{"id":61939,"date":"2026-07-03T14:31:35","date_gmt":"2026-07-03T18:31:35","guid":{"rendered":"https:\/\/overcentral.com\/en\/?p=61939"},"modified":"2026-07-03T14:31:35","modified_gmt":"2026-07-03T18:31:35","slug":"webbrain-open-source-ai-browser-agent","status":"publish","type":"post","link":"https:\/\/overcentral.com\/en\/webbrain-open-source-ai-browser-agent\/","title":{"rendered":"WebBrain Launches Open-Source AI Browser Agent for Chrome and Firefox"},"content":{"rendered":"

An open-source browser agent that runs entirely on a local model, operates inside your authenticated sessions, and never sends page data to a cloud server is now available for Chrome and Firefox. WebBrain, built by Emre Sokullu and released under the MIT license, is a free browser extension that reads pages, extracts data, and automates multi-step tasks through a side-panel chat interface. Its full source code is available on GitHub<\/a>, and it supports both local inference and cloud API endpoints depending on the user’s privacy and capability requirements.<\/p>\n

What Is WebBrain and How Does It Work<\/h2>\n

WebBrain lives in the browser’s side panel \u2014 using Manifest V3 and the sidePanel API in Chrome, and Manifest V2 with sidebar_action in Firefox. Each tab maintains its own separate conversation history. The extension operates within the user’s existing authenticated session, meaning it sees logged-in accounts exactly as the user does. It stores no data externally, adds no telemetry, and requires no separate account.<\/p>\n

The interface ships in English, Spanish, French, Turkish, and Chinese, and auto-detects the browser language on first launch. This makes it immediately usable across multiple regions without configuration.<\/p>\n

Ask Mode vs Act Mode: Two Distinct Capability Levels<\/h2>\n

WebBrain separates its functionality into two modes. Ask mode is read-only: it can read page content through ordinary content scripts but cannot change anything on the page. Act mode can click, type, scroll, navigate, and execute workflows. The difference in how these modes work under the hood is significant for anyone evaluating the extension’s reliability.<\/p>\n

Act mode drives the page through the Chrome DevTools Protocol via the chrome.debugger API. This produces trusted input events that modern websites actually honor, and it can reach cross-origin iframes and shadow DOM elements that standard content scripts cannot access. Chrome displays its standard “WebBrain started debugging this browser” banner while the debugger is attached. Firefox has no equivalent to the Chrome DevTools Protocol, so its Act mode is meaningfully more limited.<\/p>\n

Temperatures are fixed for predictability: Act mode uses 0.15, Ask mode uses 0.3, and dedicated vision screenshot descriptions use 0.<\/p>\n

The Security Model: Built for an Adversarial Surface<\/h2>\n

Browser agents operate on an adversarial surface where web pages can hide prompt injections designed to hijack an agent’s behavior. WebBrain’s design addresses this directly. The agent starts in read-only Ask mode and asks for approval before consequential actions. Users can disable these prompts in the Permissions settings, but they are on by default.<\/p>\n

For mutations \u2014 anything that creates, sends, submits, or buys \u2014 WebBrain uses the visible UI. It refuses to call REST or GraphQL endpoints directly for mutations. A per-conversation \/allow-api override exists for cases where the UI genuinely fails. Reading operations are treated separately: fetching a README or comparing prices uses background HTTP through the fetch_url and research_url tools, since reading changes nothing remotely.<\/p>\n

Use Cases: Data Extraction, Research, Form Filling, and Automation<\/h2>\n

The practical applications span several common workflows. Data extraction is the most straightforward: open a catalog and ask the agent to extract all product names and prices. The agent reads the structure and returns rows. It also works with PDFs.<\/p>\n

Research summaries are another natural fit. Users can ask WebBrain to summarize an article and then follow up with specific questions. The extension detects paywalls honestly and does not attempt to bypass them. It also dismisses common cookie-consent banners before reading.<\/p>\n

Form filling suits repetitive signups. An optional Profile auto-fill stores a short bio in local plaintext, which is sent to the configured LLM to complete low-stakes forms. The documentation advises keeping important passwords out of this feature.<\/p>\n

