Question 1

Is WebBrain a free alternative to Claude's browser plugin?

Accepted Answer

Yes. WebBrain provides similar AI browser agent capabilities — reading pages, extracting data, clicking buttons, filling forms, and automating multi-step workflows. Unlike Claude's proprietary browser plugin which requires a Claude Pro subscription and only works with Anthropic's models, WebBrain is completely free, open-source (GPL-3.0-or-later license), and supports multiple LLM providers including local models that run entirely on your machine.

Question 2

How much is a WebBrain Compass subscription?

Accepted Answer

WebBrain Compass is currently $5/month per device profile, much cheaper than Claude Pro for the time being. Usage is currently unlimited under a fair usage policy: normal personal use is fine, but abusive, automated, resale, or unusually high-volume use is not allowed. The subscription is linked to the device identifier generated by the extension for that browser + OS GUID, not to a user account. If that identifier is lost because you reset extension storage, delete the browser profile, reinstall the browser, change browser/OS, or lose the device, we cannot retrieve or transfer it and we do not issue refunds for that loss.

Question 3

How do I manage or cancel my subscription?

Accepted Answer

Manage billing at api.webbrain.one/account and use your email to log in.

Question 4

I already paid for WebBrain Compass, but the extension is asking me to pay again. Do I need to pay twice?

Accepted Answer

No — do not pay again. Your Compass subscription is linked to the browser/device identifier generated by the extension, so if that identifier changed (new browser or device, reinstalled browser, deleted browser profile, or reset extension storage), the extension no longer recognizes your existing subscription and asks for payment. Reattach your existing subscription to your current browser through the self-service portal at api.webbrain.one/portal and follow the instructions there to restore access. For receipts and other billing changes, manage billing at api.webbrain.one/account with your email.

Question 5

How does encrypted Cloud Sync work?

Accepted Answer

Cloud Sync is an optional feature for active WebBrain Compass subscribers. It syncs your WebBrain memories, Profile auto-fill text, and supported provider configuration, including API keys. It does not sync chat history, browser history, page screenshots, or legacy OAuth access and refresh tokens. Your browser encrypts the sync vault before upload with a password you choose. WebBrain Compass stores only the encrypted vault; it cannot read your memories, profile text, provider settings, or API keys, and it cannot recover the sync password for you. The email field is for WebBrain Compass magic-link billing/subscription authentication. It is not an account password. If the UI says change password in Cloud Sync, it means the sync encryption password, not a WebBrain account password. If you forget the sync password, the old encrypted cloud vault cannot be decrypted. You can reset the cloud vault and upload a new encrypted copy from a device that still has the data locally.

Question 6

How does WebBrain compare to OpenClaw, Browser-Use, and other AI agent frameworks?

Accepted Answer

They're different categories of tools. WebBrain is a browser extension — you install it in Chrome or Firefox and chat with it in a side panel, no coding required. Frameworks like OpenClaw and Browser-Use are developer SDKs for building automated browser pipelines in Python, typically using headless browsers and CDP. Think of it this way: WebBrain is for daily browsing with an AI assistant; agent frameworks are for building scraping bots and test automation. You can use both — they complement each other.

Question 7

Can I edit a WebBrain workflow?

Accepted Answer

Yes. Use the WebBrain Workflow Editor to inspect, edit, validate, and export a workflow JSON file in your browser.

Question 8

Can I use WebBrain completely offline?

Accepted Answer

Yes. WebBrain's default provider is llama.cpp, which runs a local AI model on your computer. No API keys needed, no internet required for the AI, and no data ever leaves your machine. Just download a GGUF model, start llama-server, and you have a fully private AI browser agent. You can also use Ollama with its OpenAI-compatible endpoint.

Question 9

What is the Emergency Box, and how does offline RAG work?

Accepted Answer

The Emergency Box is a curated collection of ~570 public-domain field-reference documents (medical, survival, education, communication) that WebBrain can download and index locally. Once installed, you can ask factual questions offline and get answers sourced from those documents — no internet connection required. Under the hood, WebBrain uses SQLite FTS5 for fast keyword search across both the Emergency Box and any installed Wikipedia archives, plus an optional semantic search layer using a locally downloaded multilingual E5 model (~140 MB). Results from both methods are combined and diversified, and every answer cites the exact source passage. Citations open in a local reader that verifies document integrity before displaying — they never navigate to a live web page. You can also install Wikipedia archives through Apocalypse Mode for a broader offline knowledge base. Source and language filters let you control which collections are searched.

