OpenClaw and Hermes Agent: Three One-Click Paths to a Self-Hosted AI Agent in 2026

Meet the products

OpenClaw is a personal AI assistant you run on your own devices. The GitHub repo (openclaw/openclaw) carries over 379,000 stars and lists OpenAI, GitHub, NVIDIA, Vercel, Blacksmith, and Convex as sponsors. The actual install on a developer machine is one command: npm install -g openclaw@latest && openclaw onboard. The Gateway is the control plane; the assistant is the product. OpenClaw runs on macOS, Linux, Windows, iOS, and Android with a live Canvas you can speak and listen on. The Cloudflare integration was originally introduced as MoltWorker; it has since consolidated under the OpenClaw name (with the older Moltbot and Clawdbot as historical names).

OpenClaw supports 22 messaging channels: WhatsApp, Telegram, Slack, Discord, Google Chat, Signal, iMessage, IRC, Microsoft Teams, Matrix, Feishu, LINE, Mattermost, Nextcloud Talk, Nostr, Synology Chat, Tlon, Twitch, Zalo, WeChat, QQ, WebChat. There is an active skill marketplace (VoltAgent/awesome-openclaw-skills catalogues 5,400+ skills, with broader compatibility through the open agentskills.io standard).

Hermes Agent is the alternative: a self-improving assistant from Nous Research with around 199,000 GitHub stars. It runs on Telegram, Discord, Slack, WhatsApp, Signal, and the CLI, with voice memo transcription and cross-platform continuity. The differentiator is the learning loop: skills self-improve during use, Honcho dialectic user modelling builds a persistent picture of who you are, and FTS5 session search plus LLM summarisation make past conversations recallable. It plugs into 200+ models (Nous Portal, OpenRouter, NovitaAI, NVIDIA NIM/Nemotron, Xiaomi MiMo, z.ai/GLM, Kimi/Moonshot, MiniMax, Hugging Face, OpenAI) switchable with hermes model. Six terminal backends (local, Docker, SSH, Singularity, Modal, Daytona); Modal and Daytona offer serverless hibernation so the agent costs near-zero when idle.

# OpenClaw (Node 24 recommended, 22.19+ works)
npm install -g openclaw@latest
openclaw onboard --install-daemon

# Hermes Agent (Linux, macOS, WSL2, Termux)
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash

Side by side

Pick by trade-off, not by trend.

Path 1: OpenClaw on AWS Lightsail

Best fit if you live in the AWS ecosystem already. Lightsail's OpenClaw blueprint comes pre-configured with Amazon Bedrock as the default model provider, so the assistant works the moment IAM permissions are set up.

1. Create the instance. Lightsail console → Create instance → Linux/Unix → Blueprint: OpenClaw. The 4 GB plan is the recommended starting size.
2. Wait for Running. The instance comes up in a couple of minutes. Lightsail issues a Let's Encrypt cert for the public IP automatically (no manual TLS setup).
3. Pair your browser. Open the instance, hit the Getting started tab, click Connect using SSH. The MOTD prints a Dashboard URL and an Access Token. Open the URL in a new tab, paste the token, click Connect, approve the CLI and the device pairing.
4. Enable Bedrock. Back on the Getting started tab, click Copy the script under "Enable Amazon Bedrock as your model provider", then Launch CloudShell. Paste, hit Enter. The script creates LightsailRoleFor-<instance-id> with Bedrock + Marketplace permissions. Wait for "Done".
5. First-time Anthropic access. Default model is Claude Sonnet 4.6. If this is your first Anthropic use in your AWS account, complete the FTU form in the Bedrock console (Model catalog → pick Anthropic). One-time per account.
6. Optional: attach a static IP. Networking tab → create + attach. Lightsail re-issues the Let's Encrypt cert against the new IP automatically. You will need to re-pair browsers after this.
7. Optional: connect a channel. SSH back in, run openclaw channels add, pick Telegram or WhatsApp, follow the bot-token or QR flow.

Path 2: OpenClaw or Hermes Agent on Hostinger

Cheapest start, both products are first-class one-click installs in the Hostinger marketplace. Pick OpenClaw if you want the simpler chat-gateway shape, or Hermes Agent if you want the 200+ LLM range and the multi-channel breadth.

