Vertra Cloud does not host a model. You bring the key of a provider and pay for the model usage to that provider. This template is a Vertra Cloud packaging and is not affiliated with the OpenClaw project (MIT license).
Requirements and plan
1. Get the provider key and choose the model
OpenClaw works with OpenAI, Anthropic, Google Gemini, OpenRouter or an OpenAI-compatible endpoint, and dozens more from the panel (Settings > Models). Create a key in your provider’s account and copy it: most providers show it only once. In the wizard you give the key and, optionally, the model. Empty means the provider’s default. With a compatible endpoint the base URL (ending in/v1) and the model are required. On OpenRouter any model id works, including :free ones. OpenClaw starts even with an invalid key; the key is only used when the assistant answers, so test it in the first chat.
2. Create the assistant
1
Open the template
In the dashboard open Templates, pick OpenClaw and click Use template.
2
Provider, key and model
Choose the provider, paste the key and, if you want, type the model.
3
Browser: yes or no
Can the assistant use a browser? Yes (default) adds Chromium to the app and needs 3 GB of memory at least. No gives a text-only assistant with 2 GB. It applies on the first start.
4
Channels (optional)
Telegram, Discord and Slack each ask for the token and who may talk to the bot. Without an allowed list the channel is not enabled. See step 4.
5
Create
Name the app and click Create. The outputs show the app address and the gateway token, shown once. Copy the token.
3. First access: the gateway token
1
Open the address
Open the app address. The panel (Control UI) asks for the token.
2
Paste the token
In Settings > Gateway paste the gateway token from the outputs. The token also lives in the encrypted variable
OPENCLAW_GATEWAY_TOKEN.3
Wait for the pairing
Wait up to 15 seconds: the first browser you use is approved automatically. After that, talk to the assistant in the panel.
Approve a second browser or device
Only the first device is approved by itself. When you open the panel from another browser or computer and paste the token, the device stays waiting for approval. From a device that is already connected, open Settings > Devices and approve the new one.Automatic pairing lowers “token plus approval” to “token” for that first device only. If you prefer to approve every device, set
OPENCLAW_BOOTSTRAP_PAIRING to off in the app variables before the first access.4. Connect a channel
You can set channels in the wizard or later in the panel.- Telegram
- Discord
- Slack
- WhatsApp
- In Telegram, talk to @BotFather, send
/newbotand copy the bot token. - Find your numeric id with @userinfobot.
- Enter the token and your id.
5. Use the browser
Ask for something that needs the web, such as “open example.com and tell me the title”. In Chat > + > Browser you see the window the agent is using.- The agent browser blocks private and local networks, on purpose.
- It runs without Chromium’s own sandbox: the container is the isolation.
- It needs memory. With several tabs Chromium goes past 2 GB, so a plan with room matters.
The panel terminal
The panel has a terminal that opens a shell in the app’s own container, with no extra privilege. Use it to look around, but remember that the container’s file system outside/app is read-only.
Cron, skills and voice
Use Settings > Cron for scheduled tasks. The ready skills appear in Skills. Voice needs your browser’s microphone and a speech provider.Change provider or model later
Use Settings > Models in the panel. The variablesOPENCLAW_PROVIDER and OPENCLAW_MODEL_ID also follow the wizard on every start. The file /app/.openclaw/openclaw.json is generated once; after that it belongs to you and the panel rewrites it. The gateway token, the public origin and the model follow the app variables.
The cost of the AI is yours
The plan charges for the memory. Your provider bills the model calls. Browser tasks and scheduled jobs can use many tokens, so set a spending limit in your provider’s account.Update and roll back
New versions arrive as new versions of the template. A snapshot is taken before the update, and rolling back brings the data along, because OpenClaw migrates its local schema when it starts a new version. See Update and roll back. Zero-downtime deploy is off: two copies on the same disk cannot coexist.Common problems
The panel does not connect
The panel does not connect
Check the token in Settings > Gateway, and that the address you opened is the app’s own address (
PUBLIC_ORIGIN).A second browser stays waiting
A second browser stays waiting
Approve it in Settings > Devices from a device that is already connected.
Timeouts in the browser tools
Timeouts in the browser tools
Not enough memory: Chromium passes 2 GB with several tabs. Raise the app’s memory or use fewer tabs.
Discord or Slack take a long time to come up
Discord or Slack take a long time to come up
Normal on the first boot (npm plugins). If it fails, check the token and the access to npm in the app log.
OPENCLAW_MODEL_ID is required
OPENCLAW_MODEL_ID is required
With a compatible endpoint the model is required. Set
OPENCLAW_MODEL_ID and restart.Related
OpenClaw in the catalog
What the template creates and its variables.
Update and roll back
The safety snapshot and rollback with data.
Public page
The template in the public gallery.
Use a template
The gallery, the wizard and the outputs.