OpenClaw Brev Launchable

This launchable provides a Brev-oriented bootstrap flow for bringing up OpenClaw on a fresh Ubuntu or Debian-based NVIDIA Brev environment. By default, the launchable deploys to a low-cost CPU instance to optimize costs for long-running agents. OpenClaw is configured to call remote model endpoints, so no local GPU is required.

Launchable Quickstart

Prerequisites

To deploy this launchable, you will need:

• An NVIDIA Brev account
• An NVIDIA Build API key

To generate an API key, visit build.nvidia.com/settings/api-keys. Click the "Generate API Key" button. In the dialog, enter any key name (e.g., "brev-openclaw") and click "Generate Key". Copy the key and store it in a secure location. See Troubleshoot if you run into issues.

Deploy

1. Click the Launchable link to open the one-click deployment page:

   Deploy OpenClaw on Brev

2. Review the Instance Configuration. The Launchable comes pre-configured with the recommended CPU instance and environment settings. Adjust if needed.

3. Click Deploy. Brev will provision a cloud instance, install OpenClaw and code-server, and prepare the environment automatically.

4. Navigate to the Environment Page. Click the "Go to instance page" button, which replaces the "Deploy Launchable" button after deployment starts.

Configure

1. Wait until the environment shows the setup script has completed (you will see a "Completed" status pill near the top of the page).

2. Click the "Code Server" button to finish the setup. This opens the code server in a new tab.

3. A terminal will automatically open at the bottom of the page. The script ./configure.sh from the cloned launch-openclaw repo should run automatically.

4. You will be prompted to add your NVIDIA API key here. Paste it ONCE and press the Enter key. Note: You cannot see the pasted API key. Pasting it twice will result in an invalid key.

5. The script will finish and show:

   OpenClaw Gateway Started
   ========================
   
   URL:
   https://openclaw0-xxxxxxxx.brevlab.com/#token=xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
   
   API Token:
   xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
   
   Hostname:
   brev-xxxxxxxx
   
   Origin:
   https://openclaw0-xxxxxxxx.brevlab.com
   

6. Copy the full URL (looks like https://openclaw0-xxxxxxxx.brevlab.com/#token=xxx...) and open it in a new tab.

7. Click the "Connect" button to open the OpenClaw chat interface. You may need to click this button more than once.

8. Start chatting with the agent.

Troubleshoot

build.nvidia.com API key issues

1. You may need to verify your account before creating an API key.
2. If you see "You do not have permissions to create API Keys in this Organization. Switch Orgs or contact your admin.", click "Switch Org" and follow the instructions to generate a key.

Error: 403 status code (no body) If chatting with the agent results in a 403 error, check the API keys set in .openclaw/.env. Confirm they are correct. If you modify those keys, run ./configure.sh again to update OpenClaw, and copy the link to reopen the OpenClaw UI.

How It Works

The launchable is split into two stages:

• launch.sh performs host bootstrap, installs OpenClaw and code-server, and starts the gateway when configuration already exists.
• configure.sh runs once from an auto-opened code-server terminal in a local clone of this repo, prompts for an NVIDIA API key, performs non-interactive OpenClaw onboarding, and then hands control back to launch.sh.

What It Does

launch.sh:

1. Ensures Node.js 22 or newer is installed.
2. Installs the pinned OpenClaw npm package version 2026.3.13, which corresponds to GitHub release v2026.3.13-1.
3. Verifies the openclaw CLI is available.
4. Clones or refreshes https://github.com/liveaverage/launch-openclaw.git into ~/launch-openclaw by default.
5. Installs code-server, the custom NV Theme, and the fabiospampinato.vscode-terminals extension.
6. Configures code-server to open the cloned repo by default, load its README.md on startup, and auto-open configure.sh from that local clone on first launch.
7. If OpenClaw is already configured, sources ~/.openclaw/.env, starts openclaw gateway, runs a 20-minute device auto-approval loop, and prints connection details.

configure.sh:

1. Prompts for an NVIDIA API key.
2. Runs openclaw onboard --non-interactive --accept-risk.
3. Configures the initial model route against NVIDIA’s OpenAI-compatible endpoint:
   • Base URL: https://integrate.api.nvidia.com/v1
   • Model: nvidia/nemotron-3-super-120b-a12b
4. Stores the key in ~/.openclaw/.env using the env-ref flow supported by OpenClaw.
5. Re-runs launch.sh so the gateway starts immediately after onboarding completes.

Brev Behavior

On hosts named like brev-<env_id>, the launchable derives:

OpenClaw:   https://openclaw0-<env_id>.brevlab.com/chat?session=main
code-server: https://code-server0-<env_id>.brevlab.com

If the hostname does not match the Brev naming pattern, it falls back to:

OpenClaw:   http://localhost:3000/chat?session=main
code-server: http://localhost:13337

Re-run Safety

The bootstrap is designed to be safe to run multiple times:

• It skips Node installation when a compatible version is already installed.
• It keeps OpenClaw pinned to the npm package version resolved from OPENCLAW_RELEASE_TAG and reinstalls if another version is found on PATH.
• It refreshes the local ~/launch-openclaw checkout if it already exists.
• It skips the first-run configure terminal after both ~/.openclaw/.env and ~/.openclaw/openclaw.json exist.
• It reuses a running gateway if a previously started process is still alive.
• It keeps state under ~/.local/state/openclaw-bootstrap/.

Target a GitHub release tag:

OPENCLAW_RELEASE_TAG=v2026.3.24 ./launch.sh

For piped execution:

curl -fsSL https://raw.githubusercontent.com/liveaverage/launch-openclaw/refs/heads/main/launch.sh | OPENCLAW_RELEASE_TAG=v2026.3.24 bash

Override the npm package version directly if needed:

OPENCLAW_VERSION=2026.3.13 ./launch.sh

When OPENCLAW_VERSION is unset, launch.sh now resolves it from OPENCLAW_RELEASE_TAG. It first tries the exact tag without the leading v, then falls back to removing a trailing GitHub-only -N suffix such as v2026.3.13-1 -> 2026.3.13.

Usage

Run the launchable directly:

chmod +x launch.sh configure.sh
./launch.sh

Run it as your normal user, not as root. The scripts use sudo only for package installation and code-server service management.

Output

On the first run, launch.sh prints a pending-configuration message with the code-server URL. After configure.sh completes, the launch flow prints a block like:

OpenClaw Gateway Started
========================

URL:
https://openclaw0-<env_id>.brevlab.com/chat?session=main

API Token:
<token>

Hostname:
brev-<env_id>

Origin:
https://openclaw0-<env_id>.brevlab.com

code-server:
https://code-server0-<env_id>.brevlab.com

Logs and State

Bootstrap logs are written to:

• ~/.local/state/openclaw-bootstrap/gateway.log
• ~/.local/state/openclaw-bootstrap/auto-approve.log

Gateway PID state is tracked in:

• ~/.local/state/openclaw-bootstrap/gateway.pid

Saved first-run credentials are stored in:

• ~/.openclaw/.env

Security Note

OpenClaw agents can execute shell commands, read files, and install software. This launchable is meant for fast bootstrap on development infrastructure, not as a hardened production deployment. Production deployments should isolate agent runtimes, restrict tool permissions, and add approval controls around destructive actions.