Good leads were hiding in pages nobody had time to read
We needed a steady list of people who had just said in public that they needed help, with an email address to reach them. Reading feeds by hand found a few a day, and every email was typed from scratch.
The tools on the market pull in two bad directions. Some scrape hard, which gets accounts restricted. Others send in bulk, which gets a domain blocked. We wanted something that reads like a person, ranks like an analyst and sends only what a person approved.
Read like a person, rank with AI, send on a click
A Chrome extension reads only the pages the user opens, starts when they press Start, and scrolls in small steps with reading pauses. It never posts, likes, follows or messages. It keeps the posts that carry an email address and hands them to a leads page.
The leads page ranks every lead with AI against a criterion the user types, sorts leads into the user's own lists by rules, and drafts one email per lead from a fixed template with two placeholders. Follow-ups are drafted after six days and replies are drafted from the real thread. The AI calls go to any OpenAI-compatible endpoint, so a local model works too.
A small mailer on the user's own machine sends from several Gmail accounts, picks the account with the most room left today, keeps a daily cap per account, spaces sends two to four minutes apart, never emails the same address twice and logs every send. Before a campaign, a deliverability test sends the real template to an inbox the user owns and reports the landing folder plus SPF, DKIM and DMARC.
From a few leads a day to a ranked list
Measured by us while running the tool in our own browser between 30 October and 3 November 2026.
What we would do next
The same pattern fits any team that finds leads or candidates on pages it already reads: recruiters, agencies and B2B sales. The page reader changes per site; the ranking, the drafting and the paced, approved sending stay the same.
This is our own tool, so nothing here is a client commitment. It is where we would take the same build for a client who wants one like it.
Team on this project
The same people who scoped it built it. No handover to a delivery pod.
Related work
Questions about this project
No. It only reads pages the user has opened in their own browser, while they watch. It never posts, reacts, follows, connects or messages, and it makes no calls to a site's API.
No. Every email waits for a person to click Send, and the queue can be paused at any time. That rule is in the mailer itself, not just in the interface.
Any model behind an OpenAI-compatible API. We ran the ranking on Claude through OpenRouter; a local model works by changing the base URL and model name in settings.
Plain-text email, one recipient per message, one attachment at most, daily caps per account, two to four minutes between sends, no tracking pixels and no third-party sending service. The deliverability test runs before a campaign, not after replies dry up.