exe.dev

exe.dev

Remote VM platform purpose-built for overnight-runs with AI coding agents. Provides persistent Linux instances reachable via browser or SSH, with an HTTPS proxy for exposing local services. The primary use case: running claude-code sessions that go for hours without tying up your local machine.

Core Value Prop

Most agentic coding sessions are bottlenecked by two things: the LLM reasoning and your local machine staying on. exe.dev eliminates the second one. Spin up a VM, start a Claude Code session, close your laptop. The agent keeps running.

Secondary value: consistent environment. VM instances have a known state — same OS, same installed tools — so agent sessions don't fail because of local config drift.

HTTPS Proxy

Each VM is accessible at vmname.exe.xyz. Ports 3000–9999 on the VM are reachable at vmname.exe.xyz:PORT. This lets you:

  • Preview web apps running in the VM from your browser without port forwarding
  • Share a running dev server with collaborators during an agent session
  • Test webhook integrations that need a real HTTPS URL (Stripe, GitHub, etc.)

The proxy is documented at https://exe.dev/docs/proxy.md. Use only documented features — internal endpoints are unstable.

Workflow

  1. Create an instance from exe.dev dashboard
  2. SSH in or use the browser terminal
  3. claude to start a Claude Code session
  4. Describe your task and let it run
  5. Check back when done (hook notifications help — see overnight-runs)

Pricing

  • Personal: $20/month — 50 VMs, 100GB disk, 200GB transfer
  • Team: $25/user/month
  • Disposable sandbox VMs are billed per second

Strengths

  • Persistent sessions — agent runs while you sleep
  • Stable HTTPS proxy for dev servers and webhooks
  • Clean environment — no local config interference
  • Browser-accessible terminal as backup to SSH
  • Cost-effective for sporadic large sessions vs. always-on cloud VMs

Weaknesses

  • Not a full CI/CD system — no automated triggers, just on-demand instances
  • HTTPS proxy adds a hop — not ideal for latency-sensitive testing
  • Storage and compute limits depend on plan
  • Not suited for production workloads — this is a dev/agent tool

Related

claude-code · overnight-runs · multi-agent-setup · agentic-workflows

Sources