@mlc-ai/web-agent-interface
v0.0.3
Published
A tool library for LLM agent in browsers to interact with the web.
Downloads
95
Keywords
Readme
Web-Agent-Interface Library
Overview
The web-agent-interface
library provides tools to LLM agents in browsers to interact with different websites.
Note: The project is still in development phase. It has limited coverage and its APIs may change.
Supported Tools
- General DOM Operations: Get page content and user selection.
- Overleaf: Edit Overleaf documents.
- Google Calendar (GCal): Read and create events on Google Calendar.
Installation
To install and build the library, follow these steps:
Clone the repository:
git clone https://github.com/mlc-ai/web-agent-interface.git cd web-agent-interface
Install dependencies and build the project:
npm install npm run build
Usage
1. Initialize State
import { State } from '@mlc-ai/web-agent-interface';
const state = new State();
2. Import Tools
import { tool, retriever, action } from '@mlc-ai/web-agent-interface';
3. Give Tool Description to Prompt
const system_prompt = `
You are a helpful AI agent.
You have the following tools to use:
${tools.map((t) => JSON.stringify(t.schema)).join(",\n")}
`
3. Call Tool Function to Get Observation
const { tool_name, parameters } = extractToolCall(llm_response);
const observation = tool[tool_name].implementation(state, parameters);
console.log("Got observation:", observation);