1. Pick the plan. Managed OpenClaw is the simplest (currently ₹549/mo intro, ₹999/mo renew). For root access pick the OpenClaw VPS plan (KVM 2 territory, ~₹799/mo intro, ₹1,199/mo renew) or the Hermes Agent VPS tier that matches your workload (KVM 1 to KVM 8).
2. Click Install. hPanel handles the OS, Docker, firewall, TLS, and a free first-year domain. Hostinger advertises a 60-second time-to-first-message, which is roughly the truth on a quiet day.
3. Open the admin UI. Hostinger emails a URL and credentials. OpenClaw and Hermes Agent both ship browser-based control panels.
4. Add messaging channels. Both products do this from the admin UI: Telegram + WhatsApp for OpenClaw; Telegram, Discord, Slack, WhatsApp, Signal, and email for Hermes Agent. Bot tokens or QR pairing depending on the channel.
5. Wire your LLM. Bundled credits cover the first few interactions; for production traffic plug in your own OpenAI, Anthropic, Google, or open-model provider key (Hermes supports 200+ models).
6. Verify backups. Weekly automatic backups are on by default. Trigger a test restore before you trust them.

Path 3: OpenClaw on Cloudflare (formerly MoltWorker)

This is the odd one out: there is no VPS. The agent runs in Cloudflare Workers and Sandbox Containers at the edge, with R2 for files, AI Gateway for model providers, and Cloudflare Access as the auth layer. The project was originally introduced as MoltWorker; it has since consolidated under the OpenClaw name. If you already use the Cloudflare developer platform, this is the cleanest fit.

1. Get a Workers Paid plan. Sandbox Containers require it. ($5/mo as of this writing.)
2. Clone the OpenClaw repo. The Cloudflare project initially shipped as cloudflare/moltworker; check the current OpenClaw docs for the canonical repo, since Cloudflare may have renamed or re-homed it. A safe starting point:

   # original moltworker repo (still installable):
   git clone https://github.com/cloudflare/moltworker openclaw
   cd openclaw

3. Configure AI Gateway. Cloudflare dashboard → AI → AI Gateway → create gateway. Either point it at your own provider key, or enable Unified Billing so all model usage shows up on the Cloudflare invoice.
4. Set Wrangler vars. Wrangler is the Cloudflare CLI. Fill in the variables the README lists (gateway URL, R2 bucket name, provider key reference, Access policy).
5. Deploy.

   npm install
   npx wrangler deploy

6. Front with Access. Add a Zero Trust Access policy (email OTP, Google, GitHub, SSO) so only your authenticated users can reach the Worker. The agent never sits on the public internet without auth.
7. Watch logs and costs. Workers Logs and the AI Gateway analytics dashboard show every request, every model call, and every dollar. This is the unsung win of the Cloudflare path.

Common gotchas across all three

Which one should you actually pick?

• Already in AWS? OpenClaw on Lightsail. Bedrock comes wired, IAM is one CloudShell script, TLS is automatic.
• Cheapest start, want a managed option? OpenClaw managed on Hostinger.
• Need 200+ LLMs and multi-channel breadth (Discord, Slack, Signal)? Hermes Agent on Hostinger.
• Already on the Cloudflare developer platform? OpenClaw on Workers (the project formerly known as MoltWorker). No VPS to patch, Access for auth, AI Gateway for cost visibility.

When to outgrow these

Stay on a one-click install until you hit one of these:

• You need more than one instance for availability or load.
• You need a managed database with point-in-time recovery.
• You need to spread across regions for latency or residency.
• SOC 2 or HIPAA scope arrives.
• You need to plug a custom model, custom planner, or non-trivial tool catalogue into the agent.

At that point migrate to ECS / EKS on AWS, a dedicated Kubernetes cluster on any provider, or a managed agent runtime your engineering team controls end to end. Until then, a one-click install is the right answer.

Want help with the next step?

Kiebot ships agents into production for clients every week. If you want a pair of hands on the architecture, the deploy, or the eval harness, see our Add AI to your product solution. For the production-readiness story, see Designing AI Applications That Survive Production.