The problem we're solving
Self-hosting an OpenClaw agent means provisioning a VPS, configuring Docker, setting up a reverse proxy with TLS, managing updates, and hoping nothing breaks at 3am. Most people give up halfway through.
The managed alternatives aren't much better. Most "OpenClaw hosting" providers give you a VPS with a one-click template, you still manage everything. The truly managed options cost $30-60/mo and lock you into their AI models.
We built Chelar to be the middle ground: fully managed, EU-hosted, with bring-your-own-key pricing at $9-19/mo.
What you actually get
When you sign up for Chelar, we provision a dedicated container for your AI agent on our infrastructure. Here's what that looks like:
job "tenant-abc123" {
datacenters = ["dc1"]
type = "service"
group "gateway" {
network {
mode = "bridge"
port "http" { to = 3000 }
}
task "agent" {
driver = "docker"
config {
image = "chelar/agent-runner:latest"
cap_drop = ["ALL"]
readonly_rootfs = true
pids_limit = 256
security_opt = ["no-new-privileges"]
}
resources {
memory = 1024
cpu = 500
}
}
}
}Every tenant gets:
- An isolated container,
cap_drop: ALL, read-only root filesystem, non-root execution (UID 1000), PID limit of 256 - Encrypted storage, API keys and credentials encrypted with AES-256-GCM before they hit the database
- Network isolation, three-layer firewall blocks inter-container communication and access to internal APIs
- Per-tenant data directory, JuiceFS with per-UID permissions (0700) and client-side encryption
- TLS everywhere, Caddy with Let's Encrypt wildcard certificates via Cloudflare DNS challenge
This isn't a shared VPS. Each agent runs in its own container with its own network namespace.
The stack
| Layer | What we use | Why |
|---|---|---|
| Orchestration | Nomad | Simpler than Kubernetes, handles container scheduling and health checks |
| Compute | Hetzner bare metal | EU data residency, predictable pricing, raw performance |
| Ingress | Caddy | Automatic TLS, wildcard certs, dynamic route management |
| Storage | JuiceFS | Shared filesystem with client-side AES-256-GCM encryption |
| Control plane | Go API | Tenant CRUD, orchestration, billing |
| Dashboard | Next.js 15 | Setup wizard, channel management, logs, settings |
Bring your own key
Most AI hosting platforms either lock you to their model or mark up API costs. Chelar uses BYOK, you paste your own API key from Anthropic, OpenAI, Google, Mistral, or OpenRouter. We encrypt it and inject it into your container at startup.
You pay your AI provider directly. We charge only for hosting: $9/mo (Starter) or $19/mo (Pro).
The Starter plan gives you a reactive AI assistant, it responds when you message it via Telegram or the browser. The Pro plan adds always-on mode with cron jobs and scheduled tasks, so your agent can proactively check things, send summaries, or run automations while you sleep.
What works today
We're honest about where we are:
- Telegram integration, connect your bot in the setup wizard, paste the token, start chatting
- Browser chat, built-in web interface for every agent
- BYOK with 5 providers, OpenAI, Anthropic, Google, Mistral, OpenRouter
- Container isolation, every tenant gets a hardened container
- Idle suspension, Starter containers sleep after 72h of inactivity, wake instantly on message. Pro containers stay on 24/7
- EU data residency, Hetzner Germany/Finland, no cross-border transfers
- 14-day free trial, full Pro features, no credit card required to start
What's coming
We're a solo founder project, so we ship one thing at a time:
- WhatsApp, QR pairing flow is built, rolling out post-launch
- Slack and Discord, planned, not yet available
- OpenClaw runtime, new agents default to ZeroClaw (Rust, ~256 MB per container) for efficiency and security. The full OpenClaw runtime (Node.js) is available on the Pro plan upon request
- More integrations, Gmail, Calendar, Notion via MCP servers
Why we're writing a blog
The OpenClaw hosting market has 14+ providers, and most of them publish self-serving "comparison" articles that rank themselves first. We're writing honest, technical content:
- Hosting comparisons with real pricing and trade-offs (including our weaknesses)
- Architecture deep-dives into how we isolate tenants, handle encryption, and manage infrastructure
- Cost breakdowns of what it actually costs to run an AI agent, from $0 to $750/mo
- Setup guides for connecting channels and configuring your agent
Next up: "OpenClaw Hosting Compared", a side-by-side breakdown of every major hosting platform with honest pros and cons.
We're building Chelar on Hetzner bare metal with Nomad orchestration. No VC funding, no content marketing team. Just a solo founder and an honest product.
