Ephemerals
Disposable cloud dev environments — one isolated VM per Git branch, pre-loaded with an AI coding agent
🔗 Visit Ephemerals
Description
Normally, giving every feature branch its own server would be wasteful and a pain to clean up. Ephemerals makes that cheap and automatic: spin up a branch, get a dedicated machine with an AI coding agent already installed; merge the branch, and the machine disappears on its own.
Ephemerals provisions one dedicated EC2 instance per Git branch, each with zero inbound ports and a coding agent (currently Claude Code, with Codex and Gemini CLI planned) ready to use. Developers reach it through SSH or a browser terminal, preview running apps, and collaborate without sharing infrastructure with other branches. Agent sessions persist in tmux, secrets are encrypted and scoped to the branch's lifecycle, and the whole instance tears down automatically once the pull request merges — no orphaned servers to remember to kill.
💬 Our review
The short version: automatic per-branch isolation plus a pre-installed coding agent is a genuinely useful combination for teams doing a lot of parallel AI-assisted development, but the ecosystem is still narrow.
Compared to GitHub Codespaces or Gitpod, Ephemerals' differentiators are the zero-inbound-ports security model and the fact that an AI agent comes pre-wired rather than something you install yourself — but Codex and Gemini CLI support are still "coming soon," so today it's really a Claude Code environment. Pricing is seat-based ($10/user/month) plus metered compute from $0.18/hr for 2 vCPUs, which is reasonable for short-lived branch work but adds up if branches sit around unmerged. There's also no visible documentation or FAQ, which is a real gap for a dev-tools product. Worth it for teams standardizing on Claude Code who want branch isolation without managing EC2 themselves; wait if you need Codex/Gemini today.
📊 Global score
🤖 AI-enriched data
10 $/utilisateur/mois + compute à l'usage : 2 vCPU 0,18 $/h · 4 vCPU 0,36 $/h · 8 vCPU 0,72 $/h · 16 vCPU 1,44 $/h. Palier gratuit : 30 min/jour sur machine 2 vCPU.
Pros
Isolation par branche : une VM dédiée, zéro port entrant, aucun conflit de ressources
Agent de code IA (Claude Code) préinstallé, prêt à l'emploi
Nettoyage automatique des machines à la fusion de la pull request
Plusieurs tailles de machine (2 à 16 vCPU) selon la charge
Cons
Codex et Gemini CLI encore annoncés "bientôt", seul Claude Code est vraiment disponible
Aucune documentation publique visible (pas de centre d'aide, blog ou FAQ)
Architecture zéro port entrant rend les serveurs classiques ou services longue durée peu pratiques à tester