Skip to main content

LLM Chatbot

The LLM Chatbot example demonstrates how an ICP smart contract can be used to interact with a large language model (LLM) to generate text. The user can input a prompt, and the smart contract will use the LLM to generate a response. The response is then returned to the user, and the user can submit some follow-up prompts to continue the conversation.

This application's logic is written in Rust, a primary programming language for developing canisters on ICP.

Deploying from ICP Ninja​

When viewing this project in ICP Ninja, you can deploy it directly to the mainnet for free by clicking "Deploy" in the upper right corner. Open this project in ICP Ninja:

Project structure​

The /backend folder contains the Rust smart contract:

  • Cargo.toml, which defines the crate that will form the backend
  • lib.rs, which contains the actual smart contract, and exports its interface

The /frontend folder contains web assets for the application's user interface. The user interface is written using the React framework.

Continue building locally​

To migrate your ICP Ninja project off of the web browser and develop it locally, follow these steps. To open this project in ICP Ninja, click here.

1. Download your project from ICP Ninja using the 'Download files' button on the upper left corner under the pink ninja star icon.​

2. Setting up Ollama​

To be able to test the agent locally, you'll need a server for processing the agent's prompts. For that, we'll use ollama, which is a tool that can download and serve LLMs. See the documentation on the Ollama website to install it. Once it's installed, run:

ollama serve
# Expected to start listening on port 11434

The above command will start the Ollama server, so that it can process requests by the agent. Additionally, and in a separate window, run the following command to download the LLM that will be used by the agent:

ollama run llama3.1:8b

The above command will download an 8B parameter model, which is around 4GiB. Once the command executes and the model is loaded, you can terminate it. You won't need to do this step again.

3. Open the BUILD.md file for further instructions.​