elitan

Heroku's git-push-to-deploy was magic in 2012. Sad to see it go out like this.

I've been building Frost https://github.com/elitan/frost, open source and self-hosted. Same idea, git push to deploy, automatic SSL, custom domains, but on your own VPS. Docker-native, no vendor lock-in, no pricing surprises.

The angle that's a bit different: it's designed for AI coding agents. Simple config they write correctly, clear errors, no K8s complexity to hallucinate. You give your agent the install URL and it sets up the whole server.

Building Frost, a simple self-hosted deployment platform. Docker-only, single-user, no Kubernetes. You point it at a repo or image, it handles the build → health check → deploy → stop old container flow.

Stack: Bun, Next.js, SQLite, Kysely. Been building it almost entirely with Claude Code which has been a surprisingly effective workflow for this kind of infra tooling.

https://github.com/elitan/frost

Been using Claude Code to build a small deployment tool (Frost) for exactly this use case. The meta experience is interesting - using an AI agent to build tooling that makes self-hosting easier.

What I've found: Claude Code is great at the "figure out this docker/nginx/systemd incantation" part but the orchestration layer (health checks, rollbacks, zero-downtime deploys) still benefits from purpose-built tooling. The AI handles the tedious config generation while you focus on the actual workflow.

github.com/elitan/frost if curious

Amazon (AWS) is not cheap! :D

interesting, you have a link?

Plagiarized from what? Happy to address if you can point to what you're referring to.

For those who can't wait for PG18 or need full instance isolation: I built Velo, which does instant branching using ZFS snapshots instead of reflinks.

Works with any PG version today. Each branch is a fully isolated PostgreSQL container with its own port. ~2-5 seconds for a 100GB database.

https://github.com/elitan/velo

Main difference from PG18's approach: you get complete server isolation (useful for testing migrations, different PG configs, etc.) rather than databases sharing one instance.

helpful

This looks really good, love the local LLM support. Will try this instead of Notion which is my current note taking app. Thank you!

It's a killer feature indeed. Glad you liked it! :)

Hey HN! I built Velo to solve a problem I kept running into: testing database migrations and debugging production issues without copying around massive database dumps or waiting for slow clones.

The core idea: use ZFS copy-on-write snapshots to branch PostgreSQL databases the way you branch Git repos. A 100GB database branches in ~2 seconds and takes ~100KB of space initially. Each branch is a complete, isolated PostgreSQL instance running in Docker.

How it works:

- Runs CHECKPOINT to flush PostgreSQL buffers (ensures consistency)

- Creates a ZFS snapshot (filesystem-level, instant)

- Clones the snapshot to a new dataset (CoW means no data copying)

- Spins up a new PostgreSQL container pointing at the clone

The mental model is: Project = Git repo, Branch = Git branch. Commands like `velo branch create api/test-migration` or `velo branch reset api/dev`.

I also added point-in-time recovery using WAL archiving, so you can create branches from specific timestamps (e.g., "5 minutes before the incident").

Requirements: Linux + ZFS + Docker.

Built using Bun and TypeScript.

Let me know what you think :)

Sure is! Thank you :)

Two things related to Postgres:

* Velo - Postgres with instant branching (https://github.com/elitan/velo)

* Terra - Declarative schema management for Postgres (https://github.com/elitan/terra)

Some fun side projects i hack on during the evenings and weekends.

I'm building #4:

> A hybrid of Strong (the lifting app) and ChatGPT where the model has access to my workouts, can suggest improvements, and coach me. I mainly just want to be able to chat with the model knowing it has detailed context for each of my workouts (down to the time in between each set).

here: https://j4.coach/

Still early, have ~30 min per day to work on it but it's usable and improving every week :)

I have the opposite feeling.

i'm fine with it. happy to seed.

have you ever hired humans?

For sure.

I had the same thought when I was dog fooding a CLI tool I've been vibe coding. It's a CLI for deploying Docker apps on any server. An here is the exact PR "I" did.

https://github.com/elitan/lightform/pull/35

One of the advantages of vibe coding CLI tools like this is that it's easy for the AI to debug itself, and it's easy to see where it gets stuck.

And it usually gets stuck because of:

1. Errors 2. Don't know what command to run

So:

1. Errors must be clear and *actionable* like `app_port` ("8080") is a string, expected a number. 2. The command should have simple, actionable and complete help (`--help`) sections for all commands.

Thank you! :)

I hope all our competitors find your answer inspirational!

This is the #1 feature of my Meta Ray-Ban glasses.

I follow the same structure, and I use Notion, Google Calendar, and Flow App.

What's the same rate for humans?

It works well here in Sweden: https://en.wikipedia.org/wiki/Freedom_to_roam

I wouldn't be surprised if Region Blekinge were using something much worse and much more expensive than Whisper for their transcription.

I've been transcribing A LOT of SR (Swedish Radio) shows as part of https://nyheter.sh/, and Whisper (self-hosted) has been very accurate.

Vector search in SQLite is what's keeping me from switching from Postgres.

same

Great take!

Is there a demo available?