AI agents are deploying real applications. They shouldn't have to fight dashboards they can't click, SSH sessions they can't manage, or sandboxes that charge $83/month for what any VPS provider gives you for $5.
Today's AI agents — Claude Code, Codex, Gemini CLI — are doing real work: deploying applications, running tests, managing infrastructure. But the tools they're using were never designed for them.
E2B and Daytona are great for ephemeral coding sessions. They weren't built for persistent workloads — and the pricing proves it. Any VPS provider will sell you that compute for $5.
Hetzner is cheap. DigitalOcean has a decent API. But neither was designed for machine operators. Human-readable output, manual SSH setup, no budget controls, no exec primitive.
AWS and GCP are powerful. They're also built to make the bill confusing enough that you stop questioning it — and to make you dependent on infrastructure you'll never own.
Remote execution requires key management. Status requires parsing human-formatted output. Budget enforcement is impossible at the API level. The agent is an afterthought.
Edi is CLI-first cloud compute built for AI agents and the developers who work alongside them. Every command returns JSON. No dashboard. No SSH key management. No fighting the tool.
Every command, every response. Structured errors always include a code, message, and resolution. Agents can parse everything without heuristics.
Servers boot in under a second from ready-to-run templates: Node, Python, Go, Rust, and more. No waiting for package installs mid-agent-loop.
Run a command on any server, get structured output back. No SSH. No key management. Just POST /exec and a JSON response.
Set monthly, daily, and per-server limits in one command. Hard stops — not alerts you'll miss. Hand your agent real infrastructure access and actually sleep at night.
Provision and manage dozens of isolated environments from a single command. Each agent gets its own VM. Declarative state via edi apply.
Pause a VM and pay only for disk. Resume and pick up exactly where the agent left off. Stop paying for compute you're not using.
Remote execution is the single most important feature for agent-native infrastructure. Edi makes it a first-class API call — not an afterthought bolted onto SSH.
Your agent sends a command. Edi runs it inside the VM via vsock, collects the output, and returns structured JSON. Streaming output, working directory, timeout — all configurable. No key rotation, no host verification, no session management.
The entire path: API → gRPC → node agent → vsock → guest agent → process → response. End-to-end in milliseconds.
No other cloud provider offers hard budget enforcement at the API level. Not AWS, not DigitalOcean, not anyone. Edi does — as a core primitive, not a dashboard toggle you'll forget about.
The gap in the market is specific: persistent agent compute at commodity prices. Ephemeral sandboxes are 7-16× more expensive. Traditional VPS providers have no agent-native features.
| Feature | Edi | E2B / Daytona | Fly.io | DigitalOcean | Hetzner |
|---|---|---|---|---|---|
| 1 vCPU / 1GB always-on | $5/mo | $60–83/mo | ~$83/mo | $6/mo | ~$4.50/mo |
| Agent-native API (JSON everywhere) | ✓ Yes | ✓ Yes | Partial | No | No |
| Hard budget controls (API-level) | ✓ Yes | No | No | Dashboard only | No |
| Exec primitive (no SSH) | ✓ Yes | SDK only | No | No | No |
| Always-on persistent VMs | ✓ Yes | No (max 24h) | ✓ Yes | ✓ Yes | ✓ Yes |
| Fleet management | ✓ Yes | No | No | No | No |
| Snapshot / pause / resume | ✓ Yes | No | Partial | Partial | No |
We charge for compute. That's it. No marketplace tax, no platform fee, no percentage of your success. Paused VMs cost $0.10/GB/month — disk only. Free tier included.
Paused VMs: $0.10/GB/month (disk storage only). More sizes available at launch.
The tech industry stopped building for us. We haven't forgotten what that feels like.
We charge for compute. That's it. No marketplace tax, no platform fee, no percentage of your success. The infrastructure is yours — we just run it reliably and cheaply.
A small team. Leased Hetzner bare metal. Real margins. We'll tell you what Edi is not the right tool for — right now, that's GPU compute, SOC 2-required enterprise workloads, and stateless serverless. We'll be honest when that changes too.
Edi is designed to work alongside other tools — Vercel, Cloudflare, your current provider. We're not trying to be everything. We're trying to be the right thing for persistent agent compute. Use us for that. Use others for the rest.
AI is displacing real people. We're not pretending otherwise. A portion of every dollar Edi makes goes toward reskilling resources for people navigating that shift — not as a vague "we give back" gesture, but as a founding principle with a real number attached.
Edi speaks JSON natively. Every response your agent reads will be structured, predictable, and machine-readable — not scraped from human-readable output.
Free tier includes 1 always-on VM. No credit card required to start.