pods.ml

Spin up an entire agent stack in thirty seconds.

Self-hosted, one-click pods for AI agents. Your box, your data, your keys.

Website · Dashboard · X · Discord

---

What is podsML

Pick a template, hit deploy, get a public HTTPS URL. Reproducible runtime, your data, one box.

$ pods deploy hermes --auto-domain
→ https://juno.pods.ml   (live in ~27s)

The entire platform runs inside your own Azure, Hetzner, or bare-metal box. We never see your traffic, prompts, or model keys.

How it works

Step	What happens
1. Pick a template	Hermes agent, code sandbox, n8n, Minecraft. Each pod has image, resources, ports, and post-deploy surface defined ahead of time.
2. Hit deploy	Provisioned in an isolated container with its own filesystem. Live install progress streams to your browser.
3. Ship the URL	Public HTTPS endpoint in ~27s (p50). In-app terminal, live CPU and memory, swap providers without leaving the dashboard.

Pods available today

Hermes Agent — agent

Always-on AI agent with persistent memory, skills, MCP servers, and 30+ messaging connectors (WhatsApp, Discord, Slack, Telegram). OAuth handoff, streaming terminal, managed mailbox per pod. The flagship.

port 8080 · 1 vCPU · 4 GB RAM · 15 GB disk · https

Code Sandbox — sandbox

Ubuntu pod with VS Code in the browser, Claude Code CLI, or plain shell. Persistent home directory, sudo, public HTTPS URL.

port 8080 · 1 vCPU · 4 GB RAM · 20 GB disk · https

n8n — automation

Visual workflow automation with drag-and-drop nodes, 400+ integrations, queue mode, public webhook URL out of the box. Editor protected by basic auth.

port 5678 · 1 vCPU · 2 GB RAM · 8 GB disk · https

Minecraft (Paper) — game

High-performance Bukkit/Spigot compatible server. Latest stable Paper build. Console, file access, plugin folder, server.properties editor, version pinning.

port 25565 · 2 vCPU · 4 GB RAM · 10 GB disk · tcp

The thesis

"If your agent platform can read your prompts, it's not your agent platform."

1. Self-hosted from byte one. Whole platform runs inside your own infra. We never see traffic, prompts, or keys.
2. No middleman cloud. Pod data lives on your disk. No managed database, no telemetry pipe, no third-party inference relay.
3. One box, many pods. A single 4-vCPU box runs an agent, an automation server, a code IDE, and a game server side by side. Pay your cloud bill, not ours.

Architecture

┌─────────────────────────────────────────────────┐
│  https://*.pods.ml                              │
│  ↓ automatic TLS · per-pod subdomain            │
│                                                 │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐      │
│  │  hermes  │  │  n8n     │  │  code    │  ... │
│  │  :8080   │  │  :5678   │  │  :8080   │      │
│  └──────────┘  └──────────┘  └──────────┘      │
│  isolated runtimes · hard-capped resources     │
│                                                 │
│  your disk · your provider · your key          │
└─────────────────────────────────────────────────┘
                  your Azure / Hetzner / bare-metal

Get started

Sign in, pick a template, and you have a running agent with a public URL before your coffee cools. Free during the prototype.

→ pods.ml/login

Community

• X: @podsml
• Discord: discord.gg/jgC5TFmQ
• Email: hello@pods.ml

---

_{one click · one pod · your data
© pods.ml · 2026 · prototype}