v5.0.8 OpenClaw ยท Claude Code ยท Codex ยท Gemini ยท Grok ยท Ollama

Your AI workstation, on your VPS.
Get a full AI workstation.

One command installs OpenClaw + a complete browser-based AI workstation. Agent chat, shared browser, code sandbox, remote desktop, terminal, email, skills marketplace, and automations โ€” connect providers with account sign-in where supported or API keys where preferred. Free and open source.

Choose your install path
$ curl -fsSL https://bridgesllm.ai/install.sh | sudo bash

Ubuntu 22.04+ or Debian 12+ on a compatible host. Installation time depends on downloads, host resources, and enabled features. Review any interruption before retrying.

PS> irm https://raw.githubusercontent.com/BridgesLLM-ai/portal/main/installer/install-windows.ps1 | iex

Runs the public Windows bootstrapper for the experimental WSL 2 local preview. Not a supported production hosting profile.

See what's included Why VPS over hardware?
Works on every device

Your workspace. Any device.

Chat with AI agents, preview live HTML, switch models, and manage projects โ€” from your phone, tablet, or desktop. No app install needed.

โœ“ Touch-optimized chat interface
โœ“ Inline HTML previews & sandboxed renders
โœ“ Model switching on the fly
โœ“ Full desktop experience on tablet
โœ“ Works in modern browsers โ€” no app required
https://your-vps.example.com

What is BridgesLLM Portal?

It turns any cheap server into a full AI workspace you control from your browser. Chat with multiple AI models. Let your AI browse the web while you watch. Run code safely in sandboxes. Manage files, send email from your own domain, install new abilities from a marketplace, and schedule tasks to run while you sleep. The Portal is free; you provide the server, provider access, and any domain or mail configuration.

Lower the friction

Your tools, together.
Setup you can follow.

One installer provisions the workstation and guided setup connects the capabilities your host and provider accounts support.

Current release โ€” 5.0.8

Clearer chats. More reliable recovery.

Portal 5 brings Project Work and broader harness support. Version 5.0.8 fixes duplicate message and tool activity, preserves repeated replies, and adds clearer chat-scoped Tasks progress. Recent patches also improve signed updates, backup verification, and interrupted-maintenance recovery.

Shared Browser

Watch the agent work, live

Your agent drives a real Chrome session while you watch on the same screen. Take the mouse back whenever you want โ€” it is collaboration, not a black box.

Agent Tools

Extend what the agent can reach

Install reviewed skills, wire up MCP tools, and schedule recurring agent work โ€” then inspect the runs instead of treating background actions as fire-and-forget.

01

Answer agents from anywhere

Pending Agent Chat and Project Chat questions appear as persistent, expandable notifications. Choose an option, type a response, send, or deliberately skip โ€” without navigating away from the page you are working on.

Answer in placeDistinct alertRequest ownership
02

Durable long-running chat

Reconnect replay, bounded history, exact-run steering, cancellation, and stable tool and thought timelines make long turns usable after ordinary browser disruption.

Reconnect replayExact-run steering
03

Project-bound execution

Compatible hosts get non-root, project-bound runtimes with one writable workspace, live attestation, and controlled public egress. Providers qualify before they can run there.

Non-rootAttestedScoped egress
04

Apps and project data survive maintenance

Backups include project and standalone app sources. Update and restore paths preserve data, reconcile running-app intent, and keep a last known-good deployment available when promotion fails.

Transactional deploySource recovery
05

Updates prove more before claiming success

The Owner sees signed release details and starts the update. Portal checks backup freshness, serializes maintenance, applies migrations, verifies service and runtime health, and rolls back a candidate that cannot become healthy.

Signed artifactsBackup preflightRollback
06

Setup that starts with ownership

Fresh installs create a secure Owner first. Readiness cards then guide optional domain, mail, provider, local model, Remote Desktop, and other host capabilities.

