This 100% private AI Agent just destroyed Clawdbot

Agent Zero on VPS: A Comprehensive Guide

Key Concepts:

• Agent Zero: A powerful, open-source AI agent capable of autonomous file analysis, video editing, browser automation, and utilizing AI models.
• VPS (Virtual Private Server): A virtualized server providing dedicated resources for running applications 24/7.
• Docker: A platform for developing, shipping, and running applications in isolated containers.
• Docker Compose: A tool for defining and running multi-container Docker applications.
• OpenRouter: An API aggregator providing access to various AI models (e.g., Opus 4.6).
• AOT (Agent Zero Token): A token used for distributed governance and unlocking free inference via the Agent Zero API.
• Venice API: Agent Zero’s own API offering free inference based on staked AOT tokens.
• Knowledge Files: Markdown files used by Agent Zero to store and recall information for future use.
• Secrets Management: A secure system for storing API keys and other sensitive information without exposing them directly to the agent.
• Project Feature: A system within Agent Zero for organizing tasks, instructions, and memories into separate, isolated environments.

1. Introduction & Agent Zero Capabilities

Agent Zero is presented as the most powerful AI agent currently available, with enhanced capabilities when deployed on a VPS. Its core functionalities include autonomous file analysis, code-based video editing, human-like browser interaction, and access to free AI model inference. The agent is open-source, private, and free to use, but requires responsible and ethical application due to its powerful nature.

2. VPS Setup & Docker Installation

The tutorial focuses on setting up Agent Zero on a VPS for 24/7 operation. The initial step involves verifying Docker installation using docker version. If not installed, a curl command (provided below the video) is used to download and execute the installation script. A second sudo command is necessary to run the script. Successful installation is confirmed by running docker version again, resulting in version 29.2.1 in this instance.

3. Docker Compose Configuration

The configuration is managed using docker-compose.yaml. The contents of this file are obtained from a GitHub gist (link provided below the video). Crucially, three modifications are required:

• Login & Password: Replace the default "admin admin" credentials with secure alternatives.
• OpenRouter API Key: An OpenRouter API key is needed to utilize the newly released Opus 4.6 model, which excels in tool calling and terminal usage. The key is generated via the OpenRouter website (top right -> Keys -> Create API key -> agent zero bps) and must not be shared.
• Saving & Exiting: The pasted configuration is saved using Ctrl+O (confirming the filename) and exited with Ctrl+X in the nano editor.

4. Starting Agent Zero with Docker Compose

After configuring docker-compose.yaml, the command docker compose up -d is executed. This pulls the Agent Zero image (several gigabytes in size, containing a full operating system) from Docker and starts the container in detached mode. Docker Compose simplifies stopping and starting the container without losing configuration data.

5. VPS Provider Recommendation: Hostinger

Hostinger is recommended as an affordable and easy-to-use VPS provider. The KVM2 plan is highlighted, offering 2 vCPU cores, 8GB of RAM, and 100GB of disk storage. A 10% discount is available using the code "David" during checkout. The recommended operating system is Ubuntu (latest version) due to its extensive documentation and support.

6. Accessing the Agent Zero Interface

Once the VPS is set up and Agent Zero is running, the interface is accessed via a web browser using the VPS’s IP address followed by port 5080 (e.g., [VPS IP Address]:5080). The login credentials are those defined in the docker-compose.yaml file.

7. Model Configuration: Opus 4.6 & Utility Models

The tutorial demonstrates switching to the newly released Opus 4.6 model. This is done within Agent Zero’s settings (Settings -> Chat Model). A key advantage of Agent Zero is its ability to delegate tasks to different models based on their strengths. For utility tasks, a cheaper model like K 2.5 is recommended to reduce token costs. This contrasts with Cloudbot (OpenClaw), which consumes more tokens due to its lack of model delegation.

8. Secrets Management & OpenRouter Integration

Agent Zero’s secrets management system is highlighted as a secure way to store API keys. Instead of directly exposing the key, the agent recognizes the secret keyword without revealing the actual value. An OpenRouter API key is added as a secret, allowing Agent Zero to access tools like NanoBanana Pro. The process involves:

• Creating a new secret in Settings -> External Services -> Secrets Management.
• Adding the OpenRouter API key.
• Utilizing Python documentation (obtained from OpenRouter’s quick start section) to instruct Agent Zero to create a knowledge file for NanoBanana Pro.

9. Knowledge Files & Memory Management

Agent Zero stores reusable knowledge in markdown files within the /ao/user directory (mapped from the root directory during setup). The agent can be instructed to "remember" information, which is then saved as a knowledge file. The utility model (K 2.5) is used for searching memories, while the main model (Opus 4.6) generates the final response.

10. Generating Images with NanoBanana Pro

The tutorial demonstrates generating an image of a "flying cat over Dubai" using NanoBanana Pro through Agent Zero. The agent utilizes the stored OpenRouter API key (accessed securely via the secrets management system) and the knowledge file created earlier.

11. Project Feature & Context Isolation

Agent Zero’s project feature allows for organizing tasks and maintaining separate contexts. Projects have their own instructions, memory, and file structure. This enables users to switch between different tasks without losing context or interfering with each other.

12. Free Inference with the Agent Zero API (Venice API)

Agent Zero offers free inference through its own API (Venice API) to users who stake AOT tokens. The process involves:

• Connecting a Web3 wallet (e.g., MetaMask) to the Agent Zero website.
• Staking AOT tokens (with optional lock time for increased stake score).
• Generating an API key from the Agent Zero API dashboard.
• Adding the API key to Agent Zero’s settings.

The amount of free inference is determined by the amount of AOT staked. The Venice API does not train on user data and maintains complete privacy.

13. Privacy & Security Emphasis

A core principle of Agent Zero is privacy and security. Unlike competitors like OpenAI and Anthropic, Agent Zero does not train on user data or store chat logs. It is designed to be open-source, runnable locally, and free to use, prioritizing user data protection.

Notable Quotes:

• “Agent Zero is the most powerful AI agent on the market.”
• “Make sure to use it ethically.”
• “Everything is working. The setup is the hardest part.”
• “If you care about privacy and security, you really shouldn't be using anything else other than agent zero.”
• “We will never steal your data and we don't even train models, nothing like that, which is something that obviously every other company does.”

Conclusion:

This tutorial provides a detailed guide to setting up and utilizing Agent Zero on a VPS. The agent’s powerful capabilities, combined with its focus on privacy, security, and open-source principles, position it as a compelling alternative to existing AI solutions. The ability to leverage free inference through AOT staking further enhances its value proposition. The project feature and robust secrets management system provide advanced control and security for users.