oakery.io / woodpecker early access

Woodpecker + Peck

A human-in-the-loop browser agent that automates your web workflows and pays for resources — APIs and data — on your behalf.

unzip, open chrome://extensions, turn on Developer mode, then Load unpacked twice — once for each folder

the side panel at work — real screenshots in use cases below

00Overview

Woodpecker is a browser agent that lives in your side panel. Open it on any page and ask your preferred LLM to:

  • Cut through the UI. Skip the clutter of menus and infinite scroll — describe what you're looking for, and Woodpecker pulls just that from the page.
  • Own your data, everywhere. Save your browsing preferences and the data you gather across sites to your own data vault.
  • Work as your personal assistant. Define a workflow once, and let Woodpecker automate the actions for you.

Peck is a companion extension that runs the user scripts Woodpecker writes, directly in your browser.

01Use cases

Track eBay sold prices without writing a scraper

Say you're a coin collector tracking what certified coins actually sell for.

  • Set up your eBay search filters the way you normally would.
  • Tell Woodpecker which fields you want from each listing — grading service, grade, sold price, delivery fee, sale date, seller.
  • Woodpecker turns the results page into a table you can copy or export as JSON.
  • Run it again anytime and pick up just the new listings since your last run.

From there the data goes wherever you want — a spreadsheet, a database, your own app.

Build your own feed with a "Save" button

You can build your own RSS feed from any site you like. Whether it's X, Reddit, Hacker News, or a blog you follow, ask Woodpecker to add a Save button right on the page. Clicking it saves the post as structured data to Oakery, the personal data vault we provide. Over time, you end up with a reading feed collected from everywhere you care about, stored in one place that belongs to you.

02How it works

Woodpecker runs the agent loop entirely in your side panel, and any action that changes a page waits for your approval unless you enable full access. When a task needs a script, Woodpecker hands it to Peck through a dedicated connector — Peck only accepts requests from Woodpecker, runs the script, and streams results back over a one-way channel.

03Why we're building this

The web experience

Most websites are optimized for their revenue, not your time. With an LLM in your browser, you extract only the information you come for, removing ads and engagement hooks.

Agentic payments

Agents are starting to consume the web through APIs rather than interfaces, and the existing payment model wasn't built for that. With an embedded crypto wallet, the agent signs a payment the moment a service asks for one — no account, no stored credentials. Emerging standards like x402 are already making this real, and we're building Woodpecker for that world from the start.

Own your data

What your assistant learns about you shouldn't live inside one AI company. With Woodpecker, your data stays in a vault you own, and you can use whichever LLM you prefer.

04Installation

We're in early access — install is manual for now, about two minutes. A Chrome Web Store release is on the way.

Step 1 · Load the two extensions

  1. Download both ZIPs above and unzip them.
  2. Open chrome://extensions and switch on Developer mode (top-right corner).
  3. Click Load unpacked twice — once for each unzipped folder. Each folder already has manifest.json at its root, so just select the folder itself.
  4. Open Peck → Details and turn on Allow User Scripts. Chrome requires this for MV3 userscripts — Peck can't run anything without it.

Step 2 · Connect Peck and the Data Vault

Click the Woodpecker icon in Chrome's toolbar to open the side panel. Go to Connectors and click the One-tap presets for Peck and Data Vault.

  • Peck — lets Woodpecker write userscripts and hand them to Peck to run on the page.
  • Data Vault — storage you own for results and preferences, kept across sessions and sites.

Step 3 · Choose how the model gets paid for

Woodpecker needs a model to think with. Open Models → Add LLM:

  • Your ChatGPT subscription — pick the openai-codex provider and sign in; it runs on the plan you already have.
  • Your API key — OpenAI, Anthropic, OpenRouter, or any OpenAI-compatible endpoint. Stored locally.
  • Your Claude subscription — through ClaudeService, below.

Woodpecker credits are coming soon.

Step 4 · Take it for a spin

Open any page, describe what you want, and go.


Optional · Running on a Claude subscription

ClaudeService wraps claude -p in an OpenAI-compatible server on localhost, so a Claude Code subscription works without an API key. Install the Claude Code CLI, run npm start in the repo, then add it under Models → Add LLM with Provider: custom:

  • Base URL — http://127.0.0.1:8787/v1 (note the /v1).
  • Model — sonnet, opus, or haiku; they pass straight through to the CLI.
  • API key — leave it empty unless you set OPENAI_API_KEY on the service.
  • Context window / max output tokens — default low (32,768 / 8,192). Raise them to match the Claude model, or long pages get compacted early.
  • Supports image — turn on for screenshots. Leave reasoning off: thinking effort is set on the ClaudeService side, not per request.