Owner firstReadiness cards
Install now โ€” it's free See setup guide

Allow time for package downloads and runtime provisioning. Start with at least 3.5 GB RAM and 35 GB disk; 4 GB+ RAM is recommended.

Lower the cost

Self-hosted AI on a cheap VPS.
No hardware. Flexible provider costs.

The portal is free. Your monthly cost is your VPS plus whatever model access you choose โ€” subscription-style sign-in where supported, API keys where needed, or local Ollama.

Best-value VPS + BridgesLLM Portal

VPS (e.g. Hetzner, DigitalOcean)$20โ€“40/mo
Portal softwareFree
Codex / GeminiSubscription-style sign-in available
ClaudeAccount-dependent pricing
API-key providersUsage-based if you choose them
Local OllamaOptional
Hardware upfront$0
Monthly total
Depends on your provider mix and usage
Flexible

Mac Mini M4 + API Keys

Mac Mini M4 (16GB)$800 upfront
Electricity (~150W average)~$13/mo
Claude API (heavy use)$50โ€“200/mo
OpenAI API (heavy use)$50โ€“150/mo
Amortized hardware cost~$67/mo
12-month total
$2,600โ€“$6,200+
$217โ€“517/mo

LIVE Gaming PC + API Keys

Mid-range gaming PC$1,200 upfront
Electricity (~400W average)~$35/mo
Claude API (heavy use)$50โ€“200/mo
OpenAI API (heavy use)$50โ€“150/mo
Amortized hardware cost~$100/mo
12-month total
$3,420โ€“$7,620+
$285โ€“635/mo

Cloud IDEs (Codespaces etc.)

GitHub Codespaces (4-core)~$0.36/hr
8h/day, 20 days/mo~$58/mo
No persistent environmentโ€”
Limited AI agent supportโ€”
12-month total
$696+ (no local agents)
$58+/mo

* Provider costs depend on your account, harness, and access method. Subscription sign-in, metered APIs, and local compute have different limits and charges; no flat-rate provider access is included.

Feature comparison

What you actually get.

What Portal provides, and what you still bring.

CapabilityPortalRequirement or boundary
Agent Chat and Project WorkBrowser-based conversations and workspacesAvailable harness, supported host, and provider access
Shared browser and desktopChrome and noVNC integrationProvisioned desktop; separate provider sign-ins where needed
Project isolationScoped containers and controlled egressCompatible Docker, kernel, and AppArmor; operator chat is separate
Mail and public sharingOptional domain mail and scoped share linksDomain, DNS, TLS, and deliberate public exposure
UpdatesOwner-started signed Portal updatesCompatible AI tools use separate explicit maintenance
BackupsPortal data; optional agent reference exportsNo mail, provider logins, runtime databases, or whole-server restore
Costs and privacyMIT-licensed, self-hosted workstationVPS/provider costs and external AI traffic remain yours
Built for trust

Your server. Your data. Your control.

Portal data is self-hosted. External AI requests leave your server; optional operational telemetry, update checks, and installer events are explained in the privacy notice.

HTTPS everywhere

Automatic SSL certificates via Let's Encrypt. HSTS headers, CSP policies, and strict security headers out of the box.

Sandboxed execution

Project Chat uses project-scoped containers and controlled egress on supported hosts. Main Agent Chat and Terminal are separate operator-level surfaces, not tenant sandboxes.

Role-based access

Owner, admin, user, and viewer roles. New accounts require approval. Disabled or banned accounts are locked out immediately.

Path traversal protection

Dedicated middleware blocks directory traversal, symlink escapes, and access to system paths. Repeated violations are tracked and escalated.

Firewall by default

The VPS profile configures UFW. Internal APIs, browser control, and database services stay private; configured public mail protocols and share links have their own exposure.

Malware scanning

Uploaded files are scanned with ClamAV before storage. Detected threats are rejected automatically.

Full details in our security policy โ€” including known limitations and hardening recommendations.

Powered by OpenClaw

Bring your own providers.
Choose the right harness.