Question 10

Which AI models does WebBrain support?

Accepted Answer

WebBrain supports four provider types: llama.cpp (any local GGUF model), OpenAI (GPT-4o, GPT-4, etc.), Claude (Claude Opus, Sonnet, Haiku via native API), and OpenRouter (access to 100+ models from various providers). Any OpenAI-compatible API endpoint works, so you can also use services like Together AI, Groq, Mistral, or any local server with an OpenAI-compatible interface.

Question 11

What's the most recommended model?

Accepted Answer

As of April 21, 2026, the top recommendation is Qwen 3.6 35B. Why: in our vision benchmark (vision-model-shootout), it outperformed Gemma 4 on screenshot understanding while remaining practical for local inference. For consumer GPUs, RTX 5090 is ideal, and RTX 4090 is often workable with INT4 AutoRound quantization via Intel/Qwen3.6-35B-A3B-int4-AutoRound. For max speed, we recommend serving on vLLM. Sample command: python -u -m vllm.entrypoints.openai.api_server --model Intel/Qwen3.6-35B-A3B-int4-AutoRound --served-model-name qwen3.6-35b --quantization auto --dtype bfloat16 --max-model-len 65536 --max-num-batched-tokens 32768 --max-num-seqs 4 --host 0.0.0.0 --port 8000 --gpu-memory-utilization 0.92 --enable-prefix-caching --enable-chunked-prefill --limit-mm-per-prompt '{"image": 4, "video": 1}' --mm-processor-cache-type shm --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder --trust-remote-code --allowed-origins '["*"]' --speculative-config '{"method": "dflash", "model": "z-lab/Qwen3.6-35B-A3B-DFlash", "num_speculative_tokens": 15}' --attention-backend flash_attn DFlash speculative decoding is optional.

Question 12

Why is WebGPU only available in Apocalypse Mode?

Accepted Answer

Because we do not think browser-based WebGPU text inference is good enough yet for most users, and we do not want you to suffer through a poor experience. If you want to use the same LFM2.5 2.6B model locally, run inference on your machine with llama.cpp or Ollama and point WebBrain to that local endpoint instead of relying on WebGPU. The WebGPU vision fallback remains available because its model is much smaller and the fallback is used only occasionally.

Question 13

I'm getting "Failed to fetch" when connecting to a local LLM server (vLLM, Ollama, llama.cpp) on my network

Accepted Answer

If your LLM server is on a different machine on your local network (e.g. http://192.168.1.x:8000), Chrome blocks the request unless the server sends CORS headers. The fix depends on your server: vLLM: Start with --allowed-origins '["*"]' (the value must be a JSON list). Ollama: Set the environment variable OLLAMA_ORIGINS=* before starting. llama.cpp: CORS is enabled by default — no changes needed. If your server runs on localhost (same machine as the browser), CORS is usually not required. The issue only affects cross-machine connections on the local network. Make sure your base URL in WebBrain settings ends with /v1 (e.g. http://192.168.1.47:8000/v1). Exception: Ollama used from a browser extension may still need OLLAMA_ORIGINS; see the Ollama FAQ below.

Question 14

Why does Ollama on localhost return 403 from the WebBrain extension?

Accepted Answer

Recent Ollama versions can reject requests from browser-extension origins even when Ollama is running on the same machine. The extension sends an Origin such as chrome-extension://... or moz-extension://..., and Ollama may answer 403 unless those origins are allowed. Quit any Ollama desktop app that is already holding port 11434, then start Ollama with one of these commands: OLLAMA_ORIGINS="*" ollama serve OLLAMA_ORIGINS="chrome-extension://*,moz-extension://*" ollama serve Keep WebBrain's Ollama base URL set to http://localhost:11434/v1. Terminal curl checks can still work without this setting because they do not send a browser-extension origin.

Question 15

Does WebBrain work on Firefox?

Accepted Answer

Yes. WebBrain ships with both a Chrome version (Manifest V3, using the sidePanel API) and a Firefox version (Manifest V2, using sidebar_action). Both versions have identical features. The Firefox version can be loaded as a temporary add-on for development, or published to addons.mozilla.org for permanent installation.

Question 16

Can I move the Firefox sidebar from the left to the right, like Chrome's side panel?

Accepted Answer

Yes — Firefox's sidebar defaults to the left, but you can flip it. Right-click anywhere in the sidebar header and choose Move Sidebar to Right (or use View → Sidebar → Move Sidebar to Right from the menu bar). The position persists across restarts. Chrome's sidePanel defaults to the right and isn't user-movable from the panel itself.

