Waffy - Open-Source AI Browser Extension

TL;DR

TL;DR: Waffy is a free, open-source Chrome extension that drives a browser with AI using your own API keys. It automates multi-step web tasks — clicking, typing, scrolling, extracting data — by following a Plan → Execute → Validate → Output pipeline. No backend, no subscription required.

Source and Accuracy Notes

This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: waffy.io — verified by visiting site on 2026-07-26
• Source repository: github.com/upzare/Waffy — verified README read on 2026-07-26
• License: Apache 2.0 — verified from LICENSE file on main branch
• HN launch thread: news.ycombinator.com/item?id=48901133 — Show HN from 2026-07-26
• Chrome Web Store: chromewebstore.google.com

What Is Waffy?

Waffy is an open-source Chrome extension (also works in Brave, Edge, and other Chromium-based browsers) that acts as an autonomous browser agent. You describe a task in plain English — “Go to example.com, click login, fill in my credentials, and download the report” — and Waffy executes it by controlling the browser directly.

It is built by UPZARE and released under Apache 2.0. The key differentiator from services like OpenAI Operator or Anthropic’s computer use is that Waffy runs entirely inside your browser with no backend server handling inference. You bring your own API keys, and the keys never leave your machine.

The product positions itself as “AUTONOMOUS BROWSER AGENT” — and its feature set backs that up with a four-stage pipeline:

| Stage | Purpose | |---|---| | Planning | Decides whether to automate and drafts a task plan | | Execution | Drives browser actions from screenshots | | Validation | Checks whether the task succeeded | | Output | Summarizes results for you |

Setup Workflow

Step 1: Install the Extension

Option A — Chrome Web Store (recommended):

Install directly from the Chrome Web Store listing.

Option B — Build from source:

git clone https://github.com/upzare/Waffy.git
cd Waffy
bun install
bun run build

Then load the extension in your browser:

1. Open chrome://extensions
2. Enable Developer mode (top-right toggle)
3. Click Load unpacked and select the build/ folder
4. Open Waffy from the toolbar icon, or press Ctrl+I to open the side panel

Step 2: Configure API Keys

1. Right-click the Waffy icon in the toolbar → Options, or open Extension settings from the side panel
2. Go to API Keys and add at least one provider key (OpenAI, Anthropic, Google AI Studio, xAI, Groq, or OpenRouter)
3. Go to Models and assign models to each pipeline stage

For the Execution stage, Waffy recommends a vision model with spatial reasoning — specifically gemini-3.5-flash via Google AI Studio or OpenRouter. The README notes that browser built-in AI is not recommended for this stage.

Step 3: Run Your First Task

Open the Waffy side panel and type a natural language instruction. Examples from the docs:

• “Go to example.com and click the login button”
• “Find all product prices on this page and summarize them”
• “Fill out the contact form with the data from my last email”

Deeper Analysis

Multi-Provider Support

Waffy supports seven provider families:

| Provider | Example Models | |---|---| | OpenAI | GPT-4o, GPT-4.1 | | Anthropic | Claude Sonnet, Claude Opus | | Google AI Studio | Gemini 2.x | | xAI | Grok | | Groq | Llama, Mixtral | | OpenRouter | 100+ models via single key | | Browser built-in AI | Chrome Gemini Nano, Edge Phi Mini |

OpenRouter is particularly useful here — a single OpenRouter API key gives access to 100+ models through one interface, which pairs well with Waffy’s per-stage model selection.

On-Device AI Option

For lightweight stages (Planning, Validation, Output, Title/Step generation), Waffy supports Chrome’s built-in Gemini Nano and Edge’s built-in models. These require no API key and run locally. This is meaningful for cost-conscious users who want to use a paid vision model only for the Execution stage.

Automation Modes

Waffy has two primary modes accessible from the extension UI:

• Chat mode — Ask questions about the current page. Waffy reads the content and answers without changing anything.
• Automate mode — Give Waffy a goal and it takes over: clicking, typing, scrolling, and managing tabs until the task is complete.

The validation step independently checks whether the task succeeded before reporting back, which reduces the risk of the agent silently failing mid-workflow.

Privacy Architecture

Because Waffy runs inference on keys you provide directly to the extension (not a Waffy server), your API usage and the pages you automate are not logged by any Waffy backend. The README states: “Keys stay in your browser.” For users automating sensitive web apps (banking, HR systems, internal tools), this is a meaningful separation from cloud-hosted computer use services.

Practical Evaluation Checklist

• Is this extension actively maintained? Check the GitHub releases page for recent activity.
• Does the vision model you plan to use support screenshot-based UI grounding? The README specifically calls out that the Execution stage needs a vision model with spatial reasoning.
• Have you set per-stage model budgets? Using GPT-4o for every stage is expensive; swap Planning/Validation to a cheap text model like gpt-4o-mini or a Groq-hosted Llama.
• Does your browser support manifest v3 extensions? Waffy requires a Chromium-based browser (Chrome, Brave, Edge, Opera).

Security Notes

• API keys are stored in the extension’s local storage. If you use browser sync or share your Chrome profile, those keys are accessible to whoever has access to that profile.
• Waffy’s automation can execute actions on any website the active tab can reach. Only grant extension access to sites you intend to automate.
• The extension’s open-source nature (Apache 2.0) allows security review of the codebase before trusting it with credentials.
• No backend means no server-side logs of your automation activity — but your API provider (OpenAI, Anthropic, etc.) still logs inference calls made with your key.

FAQ

Q: Is this different from OpenAI Operator or Claude Computer Use? A: Yes. Operator and Claude Computer Use run on remote servers — your browser session is hosted by OpenAI/Anthropic. Waffy runs entirely inside your local Chrome browser, controlling the actual browser you are looking at. You also bring your own API keys instead of paying a separate subscription.

Q: Does it work on Firefox? A: No. Waffy is a Chromium extension and explicitly requires a Chromium-based browser (Chrome, Brave, Edge, Opera). Firefox uses a different extension API.

Q: What does it cost to run? A: The extension itself is free. You pay for your own API usage. The recommended vision model (gemini-3.5-flash) via Google AI Studio is currently $0.075/million input tokens — significantly cheaper than GPT-4o for vision tasks.

Q: Can it handle CAPTCHAs or anti-bot measures? A: No specialized anti-CAPTCHA handling is documented. Like all browser automation tools, sites with aggressive bot detection may interfere with Waffy’s operation.

Q: Is there a self-hosted option? A: No backend is required, so the extension itself is self-contained. There is no separate self-hosted server component. You can load the extension from source without needing any external service.

Conclusion

Waffy fills a specific gap in the browser automation landscape: fully local, open-source, BYOK execution with a clean multi-stage agent pipeline. It is not trying to replace cloud-hosted computer use services — it is aimed at users who want to keep keys and data on their own machine.

For developers comfortable with API key management and looking for a free alternative to browser-based AI agents, Waffy is worth installing. The build-from-source path is straightforward if you have Bun installed, and the Chrome Web Store route takes under two minutes.

If you want to try it, install from the Chrome Web Store and configure an OpenRouter key (which unlocks 100+ models) to get started without creating multiple provider accounts.