BridgesLLM Portal runs on OpenClaw โ€” the open-source agent framework for running agent workflows on infrastructure you control. Connect Claude, Codex, Gemini, Grok, or local Ollama models through account sign-in, setup-token, API keys, or local-model paths depending on the provider. Predictable infrastructure cost, clearer ownership, and a full browser workspace.

OpenClaw logo OpenClaw (agent framework)
Claude Code logo Claude Code (Anthropic)
Codex logo Codex (OpenAI)
Gemini logo Gemini (Google)
Grok logo Grok (xAI) New
Agent Zero logo Agent Zero New
Ollama logo Ollama (local models)

Provider availability depends on your account and the connection path you choose. OpenClaw, Claude, Codex, Gemini, Grok, Agent Zero, and Ollama names and logos are trademarks of their respective owners and are shown for identification only.

MultiHarness support
1Install command
Your VPSWorkstation hosting
MITOpen source
Guided installation

From a supported VPS
to your AI workstation.

Get a compatible VPS, run the installer, and follow setup. Some host prerequisites and recovery operations require administrator access.

1

Get a VPS

Hetzner, DigitalOcean, Vultr, Linode โ€” any supported Ubuntu or Debian VPS. 3.5GB+ RAM, 35GB disk, typically $20โ€“40/mo.

2

Run the installer

One installer command provisions the stack and prints the setup URL. Timing varies with the host and downloads.

3

Setup wizard

Domain, SSL, admin account โ€” all in the browser. No config files to touch.

4

Connect AI & go

The setup wizard connects your AI providers โ€” paste an API key or sign in with OAuth. Start chatting immediately.

Detailed walkthrough

Step-by-step instructions

Everything you need for each step, including commands and tips.

Watch: Full install + setup walkthrough

1

Run the installer

SSH into your VPS and paste the install command. The script installs all dependencies, sets up the database, configures SSL, and starts the portal.

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

Choose Install on a new host or Update / Repair on an existing installation. Follow the printed setup URL to create the Owner in the browser.

Tip: If installation is interrupted, inspect its recorded status and error first. Never run a second installer over an active transaction or delete its recovery records.
2

Complete the setup wizard

Open the portal URL in your browser. The setup wizard walks you through domain configuration, SSL certificates, and your admin account. Everything is handled in the browser โ€” no config files to edit.

3

Connect your AI providers

The setup wizard walks you through connecting AI providers. Depending on the provider, you can use:

  • Account sign-in โ€” available for Codex and Gemini. Sign in with your existing account instead of pasting an API key.
  • Guided token or API key setup โ€” Claude uses a guided setup-token flow, and key-based providers use API keys from the provider dashboard.
  • Local models โ€” connect Ollama if you want a self-hosted model path.

You can add, change, or remove providers any time from the Settings page in the portal.

Alternative: If you prefer the command line, you can also run openclaw onboard in the portal's Terminal to configure providers manually.
4

Test agent chat

Open Agent Chat, choose an available harness/model, and confirm a real reply. Readiness indicators describe checks; they do not guarantee account entitlement or a successful turn.

That's it โ€” your AI workstation is live. Start chatting, create projects, or explore the file manager and remote desktop.

5

Optional: Give your agent portal awareness

Copy the prompt below and paste it into your OpenClaw agent's workspace file (e.g. AGENTS.md or system prompt). This teaches the agent how your portal works โ€” where files live, how projects are structured, and how to interact with users through the portal interface.

โ†“ Scroll down to the "Agent Portal Prompt" section to copy it.

Troubleshooting

Common issues and quick fixes

Agent chat cannot connect

Check the selected harness readiness and recorded error first. A gateway startup, missing login, incompatible plugin, or interrupted maintenance needs its own remedy. Inspect services without restarting an active conversation:

systemctl status openclaw-gateway bridgesllm-product --no-pager

SSL certificate not working / domain not resolving

Make sure your domain's DNS A record points to your VPS IP. After DNS propagates, re-run the installer to pick up the domain and provision SSL:

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

The installer detects the existing installation and only updates what's needed.

