GitHub - elitan/frost: Vercel experience. VPS pricing.

2 min read Original article ↗

Frost

Vercel experience. VPS pricing.

Open source · Self-hosted

The Problem

AI lets you build apps in hours. Getting them online still takes forever.

  • 47-step deployment guides
  • YAML configs AI hallucinates
  • IAM policies that take hours
  • Cryptic errors, impossible to debug

Frost fixes that.

Install

With AI agent — give your agent: https://frost.build/install.md

Manual — run on your server:

curl -fsSL https://frost.build/install.sh | sudo bash

Need a server? See INSTALL.md for AI-assisted VPS provisioning.

Features

AI-native by design

  • Simple config AI writes perfectly
  • Clear errors, actionable feedback
  • No K8s complexity to hallucinate
  • Just Docker. Predictable.

Everything you need

  • Git push → deployed
  • Automatic SSL (Let's Encrypt)
  • Custom domains
  • GitHub webhooks
  • PR preview environments
  • Instant rollbacks
  • Health checks
  • Resource limits
  • Full REST API

Deploy Anything

Docker-native. If it has a Dockerfile, Frost runs it.

  • Web apps — Next.js, Rails, Django, Go, etc.
  • Databases — Postgres, MySQL, Redis, MongoDB
  • Multi-service projects — frontend, API, workers on shared Docker network
  • Private images — pull from GHCR, Docker Hub, custom registries
  • Long-running jobs — workers, queues, background processes

Stack

  • Next.js + Bun
  • SQLite + Kysely
  • Tailwind + shadcn/ui
  • Docker

Development

bun install
bun run dev
bun run e2e:local
bun run e2e:smoke
bun run e2e:changed
bun run e2e:changed:fast
bun run e2e:profile:week1

Open http://localhost:3000

E2E Speed + Profiling

# Fast local loop (changed groups + retry + report)
bun run e2e:changed:fast

# Full profile artifacts in /tmp/frost-e2e-week1
bun run e2e:profile:week1

# Standardized knobs
E2E_GROUPS='01-basic,04-update,29-mcp' bun run e2e:local
E2E_GROUP_GLOB='group-2*.sh' bun run e2e:local
E2E_BATCH_SIZE=4 E2E_START_STAGGER_SEC=1 bun run e2e:local

Troubleshooting Transient Pull Failures

  • Runtime image pulls automatically retry with backoff (FROST_IMAGE_PULL_RETRIES, FROST_IMAGE_PULL_BACKOFF_MS, FROST_IMAGE_PULL_MAX_BACKOFF_MS).
  • Timeout signatures such as context deadline exceeded, i/o timeout, and proxyconnect tcp are treated as transient infra failures in deployment logs.
  • If a local run flakes on pull, rerun with the managed runner (bun run e2e:local) so pre-pull and retries are applied.

Requirements

  • VPS or server with Docker
  • Ubuntu 20.04+ recommended

License

MIT