

Fetching contributors list...

Frequently Asked Questions

Answers to common inquiries about BrowseCortex architecture and operations.

Is BrowseCortex completely free?

Yes. The extension code is 100% open-source and free under the MIT license. You do not pay any service fees. You only pay the LLM API provider of your choice directly for the token volume you consume.

Where are my API keys stored?

Your API keys are stored securely in Chrome's local database (`chrome.storage.local`). They are kept sandboxed within the extension, never synchronized to external servers, and only transmitted to the API endpoint you explicitly specify.

Does this support local LLM models?

Yes, it fully supports local models via tools like Ollama, LM Studio, Jan, or local LiteLLM servers. If your local engine is running on localhost, BrowseCortex will interface with it directly without requiring an internet connection or API keys.

What browser permissions does it require?

BrowseCortex requires standard API permissions like `tabs`, `scripting`, `history`, `downloads`, and `activeTab` to perform navigation actions and read page text. It only runs code in browser tabs programmatically when you actively trigger a chat agent session.

Ready to automate your browsing?

Download BrowseCortex today and build your own autonomous browser pipelines.