Meet WebBrain: An Open-Source, Local-First AI Browser Agent That Reads Pages and Automates Tasks in Chrome and Firefox

WebBrain is a free, open-source browser agent for Chrome and Firefox. It reads pages, extracts knowledge, and automates multi-step duties. Unlike most browser AI plugins, it could possibly additionally run totally on a neighborhood mannequin.

It is constructed by Emre Sokullu and licensed below MIT. The full supply lives on GitHub.

Run the agent in opposition to a neighborhood mannequin, and no web page knowledge leaves your machine. Connect a cloud API if you need extra functionality.

What is WebMind?

WebBrain lives in your browser’s facet panel. In Chrome it makes use of Manifest V3 and the sidePanel API. In Firefox it makes use of Manifest V2 and sidebar_action. Each tab retains its personal dialog historical past.

The extension operates inside your current authenticated session. It sees your logged-in accounts precisely as you do. It shops no knowledge externally and provides no telemetry or accounts.

The plugin ships in English, Español, Français, Türkçe, and 中文. It auto-detects your browser language on first launch.

Ask Mode, Act Mode, and How Actions Actually Fire

WebBrain has two modes: Ask mode is read-only and can not change the web page. Act mode can click on, sort, scroll, navigate, and run workflows.

Ask mode reads pages by way of atypical content material scripts. Act mode is completely different. It drives the web page by way of the Chrome DevTools Protocol through the chrome.debugger API. That produces trusted enter occasions that trendy websites truly honor. It additionally reaches cross-origin iframes and shadow DOM that content material scripts can not see.

That energy is scoped intentionally. WebMind attaches the debugger solely when an motion wants it, per tab. Chrome surfaces its customary ‘WebMind began debugging this browser’ banner whereas hooked up. Firefox has no CDP equal, so its Act mode is meaningfully weaker.

Temperatures are fastened for predictability. Act mode makes use of temperature 0.15. Ask mode makes use of 0.3. Dedicated imaginative and prescient screenshot descriptions use 0.

The Security Model

Browser brokers run on an adversarial floor. Web pages can cover immediate injections that hijack an agent’s habits. WebMind’s design addresses this instantly.

The agent begins in read-only Ask mode. It asks earlier than consequential actions. You can disable these prompts in the Permissions settings. They are on by default.

There can be a UI-first rule for mutations. For something that creates, sends, submits, or buys, WebMind makes use of the seen UI. It refuses to name REST or GraphQL endpoints instantly for mutations. A per-conversation /allow-api override exists when the UI genuinely fails.

Reading is handled individually. Fetching a README or evaluating costs makes use of background HTTP by way of the fetch_url and research_url instruments. Reading modifications nothing remotely, so the strict guidelines don’t apply.

Use Cases, With Concrete Examples

Data extraction is the plain one: Open a catalog and ask: ‘Extract all product names and costs from this web page.’ The agent reads the construction and returns rows. It additionally works with PDFs.
Research summaries are one other: Ask ‘Summarize this text,’ then observe up with a particular query. WebMind detects paywalls truthfully and doesn’t attempt to bypass them. It additionally dismisses widespread cookie-consent banners earlier than studying.
Form filling fits repetitive signups: An non-compulsory Profile auto-fill shops a brief bio in native plaintext. That textual content is distributed to your configured LLM to finish low-stakes kinds. Keep essential passwords out of it.
Automation spans a number of steps: Try ‘Navigate to github.com and discover trending repositories.’ In Act mode, the agent chains navigation, reads, and clicks.

Keeping Token Costs Down

Cloud tokens add up on lengthy classes. WebMind bounds the fee in 3 ways.

Screenshots are resized and iteratively JPEG-compressed earlier than they depart your machine. That retains picture tokens small.
Conversation historical past and software outputs are trimmed oldest-first because the context window fills.
You can even pair an affordable textual content mannequin for planning with a separate imaginative and prescient mannequin for screenshots.

How It Compares

WebBrain sits between browser AI plugins and full agent frameworks. Here is the plugin comparability, drawn from the undertaking’s personal documentation.

