Skip to content

quasa0/churn-to-talk

Repository files navigation

Churn to Talk

Churn to Talk is an agent that continuously finds churned users who were interested in your product and helps founders talk to them with personalized recovery emails.

Churn to Talk command center

Why

"Half the advice I give to startups is some form of 'talk to your customers.'" — Paul Graham

Founders know they should follow up with churned users, but doing it every day means digging through event logs, reconstructing what each person tried, understanding where they got stuck, and writing a non-generic message. Churn to Talk does that work automatically so the founder only has to review and send.

Hackathon

Churn to Talk was built as part of the Nozomio Hackathon, held on May 9, 2026 at the Entrepreneurs First office in San Francisco. We finished in the top 6 out of roughly 51 teams and 200+ participants. Watch our finals presentation.

How It Works

  • Tensorlake runs the churn recovery agent on a 5-minute cron, with a manual trigger for demos.
  • PostHog-style event timelines identify users who showed intent and then dropped off.
  • Nia adds codebase and product workflow context so raw events become a likely churn reason.
  • OpenAI drafts short, founder-style recovery emails for each user.
  • InsForge Postgres stores detected users, run history, drafts, and state.
  • Hyperspell stores recurring churn patterns and product issues as memory.
  • Next.js on Vercel gives the founder a command center to review, edit, save, and send.

About

[Hackathon Project] Agent that finds churned users in PostHog, infers what went wrong using codebase context, and sends personalized founder-style recovery emails. Runs as a scheduled cron on Tensorlake

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors