Moltworker: Running a Self-Hosted Personal AI Agent on Cloudflare

Editorial Note: As of January 30, 2026, Moltbot has been renamed to OpenClaw.

Recently, many individuals began acquiring Mac minis to operate Moltbot (previously Clawdbot), an open-source, self-hosted AI agent designed to function as a personal assistant. Moltbot operates in the background on a user’s own hardware, features a substantial and expanding array of integrations for chat applications, AI models, and other popular tools, and can be managed remotely. This AI can assist with finances, social media, and daily organization, all accessible through preferred messaging applications.

However, purchasing new dedicated hardware might not be desirable. An alternative exists to run Moltbot efficiently and securely online. This solution is Moltworker, a middleware Worker and adapted scripts that enable Moltbot to run on Cloudflare’s Sandbox SDK and Developer Platform APIs.

A Personal Assistant on Cloudflare: How It Works

Cloudflare Workers now offers significantly enhanced compatibility with Node.js. Previously, mocking APIs was necessary for certain packages, but these APIs are now natively supported by the Workers Runtime.

This advancement transforms how tools can be developed on Cloudflare Workers. For instance, when Playwright, a widely used framework for web testing and automation running on Browser Rendering, was first implemented, it relied on memfs. This approach was problematic due to its hacky nature and external dependency, leading to deviations from the official Playwright codebase. With improved Node.js compatibility, native node:fs usage became possible, reducing complexity and maintenance, and simplifying upgrades to the latest Playwright versions.

The range of natively supported Node.js APIs continues to expand. A comprehensive overview of this progress is detailed in the blog post “A year of improving Node.js compatibility in Cloudflare Workers.”

Progress in this area is actively measured. A recent experiment involved attempting to run the 1,000 most popular NPM packages within Cloudflare Workers, with surprisingly positive outcomes. Excluding packages that are build tools, CLI tools, or browser-only and thus not applicable, only 15 packages (1.5%) genuinely failed to operate.

A graphic illustrating Node.js API support evolution is provided below:

The results of an internal experiment on npm package support are available here for verification.

While Moltbot does not heavily depend on Workers Node.js compatibility, as much of its code runs in a container, it is important to emphasize the extensive support for packages using native APIs. This capability allows for significant application logic to run directly in Workers when developing new AI agents, bringing processing closer to the user.

Another crucial aspect is the expansion of products and APIs on the Developer Platform. This growth enables the creation and execution of diverse applications, including highly complex and demanding ones, on Cloudflare. Once deployed, applications on this platform immediately benefit from a secure and scalable global network.

These available products and services provided the necessary foundation. Key components include Sandboxes for securely running untrusted code in isolated environments, Browser Rendering for programmatic control of headless browser instances, and R2 for persistent object storage. With these building blocks, the adaptation of Moltbot could commence.

Adapting Moltbot to Run on Cloudflare

Moltbot on Workers, or Moltworker, integrates an entrypoint Worker that functions as an API router and a proxy between Cloudflare APIs and the isolated environment, with both protected by Cloudflare Access. It also offers an administration UI and connects to the Sandbox container where the standard Moltbot Gateway runtime and its integrations operate, utilizing R2 for persistent storage.

Let’s delve deeper.

AI Gateway

Cloudflare AI Gateway serves as a proxy between AI applications and various popular AI providers, offering centralized visibility and control over requests.

Recent updates include support for Bring Your Own Key (BYOK), allowing secrets to be centrally managed and used with gateway configurations, eliminating the need to pass provider secrets in plain text with every request.

For an even simpler approach, Unified Billing removes the need to manage AI provider secrets end-to-end. Users can add credits to their account and use AI Gateway with any supported provider directly, with Cloudflare handling the charges and deducting credits.

To configure Moltbot with AI Gateway, a new gateway instance is created, the Anthropic provider is enabled, and either a Claude key is added or credits are purchased for Unified Billing. The ANTHROPIC_BASE_URL environment variable is then set to the AI Gateway endpoint, requiring no code changes.

Once Moltbot uses AI Gateway, comprehensive visibility into costs, logs, and analytics becomes available, aiding in understanding how the AI agent interacts with providers.

Anthropic is just one option; Moltbot supports other AI providers, as does AI Gateway. A key benefit of AI Gateway is the ability to switch models within the gateway configuration if a superior model emerges from any provider, without needing to alter AI Agent configuration or redeploy. Additionally, model or provider fallbacks can be specified to ensure reliability and handle request failures.

Sandboxes

Anticipating the growing demand for AI agents to execute untrusted code securely in isolated environments, the Sandbox SDK was announced last year. Built upon Cloudflare Containers, this SDK offers a straightforward API for command execution, file management, background process operation, and service exposure, all from Workers applications.

Essentially, the Sandbox SDK provides developer-friendly APIs for secure code execution, abstracting the complexities of container lifecycle, networking, file systems, and process management. This allows developers to concentrate on application logic with minimal TypeScript code. An example is provided:

import { getSandbox } from '@cloudflare/sandbox';
export { Sandbox } from '@cloudflare/sandbox';

export default {
  async fetch(request: Request, env: Env): Promise<Response> {
    const sandbox = getSandbox(env.Sandbox, 'user-123');

    // Create a project structure
    await sandbox.mkdir('/workspace/project/src', { recursive: true });

    // Check node version
    const version = await sandbox.exec('node -v');

    // Run some python code
    const ctx = await sandbox.createCodeContext({ language: 'python' });
    await sandbox.runCode('import math; radius = 5', { context: ctx });
    const result = await sandbox.runCode('math.pi * radius ** 2', { context: ctx });

    return Response.json({ version, result });
  }
};