Feature	WebMind	Claude in Chrome
Open supply	MIT License	Proprietary
Price	Free without end	Requires Claude Pro ($20/mo)
Local LLM assist	llama.cpp, Ollama	No — Claude solely
Multi-provider	All OpenAI-compatible endpoints	Claude solely
Chrome	Yes (MV3)	Yes
Firefox	Yes (MV2)	No
Side panel UI	Yes	Yes
Ask / Act modes	Yes	Similar
Fully offline	Yes (with native LLM)	No — cloud required
Self-hostable	Yes	No

Frameworks like OpenClaw or Browser-Use are a distinct class. Those are developer SDKs for headless pipelines. WebMind is an end-user extension you drive from a chat panel. You can use each.

Running It: Providers and Setup

WebBrain helps native and cloud fashions by way of one interface. Local choices embrace llama.cpp, Ollama, LM Studio, Jan, vLLM, and SGLang. Cloud choices embrace OpenAI, Anthropic Claude, Gemini, Mistral, DeepSeek, and xAI Grok. It additionally helps Groq, MiniMax, Alibaba Cloud (Qwen), Nvidia NIM, and OpenRouter.

A built-in managed possibility, WebMind Cloud, wants no native setup. It prices $5 monthly per gadget profile below a fair-use coverage. For native use, llama.cpp wants no API key.

Starting a neighborhood server takes one command:

Copy Code

# llama.cpp — load at the very least a 16k-token context window
llama-server -m your-model.gguf -c 16384 --port 8080

# Ollama (OpenAI-compatible) — set the extension-origin env var
OLLAMA_ORIGINS="*" ollama serve
# then set the bottom URL to http://localhost:11434/v1 in settings

Point WebMind on the endpoint in settings. For a cross-machine vLLM server, allow CORS with –allowed-origins ‘[“*”]’.

The really useful mannequin is Qwen 3.6 35B (Qwen3.6-35B-A3B). It beat Gemma 4 on the undertaking’s screenshot benchmark. An RTX 5090 is right; an RTX 4090 works with INT4 AutoRound quantization.

Each supplier is a category that extends BaseLLMProvider. It normalizes to at least one response form:

Copy Code

null, utilization: Object

Key Takeaways

WebMind is a free, MIT-licensed AI browser agent for Chrome and Firefox, constructed by Emre Sokullu.
It runs on native fashions (llama.cpp, Ollama; Qwen 3.6 35B really useful) or any cloud API — no web page knowledge leaves your machine when native.
Ask mode reads pages read-only; Act mode clicks and sorts through the Chrome DevTools Protocol for trusted enter occasions.
Security-first by design: begins read-only, approves consequential actions, and makes use of the UI as a substitute of direct API requires mutations.
Free without end self-hosted, or $5/month per gadget profile for the managed WebMind Cloud below honest use.

Interactive Explainer with Demo

Demo-1

</div>
<div>
<h2>WebMind — Interactive Demo</h2>
<p class=”wb-sub”>Pick a activity, select Ask or Act, and watch the agent work.</p>
</div>
<span class=”wb-badge”>Simulated · no actual LLM calls</span>
</div>

<div class=”wb-pane”>
<div class=”wb-panehead”> WebMind
<span class=”wb-mode-pill ask” id=”wb-modepill”>Ask</span>
</div>
<div class=”wb-panel”>
<div class=”wb-prompt”>
<enter id=”wb-input” sort=”textual content” autocomplete=”off”>
<button class=”wb-run” id=”wb-run”>Run</button>
</div>
<div class=”wb-log” id=”wb-log”>
<div class=”wb-hint”>Choose a activity above, then press <b>Run</b>. Read-only Ask mode is the default.</div>
</div>
</div>
</div>
</div>