Question 17

Are there any limitations when using WebBrain in Vivaldi's Web Panel?

Accepted Answer

Yes. Vivaldi Web Panels suppress native JavaScript dialogs such as confirm(), alert(), and prompt(). WebBrain's Ask, Act, and Dev mode switch no longer depends on those dialogs, but a few secondary actions still do. Recording: Do not use /record in Vivaldi for now. If recording fails, Vivaldi can hide the error alert and leave no visible explanation. Settings, History, Traces, and Ollama handoff: These also contain native dialogs, but WebBrain normally opens them as regular tabs, where they work. If you manually place one in a Web Panel, open it in a normal tab before using dialog-dependent actions.

Question 18

Is WebBrain safe to use? Can it modify web pages?

Accepted Answer

WebBrain has three modes. Ask mode (default) is read-only and cannot modify anything on the page. Act mode enables browser actions such as clicking, typing, and navigating, while Dev mode adds page-debugging tools. Consequential actions require approval by default; if that permission gate is turned off, WebBrain displays a visible warning banner. You can stop the agent at any time with the Stop button. The extension's source code is fully open for audit on GitHub.

Question 19

What is the "WebBrain started debugging this browser" banner? Why does WebBrain use CDP?

Accepted Answer

To act on a page reliably, WebBrain uses the Chrome DevTools Protocol (CDP) through the standard chrome.debugger extension API — that's what triggers Chrome's "WebBrain started debugging this browser" banner. CDP is what lets the agent click and type with trusted input events that modern sites actually honor; synthetic events fired from a content script get rejected by many sites, Web Components, and framework-controlled inputs. It's also how WebBrain captures pixel-accurate screenshots for the vision fallback and reaches into cross-origin iframes and shadow DOM that content scripts can't see. Read-only Ask mode does not need CDP — page and accessibility-tree reads go through ordinary content scripts. It's Act mode reliability and cross-origin work that require it, and in a Manifest V3 extension there is no non-debugger API for trusted input. WebBrain attaches the debugger only when an action needs it, per tab, and the whole extension is open-source so you can audit exactly what the session does. CDP is powerful, so we treat its footprint as something to contain — attach on demand, keep reads off it — and tightening that further is on our security roadmap.

Question 20

It's asking me too many questions. Can I disable that?

Accepted Answer

Those approval prompts are for your safety. Even on sites you trust, malicious actors can mimic LLM instructions or inject content that tries to make the agent do things on your behalf that you would not approve. That is why WebBrain asks before consequential actions. If you still do not want to go through those prompts, type /dangerously-skip-permissions in the side panel, or open the side plugin, click the wheel icon in the header, go to the Permissions tab in settings.html, and switch off Ask before consequential actions.

Question 21

How do I use WebBrain for web scraping and data extraction?

Accepted Answer

Simply open any web page, open the WebBrain side panel, and ask in natural language: "Extract all product names and prices from this page", "Get all email addresses on this page", or "Summarize this article in bullet points". The AI agent reads the page content, understands the structure, and returns the extracted data. For more complex scraping, switch to Act mode and the agent can navigate between pages, click pagination buttons, and aggregate data across multiple pages.

Question 22

Does WebBrain call APIs directly, or does it always click through the UI?

Accepted Answer

By default, WebBrain always goes through the visible UI for any action that creates, modifies, deletes, sends, submits, posts, or buys anything. It will navigate to the page, fill the form, and click the button — exactly the way you would. It refuses to call REST/GraphQL endpoints directly via background fetch() for mutations. This is deliberate: API actions are invisible (you don't see what's being sent), often require separate auth tokens you may not have configured, and have a much larger blast radius than a visible mis-click. UI-first means everything is on screen, in your normal browser session, and stoppable. For reading data — fetching a README, looking up an issue, comparing prices across sites, checking a status page — WebBrain freely uses background HTTP requests via the fetch_url and research_url tools. Reading is not the same as acting; it doesn't change anything on a remote service, so the safety concerns don't apply. If you specifically want to allow API mutations for a particular task, type /allow-api at the start of your message (optionally followed by a short task description). This per-conversation override lets WebBrain fall back to API endpoints when the UI is genuinely failing or unworkable, while still preferring UI when UI works. A sticky badge stays visible above the input area while the override is active, and it clears when you reset the conversation.

Question 23

Can I use it in LM Studio too?