Automation spans multiple steps. In Act mode, users can ask the agent to navigate to a site and find specific content, and the agent chains navigation, reading, and clicking to complete the task.<\/p>\n

Cost Management: Keeping Token Usage in Check<\/h2>\n

Cloud tokens accumulate quickly on long sessions. WebBrain bounds costs in three ways. Screenshots are resized and iteratively JPEG-compressed before they leave the machine, keeping image tokens small. Conversation history and tool outputs are trimmed oldest-first as the context window fills. Users can also pair a cheap text model for planning with a separate vision model for screenshots.<\/p>\n

How WebBrain Compares to Other Browser AI Tools<\/h2>\n

WebBrain sits between lightweight browser AI plugins and full agent frameworks. Unlike proprietary plugins that require a subscription, WebBrain is free forever under the MIT license and supports local LLM inference through llama.cpp or Ollama. It works with all OpenAI-compatible endpoints, not just a single provider. Chrome support uses Manifest V3, and Firefox support uses Manifest V2 \u2014 a combination few alternatives offer.<\/p>\n

Frameworks like OpenClaw or Browser-Use are a different category entirely. Those are developer SDKs for headless pipelines. WebBrain is an end-user extension driven from a chat panel. Both can be used together if a project requires headless automation alongside interactive control.<\/p>\n

Running WebBrain: Local and Cloud Providers<\/h2>\n

WebBrain supports local options including llama.cpp, Ollama, LM Studio, Jan, vLLM, and SGLang. Cloud options include OpenAI, Anthropic Claude, Gemini, Mistral, DeepSeek<\/a>, xAI Grok, Groq, MiniMax, Alibaba Cloud (Qwen), Nvidia NIM, and OpenRouter. A built-in managed option called WebBrain Cloud costs $5 per month per device profile under a fair-use policy and needs no local setup.<\/p>\n

For local use, llama.cpp requires no API key. Starting a local server takes a single command, and users point WebBrain at the endpoint in settings. The recommended model is Qwen 3.6 35B (Qwen3.6-35B-A3B), which the project reports beat Gemma 4<\/a> on its screenshot benchmark. An RTX 5090 is ideal; an RTX 4090 works with INT4 AutoRound quantization.<\/p>\n

Who Should Try WebBrain Now<\/h2>\n

WebBrain is immediately available as a free, self-hosted browser agent for anyone using Chrome or Firefox. Developers and power users who want to keep their browsing data entirely local can run it with llama.cpp or Ollama and a supported model like Qwen 3.6 35B. Users who prefer a managed experience can try WebBrain Cloud for $5 per month per profile. The extension’s security-first design \u2014 starting in read-only mode and requiring approval for mutations \u2014 makes it a practical option for professionals who need automation without compromising on data privacy. Download the extension from the Chrome Web Store<\/a> or Firefox Add-ons<\/a> marketplace, or clone the repository from GitHub to inspect the source or self-host.<\/p>\n","protected":false},"excerpt":{"rendered":"

An open-source browser agent that runs entirely on a local model, operates inside your authenticated sessions, and never sends page data to a cloud server is now available for Chrome and Firefox. WebBrain, built by Emre Sokullu and released under the MIT license, is a free browser extension that reads pages, extracts data, and automates […]<\/p>\n","protected":false},"author":7,"featured_media":84405,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"fifu_image_url":"https:\/\/cards.overcentral.com\/cards\/en\/61939.png","fifu_image_alt":"WebBrain Launches Open-Source AI Browser Agent for Chrome and Firefox","footnotes":""},"categories":[349],"tags":[],"class_list":["post-61939","post","type-post","status-publish","format-standard","has-post-thumbnail","category-articles"],"fifu_image_url":"https:\/\/cards.overcentral.com\/cards\/en\/61939.png","fifu_image_alt":"WebBrain Launches Open-Source AI Browser Agent for Chrome and Firefox","_links":{"self":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/61939","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/users\/7"}],"replies":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/comments?post=61939"}],"version-history":[{"count":0,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/posts\/61939\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/media\/84405"}],"wp:attachment":[{"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/media?parent=61939"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/categories?post=61939"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/overcentral.com\/en\/wp-json\/wp\/v2\/tags?post=61939"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}