<div class=”wb-foot”>
<button class=”wb-reset” id=”wb-reset”>Reset demo</button>
<span>Actions listed here are illustrative and secure. The actual extension asks earlier than consequential actions.</span>
<span class=”model”>Built by Marktechpost</span>
</div>
</div>

var order = [‘products’,’summary’,’video’,’signup’];
var state = { mode:’ask’, scen:’merchandise’, operating:false, aborted:false };
var timers = [];

var $ = perform(id){ return doc.getElementById(id); };
perform clearTimers(){ timers.forEach(clearTimeout); timers = []; }

perform renderChips(){
var field = $(‘wb-chips’); field.innerHTML=”;
order.forEach(perform(ok){
var s = SCEN[k];
var b = doc.createElement(‘button’);
b.className = ‘wb-chip’ + (ok===state.scen?’ on’:”);
b.innerHTML = s.label + ‘ <span class=”req”>· ‘+(s.req===’act’?’Act’:’Ask’)+'</span>’;
b.onclick = perform(){ selectScen(ok); };
field.appendChild(b);
});
}

perform renderMode(){
var seg = $(‘wb-modeseg’).querySelectorAll(‘button’);
seg.forEach(perform(b){ b.classList.toggle(‘on’, b.dataset.mode===state.mode); });
var capsule = $(‘wb-modepill’);
capsule.textContent = state.mode===’act’?’Act’:’Ask’;
capsule.className = ‘wb-mode-pill ‘+(state.mode===’act’?’act’:’ask’);
}

perform selectScen(ok){
if(state.operating) return;
state.scen = ok;
var s = SCEN[k];
$(‘wb-url’).textContent = s.url;
$(‘wb-page’).innerHTML = s.web page;
$(‘wb-input’).worth = s.immediate;
if(s.req===’act’ && state.mode!==’act’){ state.mode=’act’; }
renderChips(); renderMode(); resetLog();
}

perform resetLog(){
clearTimers(); state.operating=false; state.aborted=false;
$(‘wb-run’).model.show=”; $(‘wb-run’).disabled=false;
var cease=$(‘wb-stopbtn’); if(cease) cease.take away();
$(‘wb-log’).innerHTML = ‘<div class=”wb-hint”>Press <b>Run</b> to start out. Ask mode is read-only; Act mode can act.</div>’;
var s=SCEN[state.scen];
if(s.req===’signup’||state.scen===’signup’){ /* fields already clean in web page html */ }
}

perform addStep(html){
var d=doc.createElement(‘div’); d.className=’wb-step’; d.innerHTML=html;
$(‘wb-log’).appendChild(d); postHeight(); return d;
}

perform presentStop(){
var cease=doc.createElement(‘button’);
cease.id=’wb-stopbtn’; cease.className=’wb-stop’; cease.textContent=’Stop’;
cease.onclick=perform(){ state.aborted=true; clearTimers(); finishAbort(); };
$(‘wb-run’).model.show=’none’;
$(‘wb-run’).mother or fatherNode.appendChild(cease);
}
perform coverStop(){ var s=$(‘wb-stopbtn’); if(s) s.take away(); $(‘wb-run’).model.show=”; $(‘wb-run’).disabled=false; }

perform finishAbort(){
state.operating=false; coverStop();
addStep(‘<span model=”coloration:#b91c1c;font-weight:700″>&instances;</span> Stopped by person.’);
}

perform askApproval(s){
var field=doc.createElement(‘div’); field.className=’wb-approve’;
field.innerHTML='<p>⚠ Consequential motion — ‘+s.approve+'</p><div class=”row”>’+
‘<button class=”wb-yes”>Approve</button><button class=”wb-no”>Cancel</button></div>’;
$(‘wb-log’).appendChild(field); postHeight();
field.querySelector(‘.wb-yes’).onclick=perform(){ field.take away(); end(s); };
field.querySelector(‘.wb-no’).onclick=perform(){
field.take away(); state.operating=false; coverStop();
addStep(‘<span model=”coloration:#b91c1c;font-weight:700″>&instances;</span> Action cancelled. Nothing was modified.’);
};
}

perform end(s){
state.operating=false; coverStop();
var r=doc.createElement(‘div’); r.className=’wb-result’; r.innerHTML=s.consequence();
$(‘wb-log’).appendChild(r);
addStep(‘<span class=”tick”>&test;</span> Task full · <code>performed</code>’);
postHeight();
if(s.observe){
var hyperlink=doc.getElementById(‘wb-follow’);
if(hyperlink) hyperlink.onclick=perform(e){ e.preventDefault(); runFollow(s); };
}
}