Accepted Answer

Yes. WebBrain's read-only network tools — fetch_url and research_url — also ship as a standalone LM Studio plugin at webbrain/web-tools. Install with lms clone webbrain/web-tools and toggle it on in any LM Studio chat — any tool-capable model can then call those two tools without you installing the browser extension. Pure Node, no headless browser. Source: lmstudio-plugin/.

Question 24

Can I scroll up or down while WebBrain is working on my tab?

Accepted Answer

Most of the time, yes — especially in Ask mode for read-only requests such as "Summarize this page." You can continue scrolling while WebBrain works. When should I avoid it? Auto-screenshots reflect the tab's current viewport. While WebBrain is actively inspecting or interacting with visible content in Act or Dev mode, scrolling can move a target or change what the model sees. Let that step finish before scrolling; outside those moments, you can continue using the page.

Question 25

Can I switch tabs while WebBrain is working?

Accepted Answer

Yes. As of 27.0.0, WebBrain avoids stealing focus in both Chromium-based browsers and Firefox unless foreground access is strictly necessary. The run stays attached to the tab where it started, so you can safely work in another tab. If a rare step needs the original tab in front, WebBrain may briefly focus it. Avoid clicking or typing in the same tab while WebBrain is actively acting on it.

Question 26

How does Profile auto-fill work, and is it safe?

Accepted Answer

Profile auto-fill is an optional feature in Settings → Profile. You enter a short bio — name, work email, company, and a throwaway password for low-stakes signups — and toggle it on. When enabled, WebBrain appends that text to the agent's system prompt so it can fill signup forms without asking every time. The text is stored in plaintext in your browser's local storage. It is not transmitted to the WebBrain project, but it is sent to whichever LLM provider you have configured on every turn, as part of the system prompt. Off by default. Do not put passwords for important accounts (Google, Apple, iCloud, banking, work SSO, primary email) here. Those accounts should use 2FA and shouldn't be handed to an agent anyway. A throwaway password you reuse for newsletter signups and free trials is the intended use case.

Question 27

Does the Screenshot redaction setting guarantee my private data never reaches a cloud vision model?

Accepted Answer

No — it reduces exposure, it is not a guarantee. When enabled (Settings → Assistive Models → Vision → Screenshot redaction, off by default), WebBrain pixelates form fields and text that looks like an email address or phone number on every screenshot before it is sent to a vision model. Detection runs entirely on your device using DOM heuristics; nothing extra is transmitted. It is best-effort and fail-open. If the detector cannot run on a page — for example immediately after a navigation, on PDF viewers, or on restricted browser pages — the screenshot is still sent unredacted rather than aborting your task. Content the DOM heuristics cannot see (text drawn on a canvas, PII inside images, unusual widgets) may also slip through. And the setting only covers screenshots: page text sent to the model is not redacted by it. For full vision privacy, use a local model. With an offline provider such as llama.cpp or Ollama, screenshots and page text never leave your machine, so there is nothing to redact in the first place — that is the only configuration where privacy is guaranteed. See "Can I use WebBrain completely offline?" above. Don't have a machine that can run a vision-capable local model? A good middle ground is a small local model for planning and tool calls (fast and light enough for a modest CPU/GPU) paired with a cloud provider for the vision sub-calls, with Screenshot redaction turned on. You still keep your conversation and page text on-device, and every screenshot that does leave for cloud vision has form fields and detected emails/phones blurred first — reduced cloud exposure without needing hardware for a full local vision model. Configure the vision split under Settings → Assistive Models → Vision (see "How does WebBrain keep cloud LLM bills under control?" above).

Question 28

What does WebBrain do with cookie banners and paywalls?

Accepted Answer

Cookie banners: WebBrain recognizes consent banners from common frameworks (OneTrust, Cookiebot, Didomi, Quantcast, Google Funding Choices, TrustArc) and dismisses them before reasoning about the page. Priority is "Reject all" / "Reject non-essential" / "Only necessary" when clearly visible; it falls back to "Accept all" rather than disappearing into the "Manage preferences" maze. Paywalls: WebBrain reports the paywall honestly and tells you what it could actually see (headline, dek, first paragraphs). It does not attempt to bypass paywalls — no archive.today, 12ft.io, cookie clearing, JS disabling, or reader-mode tricks. If you want the full article, log in with a subscription or ask WebBrain to look for free coverage of the same story.

Question 29

Does WebBrain support a dry-run mode?

Accepted Answer

As of 7.0.0, not yet. Dry-run mode is planned and already on the roadmap.

