AgentLLM

Introducing AgentLLM: A Revolutionary Browser-Based Language Model

AgentLLM is an innovative project that demonstrates the potential of open-source large language models (LLMs) to function as autonomous agents directly within web browsers. The primary objective of this project is to highlight how embedded LLMs can efficiently manage complex, goal-oriented tasks typically carried out by autonomous agents.

Key Features and Implementation

AgentLLM is derived from leading research highlighted by WebLLM, which leverages Chromium's cutting-edge WebGPU technology to perform tasks using the GPU. This GPU-based approach significantly improves performance compared to traditional CPU-based methods.

To create a user-friendly and accessible environment, the project builds upon the popular AgentGPT framework by swapping out ChatGPT for WizardLM and modifying the prompt system. Essentially, AgentGPT is equipped to dispatch autonomous agents capable of tackling a wide range of tasks—from simple to complex—by continuously generating and executing tasks through a looping process.

AgentLLM is particularly beneficial because its agents don't rely on external tools, which often adds complexity and unpredictability in more extensive implementations. Additionally, its configurability and rich graphical user interface make AgentLLM a fertile environment to prototype the breaking down of tasks and strategic planning. Try out the demo to explore its potential.

Motivation

Recent advancements in LLMs, such as OpenAI's ChatGPT and GPT-4, have paved the way for autonomous agents capable of independent action without human intervention. However, these LLMs usually require extensive computational resources, limiting their operation to remote and powerful servers, which involves issues related to privacy, cost, and accessibility.

Browser-native LLMs, like AgentLLM, address these concerns by localizing the operations, thereby enhancing privacy and reducing costs while ensuring consistent availability.

Getting Started

AgentLLM provides different options to explore or develop its potential. Here’s how you can get started:

Running the Demo

1. Install Chrome or Edge: Ensure the browser version is at least M113 Chromium.
2. Launch the Browser: For improved performance, launch Chrome with the command: /Applications/Google\ Chrome.app/Contents/MacOS/Google\ Chrome --enable-dawn-features=disable_robustness.
3. Access AgentLLM: Go to AgentLLM.

For the best experience, use a powerful desktop device as running the model locally may be intensive for lower-tier devices.

Docker Setup

AgentLLM can run locally through Docker:

./setup.sh --docker

Alternatively, you can use docker-compose:

./setup.sh --docker-compose

Local Development

To develop AgentLLM locally, follow the steps:

1. Fork the Project: Do this via AgentLLM's GitHub page.
2. Clone the Repository: Copy the source to your local machine.
3. Install Dependencies: Set them up with npm.
4. Environment Configuration: Create and configure a .env file.
5. Database Setup: Use SQLite by modifying the prisma schema.
6. Run the Project: Migrate the database and start the development server with npm run dev.

GitHub Codespaces

Quickly deploy AgentLLM in the cloud using GitHub Codespaces:

1. Open the repository and start a new or existing codespace.
2. Follow the terminal prompts to complete the setup.
3. Launch the application to start browsing.

By embracing this cutting-edge technology, AgentLLM offers exciting possibilities for innovation in the field of autonomous agents. This experimental project paves the way for developing scalable, browser-based LLM solutions capable of handling intricate tasks with minimal computational dependency.