perform runFollow(s){
if(state.operating) return; state.operating=true; $(‘wb-run’).disabled=true; presentStop();
var d=400;
s.observe.steps.forEach(perform(st){
timers.push(setTimeout(perform(){
if(state.aborted) return;
var el=addStep(‘<span class=”spin”></span> ‘+st.t+’ · <code>’+st.c+'</code>’);
timers.push(setTimeout(perform(){
if(state.aborted) return;
el.innerHTML='<span class=”tick”>&test;</span> ‘+st.t+’ · <code>’+st.c+'</code>’;
},380));
},d)); d+=650;
});
timers.push(setTimeout(perform(){
if(state.aborted) return;
state.operating=false; coverStop();
var r=doc.createElement(‘div’); r.className=’wb-result’; r.innerHTML=s.observe.consequence;
$(‘wb-log’).appendChild(r); postHeight();
}, d+150));
}

// mode buttons
$(‘wb-modeseg’).querySelectorAll(‘button’).forEach(perform(b){
b.onclick=perform(){ if(state.operating) return; state.mode=b.dataset.mode; renderMode(); };
});
$(‘wb-run’).onclick=run;
$(‘wb-reset’).onclick=perform(){ state.mode=’ask’; selectScen(‘merchandise’); };
$(‘wb-input’).addEventListener(‘keydown’,perform(e){ if(e.key===’Enter’) run(); });

// auto-resize: element’s personal offsetHeight + 40 (by no means scrollHeight)
perform postHeight(){
attempt{
var el=doc.getElementById(‘wb-root’);
var h=(el?el.offsetHeight:doc.physique.offsetHeight)+40;
if(window.mother or father && window.mother or father!==window){ window.mother or father.postMessage({wbHeight:h},’*’); }
}catch(e){}
}
window.postHeight=postHeight;
window.addEventListener(‘load’,perform(){ postHeight(); setTimeout(postHeight,300); });
if(window.ResizeObserver){
attempt{ new ResizeObserver(postHeight).observe(doc.getElementById(‘wb-root’)); }catch(e){}
}

// init
renderChips(); renderMode(); selectScen(‘merchandise’);
})();
</script>
</physique>
</html>
“>

Demo-2

</div>
<div>
<h2>WebMind for Developers — Interactive Demo</h2>
<p class=”wb-sub”>Read code, extract API specs, analysis sources, and act — from the facet panel.</p>
</div>
<span class=”wb-badge”>Simulated · no actual LLM calls</span>
</div>

perform renderChips(){
var field=$(‘wb-chips’); field.innerHTML=”;
order.forEach(perform(ok){
var s=SCEN[k];
var b=doc.createElement(‘button’);
b.className=’wb-chip’+(ok===state.scen?’ on’:”);
b.innerHTML=s.label+’ <span class=”req”>· ‘+(s.req===’act’?’Act’:’Ask’)+'</span>’;
b.onclick=perform(){ selectScen(ok); };
field.appendChild(b);
});
}
perform renderMode(){
$(‘wb-modeseg’).querySelectorAll(‘button’).forEach(perform(b){ b.classList.toggle(‘on’, b.dataset.mode===state.mode); });
var capsule=$(‘wb-modepill’); capsule.textContent=state.mode===’act’?’Act’:’Ask’;
capsule.className=’wb-mode-pill ‘+(state.mode===’act’?’act’:’ask’);
}
perform selectScen(ok){
if(state.operating) return;
state.scen=ok; var s=SCEN[k];
$(‘wb-url’).textContent=s.url; $(‘wb-page’).innerHTML=s.web page; $(‘wb-input’).worth=s.immediate;
if(s.req===’act’ && state.mode!==’act’){ state.mode=’act’; }
renderChips(); renderMode(); resetLog();
}
perform resetLog(){
clearTimers(); state.operating=false; state.aborted=false;
$(‘wb-run’).model.show=”; $(‘wb-run’).disabled=false;
var cease=$(‘wb-stopbtn’); if(cease) cease.take away();
$(‘wb-log’).innerHTML='<div class=”wb-hint”>Press <b>Run</b> to start out. Ask mode is read-only; Act mode can act.</div>’;
}
perform addStep(html){
var d=doc.createElement(‘div’); d.className=’wb-step’; d.innerHTML=html;
$(‘wb-log’).appendChild(d); postHeight(); return d;
}
perform presentStop(){
var cease=doc.createElement(‘button’);
cease.id=’wb-stopbtn’; cease.className=’wb-stop’; cease.textContent=’Stop’;
cease.onclick=perform(){ state.aborted=true; clearTimers(); finishAbort(); };
$(‘wb-run’).model.show=’none’; $(‘wb-run’).mother or fatherNode.appendChild(cease);
}
perform coverStop(){ var s=$(‘wb-stopbtn’); if(s) s.take away(); $(‘wb-run’).model.show=”; $(‘wb-run’).disabled=false; }
perform finishAbort(){ state.operating=false; coverStop();
addStep(‘<span model=”coloration:#b91c1c;font-weight:700″>&instances;</span> Stopped by person.’); }

