Commsies

Turns git commits and design docs into a written summary, a tweet, and a newsfeed post through a self-hosted LLM.

Year
2026
Role
Solo build — pipeline, worker, self-hosted inference
Discipline
AI Tooling

The problem

Shipping work and communicating that work are different jobs, and the second one silently does not happen. Release notes go stale, changelogs get skipped, and by the time anyone writes an update the context has been lost. Handing the whole repository to a hosted model is also not an option when the code is private.

The approach

Commsies takes a structured payload — the commits plus the markdown documents that accompany them — and runs a staged generation pipeline that produces a summary first, then derives a tweet and a newsfeed post from that summary rather than from the raw diff. Collection is deliberately kept on the sender's side, so it can be driven from a git hook, CI, or a custom script, and the service stays a pure transformation. Inference runs against a self-hosted Ollama instance in production, which keeps private code on hardware the author controls; a Cloudflare Workers AI path remains for hosted experiments.

Outcome

Runs as part of the GotchiBot fleet on a self-hosted host alongside Ollama, converting development activity into publishable updates without sending private source to a third-party model.

Details

  • Staged generation — summary first, then tweet and newsfeed derived from it
  • Self-hosted Ollama inference keeps private repositories off third-party models
  • Sender-side collection: drive it from a git hook, CI, or any script
  • Deployable as a Cloudflare Worker or a Node service

Built with

  • JavaScript
  • Node.js
  • Ollama
  • Cloudflare
  • Vercel

Need something like this? AI Tooling & Developer Infrastructure is what I do.

Related work

Last updated .