The low-code browser extension that started by letting people bend websites to their will - and grew into a platform for guiding and governing what humans and AI agents do on the web.
Most enterprise software asks you to leave the page you're on. Open a new dashboard, log into another portal, switch to yet another tab. PixieBrix, founded in New York in 2020 by Todd Schiller and Michael Mirandi, built its entire company on the opposite instinct: meet people where they already are, inside the web apps they never leave.
The mechanism is a lightweight Chrome and Edge extension. On top of any website, teams assemble what PixieBrix calls "mods" - context-aware modifications built from reusable building blocks the company calls "bricks." A brick might add a button that triggers an automation, make an API call, pull in an AI recommendation, or quietly reshape a cluttered interface into something an agent can actually use.
The original pitch was almost playful. An early viral post described it as "a wild Chrome extension that lets you bend websites to your will." Thousands of people responded. Underneath the novelty was a serious observation: the browser is, for a growing share of knowledge and service work, the operating system - and almost nobody was building tooling for that layer.
That framing has aged well. As PixieBrix moved from consumer curiosity to enterprise product, the browser stayed at the center. Where competitors installed heavy software on the endpoint or asked customers to rip and replace their tools, PixieBrix stayed in the tab. It is, the company argues, why it can deploy "in minutes, not months."
"Agent activity. Real-time enforcement. One platform."
- PixieBrix platform taglinePixieBrix's clearest early market was customer service. Contact centers and business process outsourcers run thousands of agents through browser-based tools - CRMs, ticketing systems, payment consoles - all day long. PixieBrix layered guidance and automation on top: AI-powered decision trees, workflow prompts, automation templates, and assistance drawn from large language models like OpenAI's and Anthropic's Claude.
Then came a sharper realization. In those same browser sessions, three things tend to go invisible, and each one costs money:
Off-policy refunds and discounts issued in the browser, one at a time, that no system flags until the numbers don't add up.
The agent's own actions go unrecorded, leaving compliance teams unable to reconstruct what actually happened on screen.
Sensitive information - PII, payment cards - moving through unmonitored tabs and channels with no guardrail in place.
PixieBrix's answer was to turn its automation layer into an oversight layer. A configurable policy engine can monitor, warn on, or block actions in real time: matching transactions across tabs, masking payment cards, rate-limiting snooping, and detecting data exfiltration or account-takeover patterns - all from the browser, without a heavy endpoint install.
Publicly referenced users span the outsourcing and financial worlds: TaskUs, StarTek, Teleperformance, Samsung, Nubank, IDFC and others, across contact centers, banks and e-commerce.
Low-code Chrome, Edge and Island extension for building "mods" - buttons, automations, API calls and UI changes on top of existing web apps.
Since 2020Configurable rules that monitor, warn on, or block actions - refund integrity, PII masking, exfiltration detection, takeover prevention.
2024Real-time activity logs, screen monitoring, cross-tab transaction matching and anomaly detection for compliance teams.
2024LLM integrations (OpenAI, Claude), AI decision trees and templates that guide agents through workflows in context.
2023Source-available extension that sanitizes pages before AI agents read them - blocking prompt injection, dark patterns and context pollution.
2025A gallery of pre-built mods and automation templates teams can install and customize for their own sites and internal tools.
2021The same core idea - control what happens in the browser - turned out to apply to a very different user: the AI agent. As tools began letting language models click around the web on a person's behalf, PixieBrix noticed the web could fight back. Pages carry hidden instructions, manipulative UI, and noise that derails a model.
In 2025 the company released Agent Browser Shield, a free, source-available Chromium extension that sits between the page and the model. It runs locally as a content script, with no telemetry, and sanitizes the page before the agent ever reads it: stripping page chrome, masking PII, suppressing hidden text, and blocking manipulative elements.
It targets three specific threats - prompt injection, dark patterns, and context pollution - and ships with more than 35 rules. It integrates with agent runtimes such as Browser Use and Browserbase, and as a bonus, trimming page noise cuts the token cost of every agent action.
The strategic logic is familiar for PixieBrix: give away a sharp free tool as an entry point, then offer an enterprise tier where the low-code engine lets teams write custom rules for their own business-specific sites and internal tools.
The next security perimeter isn't the network. It's the page itself.
- The thesis behind Agent Browser ShieldLed by New Enterprise Associates (NEA), with Ridge Ventures participating. Total raised across rounds: approximately $26M.
Todd Schiller and Michael Mirandi start the company to let anyone customize and automate the web.
Seed funding builds out the low-code browser extension and templates gallery.
The team publishes "Web Customization for the Masses," framing its vision for everyday users.
With Ridge Ventures participating, PixieBrix expands its automation platform.
Additional capital fuels AI integrations and enterprise adoption in customer support.
The platform repositions around real-time monitoring, policy enforcement and compliance for customer operations.
A free, source-available tool protects AI browsing agents from prompt injection, dark patterns and context pollution.
PixieBrix competes near several categories without sitting neatly in any of them. RPA platforms like UiPath and Automation Anywhere automate processes but often live outside the browser. Enterprise browsers like Island secure the whole session. Agent-assist tools guide support reps. Low-code builders help people ship apps.
PixieBrix's distinguishing choice is where it operates: inside the existing browser tab, on top of software the customer already runs, added by non-developers without vendor roadmaps or endpoint installs. That location is both the product and the strategy - it is why deployment is fast and why the same tool can serve human agents and AI agents alike.
Its transparency posture is unusual too. The core extension is source-available, so customers and users can inspect the code that is modifying their web pages - a deliberate trust signal in a market full of black boxes.
The business is still small - a lean team of roughly a dozen and modest reported revenue - but the wager is large: that the browser layer is where control over modern work increasingly belongs.
PixieBrix is a browser extension platform that lets teams add automations, AI assistance, and real-time policy enforcement on top of the web apps they already use - originally for support-agent productivity and now for browser-level oversight, compliance, and AI-agent safety.
It was founded in 2020 by Todd Schiller (CEO) and Michael Mirandi, and is headquartered in New York City.
Roughly $26M in total, including a $3.5M seed (2021) and Series A rounds ($5.4M in 2022 and a $6.9M extension in 2023), led by New Enterprise Associates with Ridge Ventures participating.
Primarily contact centers, BPOs, and enterprises in financial services, e-commerce, healthcare, and the public sector - publicly referenced users include TaskUs, Teleperformance, StarTek, Samsung, and Nubank.
A free, source-available browser extension from PixieBrix that sanitizes web pages before AI browsing agents read them, defending against prompt injection, dark patterns, and context pollution while running locally with no telemetry.
Watch & listen: founder interview on the Masters of Automation podcast, and a conversation with Mike Mirandi & Todd Schiller on Formulated Automation.