Update to the latest version

Use the Owner-only Update action on the Dashboard after checking backup recoverability. See the update guide. The installer also offers an explicit Update / Repair menu:

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

Check service status

If something isn't working, check that both services are running:

systemctl status openclaw-gateway bridgesllm-product

View logs

Portal and gateway logs are available via journalctl:

# Portal backend logs journalctl -u bridgesllm-product -n 50 --no-pager # OpenClaw gateway logs journalctl -u openclaw-gateway -n 50 --no-pager

Change AI providers after setup

Go to Settings in the portal sidebar. You can add, remove, or switch providers (OAuth or API key) at any time. Alternatively, run from the terminal:

openclaw onboard
For your AI agent

Portal-aware agent prompt

Paste this into your OpenClaw agent's workspace instructions to teach it how your portal works. Your agent will know where to find files, how projects are structured, and how to help users through the interface.

Click to expand full prompt โ–ผ
System requirements

Runs on standard Ubuntu and Debian VPSes.

Use a supported box with enough RAM and disk. Smaller instances may start, but this is the supported baseline.

โœ“ Ubuntu 22.04+ or Debian 12+
โœ“ 3.5GB RAM minimum (4GB+ recommended)
โœ“ 35GB free disk space
โœ“ 2+ CPUs recommended
โœ“ Root or sudo access
โœ“ Ports 80 and 443 open
โœ“ Any modern VPS provider
Frequently asked questions

Everything you need to know.

Can I install BridgesLLM Portal on a VPS that already has OpenClaw? +

Yes. The installer detects an existing OpenClaw installation and uses it instead of requiring a clean server. If OpenClaw is not already installed, the installer installs it for you.

If I click Update in the portal, does it also update OpenClaw? +

No. Portal Update preserves OpenClaw core and native runtime versions. A narrow question-plugin repair on supported OpenClaw 9.x may briefly restart the gateway. Update Compatible AI Tools is a separate, explicit maintenance action with its own Node prerequisites.

What size VPS do I need? +

Use Ubuntu 22.04+ or Debian 12+, 3.5 GB RAM minimum (4 GB+ recommended), 35 GB free disk space, root or sudo access, and ideally 2+ CPUs. Smaller instances may boot, but that is the supported baseline.

What does the installer set up automatically? +

The installer sets up the core portal stack for you: the portal app, OpenClaw, PostgreSQL, Caddy, and the main system services. The browser setup flow then handles guided configuration such as your admin account, provider connection, and domain/SSL steps.

Do I need API keys, or can I use subscriptions? +

It depends on the provider and harness. Supported account sign-in, API credentials, and local Ollama each have their own access requirements. Availability and charges follow your provider account; Portal does not require one universal Claude billing arrangement or grant model access.

What do I actually pay for? +

BridgesLLM Portal itself is free and MIT-licensed. You pay for the VPS, optional domain/services, and whichever model access you use: subscription access where supported, metered APIs, or local compute.

Does my data stay on my VPS? +

Yes. Your portal data, projects, files, and local services stay on your server. If you connect external AI providers, requests for those models still go to the provider you chose, but the portal itself runs on infrastructure you control.

What's the difference between BridgesLLM Portal and OpenClaw? +

OpenClaw is the underlying agent/runtime layer. BridgesLLM Portal is the browser-based workstation built on top of it, adding agent chat, shared browser, code sandbox, remote desktop, terminal, files, and guided setup.

OpenClaw + full workstation.
One command away.

Free to install. Use the provider access method you prefer โ€” OAuth where supported or API keys where needed. You pay for the VPS and any model/provider access you choose.

Install command
$ curl -fsSL https://bridgesllm.ai/install.sh | sudo bash
Privacy Policy Terms of Use Contact Support

Questions before you install? Email support@bridgesllm.com or open an issue on GitHub.