perform run(){
if(state.operating) return;
var s=SCEN[state.scen];
if(s.req===’act’ && state.mode!==’act’){
$(‘wb-log’).innerHTML=”;
addStep(‘<span model=”coloration:#4338ca;font-weight:700″>ⓘ</span> ‘+
‘Ask mode is read-only. Switch to <b>Act mode</b> to run this activity.’);
return;
}
state.operating=true; state.aborted=false;
$(‘wb-log’).innerHTML=”; $(‘wb-run’).disabled=true; presentStop();
var delay=350;
s.steps.forEach(perform(st){
timers.push(setTimeout(perform(){
if(state.aborted) return;
var el=addStep(‘<span class=”spin”></span> ‘+st.t+’ · <code>’+st.c+'</code>’);
timers.push(setTimeout(perform(){
if(state.aborted) return;
el.innerHTML='<span class=”tick”>&test;</span> ‘+st.t+’ · <code>’+st.c+'</code>’;
if(st.token) addStep(‘<div class=”wb-token”>Token-conscious: screenshot 2000&instances;1200 ‘+
‘resized & JPEG-compressed → ~380 picture tokens (illustrative).</div>’);
postHeight();
},420));
},delay));
delay+=780;
});
timers.push(setTimeout(perform(){
if(state.aborted) return;
if(s.approve){ askApproval(s); } else { end(s); }
}, delay+150));
}
perform askApproval(s){
var field=doc.createElement(‘div’); field.className=’wb-approve’;
field.innerHTML='<p>⚠ Consequential motion — ‘+s.approve+'</p><div class=”row”>’+
‘<button class=”wb-yes”>Approve</button><button class=”wb-no”>Cancel</button></div>’;
$(‘wb-log’).appendChild(field); postHeight();
field.querySelector(‘.wb-yes’).onclick=perform(){ field.take away(); end(s); };
field.querySelector(‘.wb-no’).onclick=perform(){
field.take away(); state.operating=false; coverStop();
addStep(‘<span model=”coloration:#b91c1c;font-weight:700″>&instances;</span> Action cancelled. Nothing was modified.’);
};
}
perform end(s){
state.operating=false; coverStop();
if(s.finishStep){ addStep(‘<span class=”tick”>&test;</span> ‘+s.finishStep.t+’ · <code>’+s.finishStep.c+'</code>’); }
if(s.onFinish){ attempt{ s.onFinish(); }catch(e){} }
var r=doc.createElement(‘div’); r.className=’wb-result’; r.innerHTML=s.consequence();
$(‘wb-log’).appendChild(r);
addStep(‘<span class=”tick”>&test;</span> Task full · <code>performed</code>’);
postHeight();
if(s.observe){ var hyperlink=doc.getElementById(‘wb-follow’); if(hyperlink) hyperlink.onclick=perform(e){ e.preventDefault(); runFollow(s); }; }
}
perform runFollow(s){
if(state.operating) return; state.operating=true; $(‘wb-run’).disabled=true; presentStop();
var d=400;
s.observe.steps.forEach(perform(st){
timers.push(setTimeout(perform(){
if(state.aborted) return;
var el=addStep(‘<span class=”spin”></span> ‘+st.t+’ · <code>’+st.c+'</code>’);
timers.push(setTimeout(perform(){
if(state.aborted) return;
el.innerHTML='<span class=”tick”>&test;</span> ‘+st.t+’ · <code>’+st.c+'</code>’;
},380));
},d)); d+=650;
});
timers.push(setTimeout(perform(){
if(state.aborted) return;
state.operating=false; coverStop();
var r=doc.createElement(‘div’); r.className=’wb-result’; r.innerHTML=s.observe.consequence;
$(‘wb-log’).appendChild(r); postHeight();
}, d+150));
}

