Product // AI & automation

Live

Prompt Paul

A context-aware browser AI assistant for selected text, pages, media, documents, web grounding, and guarded page actions.

TL;DR Prompt Paul is a Chrome extension by Juan Mackie for context-aware AI work in the browser. It keeps selected text, page content, media, PDFs, and dropped documents beside the current page, with optional web grounding, guarded page actions, custom instructions, provider connections, Hermes and Codex integrations, and recovery controls. It shipped publicly in 2025 with a dedicated product site; adoption is not yet measured.

Year started
2025
Role
Design, development, publishing.
Evidence stage
4/9: Publicly released
Last reviewed
2026-08-30
Problem

Browser AI work is fragmented. A useful task can involve selected text, the page around it, a PDF, an image, a document, a live source, or a controlled page action. Moving that context between tools is the new copy-paste ritual.

Who felt it

Anyone doing repeated reading, writing, research, debugging, or drafting in a browser, starting with me.

Intervention

A Chrome side panel that keeps context beside the page. Prompt Paul can work from selected text, page-aware extraction, media, PDFs, and bounded document text; optional Web grounding adds sources, and optional Act handles approved, constrained page actions.

Role

Design, development, publishing.

Started

2025

Stack

Chrome extension, provider endpoints, Codex subscription, Hermes gateway.

Constraints

Provider and model capabilities vary. Web and Act are opt-in; page actions need approvals and bounded snapshots; raw dropped documents stay local before bounded text is sent; credentials and analytics have separate privacy boundaries.

Outcome

Publicly released through a dedicated product site and the Chrome Web Store. The current public product surface documents page-aware context, media and document handling, Web, Act, Codex, Hermes, Agent Skills, voice input, and recovery controls. Adoption and usage are not yet measured.

Limitations

No published adoption metrics. Provider support varies, and Web and Act remain optional capabilities rather than proof of autonomous browser use.

Next step

Measure which context, grounding, and action workflows people actually keep using.