Robertmonkey

Posted on Mar 19

I Built a Free Web Portal for OpenClaw — Self-Hosted AI Workstation on Any VPS

#ai #opensource #vibecoding #showdev

OpenClaw has 300k+ GitHub stars. YouTube is flooded with "install OpenClaw on a VPS" tutorials. Everyone wants their own AI agent — and they should.

But two problems keep tripping people up.

Problem 1: Too Much Friction

Here's what a typical OpenClaw VPS setup looks like:

SSH into your VPS
Install Node.js
Install OpenClaw
Configure the gateway (JSON config, auth tokens)
Install and configure a reverse proxy (Nginx/Caddy)
Provision SSL certificates
Set up a database
???
Still no browser UI — just CLI

Every YouTube video glosses over steps 4–8, and you're left with a terminal-only experience. Great for power users. Not great for everyone else.

Problem 2: Too Much Cost

If you use API keys, Claude and OpenAI bill per-token. Heavy usage can easily hit $200–600/month. If you buy dedicated hardware instead (Mac Mini, gaming PC), you're paying $800–1,200 upfront plus electricity — and you still need API keys on top.

Neither option is great.

The Fix: BridgesLLM Portal

I built BridgesLLM Portal to solve both problems.

Lower the Friction

One command:

curl -fsSL https://bridgesllm.ai/install.sh | sudo bash

That single curl installs:

OpenClaw (AI agent framework)
Agent chat with Claude, Codex, Gemini, and Ollama — all through OpenClaw's gateway
Sandboxed code execution — isolated Docker containers per project
Browser-based remote desktop — full graphical desktop via noVNC
File manager with drag-and-drop uploads
Web terminal — no more SSHing in
Project management with per-project AI agents
SSL/domain wizard — Caddy handles certificates automatically
PostgreSQL database
Self-hosted email server (Stalwart)

After install, you finish setup in your browser. No config files to edit.

Lower the Cost

	VPS + BridgesLLM	Mac Mini + API Keys	Gaming PC + API Keys
Hardware	$5–20/mo VPS	$800 upfront	$1,200 upfront
AI models	~$60/mo (flat-rate OAuth subs)	$100–400/mo (per-token)	$100–400/mo (per-token)
Portal	Free (MIT)	N/A	N/A
Monthly total	~$80–140/mo	$217–517/mo	$285–635/mo

The key insight: Claude, Codex, and Gemini all offer OAuth-based flat-rate subscriptions (~$20/mo each). You connect through openclaw onboard and get unlimited usage at a predictable price. No metering. No surprise bills.

A $20/month VPS replaces $800+ hardware. Flat-rate subscriptions replace per-token billing. The portal itself is free.

Architecture

Browser → Caddy (HTTPS) → BridgesLLM Portal
                              ├── React UI (Vite SPA)
                              ├── Express API (Node.js)
                              │     ├── PostgreSQL
                              │     ├── Docker (sandboxes)
                              │     └── OpenClaw Gateway (WS)
                              │           ├── Claude (Anthropic)
                              │           ├── Codex (OpenAI)
                              │           ├── Gemini (Google)
                              │           └── Ollama (local)
                              └── noVNC (remote desktop)

The portal talks to OpenClaw's gateway over a persistent WebSocket connection. OpenClaw handles all the AI provider routing, tool execution, and agent lifecycle. The portal adds the browser layer — UI, auth, project isolation, sandboxing, file management.

What Makes It Different

This isn't another chat UI. There are plenty of those (Open WebUI, LibreChat, etc.).

BridgesLLM Portal is a workstation. It's what you'd build if you wanted a cloud IDE + AI agent platform + server management tool in one package:

Agent chat isn't just conversations — agents can execute code, read/write files, manage projects
Code sandbox gives each project an isolated Docker environment
Remote desktop streams a full graphical desktop to your browser
Project management lets you create projects with their own file trees, AI agents, and sandboxes
Self-updating — the portal updates itself from the dashboard, no SSH needed

Getting Started

Requirements:

Ubuntu 22.04+ or Debian 12+
2GB RAM minimum (4GB recommended)
Ports 80 and 443 open
A domain name (optional, can use IP initially)

Install:

curl -fsSL https://bridgesllm.ai/install.sh | sudo bash

The installer takes 2–5 minutes. It prints a one-time setup URL when done. Open it in your browser to:

Set your domain and SSL
Create your admin account
Connect AI providers via openclaw onboard
Start chatting

What's Next

The portal is actively developed. Coming soon:

Agent chat session search and history
Dashboard metrics and usage charts
Plugin system for custom agent providers
Multi-user collaboration
Scheduled agent tasks

DEV Community