$(‘wb-modeseg’).querySelectorAll(‘button’).forEach(perform(b){
b.onclick=perform(){ if(state.operating) return; state.mode=b.dataset.mode; renderMode(); };
});
$(‘wb-run’).onclick=run;
$(‘wb-reset’).onclick=perform(){ state.mode=’ask’; selectScen(‘clarify’); };
$(‘wb-input’).addEventListener(‘keydown’,perform(e){ if(e.key===’Enter’) run(); });

perform postHeight(){
attempt{
var el=doc.getElementById(‘wb-root’);
var h=(el?el.offsetHeight:doc.physique.offsetHeight)+40;
if(window.mother or father && window.mother or father!==window){ window.mother or father.postMessage({wbHeight:h, ch:CH},’*’); }
}catch(e){}
}
window.postHeight=postHeight;
window.addEventListener(‘load’,perform(){ postHeight(); setTimeout(postHeight,300); });
if(window.ResizeObserver){ attempt{ new ResizeObserver(postHeight).observe(doc.getElementById(‘wb-root’)); }catch(e){} }

renderChips(); renderMode(); selectScen(‘clarify’);
})();
</script>
</physique>
</html>
“>

WebMind is offered on the Chrome Web Store, Firefox Add-ons, and GitHub. Product particulars at webbrain website.

_{Note:Thanks to the Webbrain workforce for the thought management/ Resources for this text. Webbrain workforce has supported this content material/article for promotion.}

The submit Meet WebBrain: An Open-Source, Local-First AI Browser Agent That Reads Pages and Automates Tasks in Chrome and Firefox appeared first on MarkTechPost.

Meet WebBrain: An Open-Source, Local-First AI Browser Agent That Reads Pages and Automates Tasks in Chrome and Firefox

What is WebMind?

Ask Mode, Act Mode, and How Actions Actually Fire

The Security Model

Use Cases, With Concrete Examples

Keeping Token Costs Down

How It Compares

Running It: Providers and Setup

Key Takeaways

Interactive Explainer with Demo

Demo-1

Demo-2

7 must-know frameworks for data engineers in 2026

MIRIX: A Modular Multi-Agent Memory System for Enhanced Long-Term Reasoning and Personalization in LLM-Based Agents

Meta AI Researchers Introduce Matrix: A Ray Native a Decentralized Framework for Multi Agent Synthetic Data Generation

How a Haystack-Powered Multi-Agent System Detects Incidents, Investigates Metrics and Logs, and Produces Production-Grade Incident Reviews End-to-End

Nous Research Releases Token Superposition Training to Speed Up LLM Pre-Training by Up to 2.5x Across 270M to 10B Parameter Models

How CopilotKit Is Redefining the Agentic AI Stack in 2026

Curated by experts. Filtered for relevance.

Resources

About

Subscribe & learn more every day!

What is WebMind?

Ask Mode, Act Mode, and How Actions Actually Fire

The Security Model

Use Cases, With Concrete Examples

Keeping Token Costs Down

How It Compares

Running It: Providers and Setup

Key Takeaways

Interactive Explainer with Demo

Demo-1

Demo-2

Similar Posts

Curated by experts. Filtered for relevance.

Resources

About

Subscribe & learn more every day!