Question 30

Does WebBrain send the whole DOM to the AI model?

Accepted Answer

No. WebBrain does not send raw HTML or a raw DOM dump by default. It starts with the page URL and title, matching site guidance, and an optional viewport screenshot when vision is enabled. When a task needs page content, WebBrain reads it on demand as a reduced semantic accessibility tree or extracted text. Reads are visibility-filtered where appropriate, character-capped, and paginated. Raw page-source access is available only in Dev mode.

Question 31

Why is WebBrain 33.0.0 and later GPL-licensed?

Accepted Answer

WebBrain 33.0.0 and later is distributed under GPL-3.0-or-later because the browser extension bundles and integrates the GPL-licensed Xapian/libzim WebAssembly runtime for full-text Wikipedia ZIM search. The extension packages include the applicable GPL notices, and complete corresponding source for the bundled runtime is published alongside each release. Releases before 33.0.0 remain MIT-licensed, and independently published subprojects and third-party components retain their stated licenses. See the licensing decision record for details.

Question 32

How does WebBrain keep cloud LLM bills under control?

Accepted Answer

Three independent layers: Token-conscious screenshots. Before any image leaves your machine, WebBrain resizes it (shorter side capped, preserving aspect ratio) and iteratively JPEG-compresses it until it fits a per-turn image-token budget. A 2000×1200 screenshot that would cost you ~1,500 input tokens on GPT-4o gets compressed down to ~300–500 tokens with no practical loss for page-reading tasks. Implemented in _fitImageDimensions with unit tests for the budget math. Smart context trimming. Conversation history, tool outputs, and inline DOM dumps are bounded per turn and trimmed oldest-first when the active model's context window is approaching full. You won't see a run silently balloon from 10k tokens to 100k because a read_page returned a novel-length article. Dedicated vision model. Pair a cheap text model (e.g. GPT-4o-mini) for planning and tool calls with a separate vision-capable model (e.g. GPT-4o) only for screenshots, so you don't pay multimodal-model prices on every single turn. Configure under Settings → Assistive Models → Vision. Net result: long sessions with cloud providers stay predictable. For full control, use llama.cpp locally — zero per-token cost.

Question 33

Can I contribute to WebBrain?

Accepted Answer

Absolutely! WebBrain is GPL-3.0-or-later-licensed and welcomes contributions. Check out the GitHub repository for issues, feature requests, and contribution guidelines.

Feature	WebBrain	Claude in Chrome
Open Source	GPL-3.0-or-later License	Proprietary
Price	Free forever	Requires Claude Pro ($20/mo)
Local LLM support	llama.cpp, Ollama	No — Claude only
Multi-provider	All OpenAI-compatible endpoints	Claude only
Side panel UI	Yes	Yes
Fully offline	Yes (with local LLM)	No — cloud required
Self-hostable	Yes	No

Aspect	WebBrain	OpenClaw / Browser-Use / etc.
What is it?	Browser extension (end-user tool)	Agent framework / SDK (developer tool)
Target user	Anyone — no coding needed	Developers building automations
Installation	One-click browser install	Python/Docker setup required
UI	Built-in side panel chat	No UI — code or API only
Browser control	Content script (lightweight)	CDP / Playwright (full control)
Multi-tab workflows	Per-tab conversations	Programmable multi-tab orchestration
Headless mode	No — runs in your browser	Yes — headless automation
Extensibility	Add custom LLM providers	Full Python SDK, custom tools
Best for	Daily browsing AI assistant	Automated scraping / testing pipelines

The Open-Source AI Browser Agent

The Future of Local AI

Quarterly Tax Return — Form QT-4

Product Catalog

How WebBrain keeps you in control

Watch WebBrain in action

How WebBrain keeps you in control

Local WebBrain beats Claude in Chrome, side by side

Word on the web

Bring your own AI

Everything you need in a browser AI

Page Understanding

Full Browser Agent

Data Extraction

Token-Conscious

Smart Context

Dedicated Vision Model

Cookie & Paywall Aware

Profile Auto-fill

Optional CAPTCHA Solver

How does WebBrain compare?

vs. Browser AI Plugins

Frequently Asked Questions

Install WebBrain

Chrome & Chromium

Firefox

Microsoft Edge

Your coding agent, inside the browser you already use.

Delegate the goal. Keep the session.

vs. AI Agent Frameworks (different category)

WebBrain, ready when the internet isn’t.

Local WebBrain beats Claude in Chrome, side by side

WebBrain, ready when the internet isn’t.

Spread the word, share the love