Chatbot API Wrapper

Exposes a JSON API for a web chatbot for testing with the Mindgard CLI.

Accepts HTTP POST of type application/json to /chatbot in the form {"system_prompt": "{system_prompt}", "prompt":"{prompt}"} where {system_prompt} and {prompt} are placeholders for the system prompt and prompt to use for testing.

Responds with a JSON array with one element containing the chatbot's response. e.g. ["Hello I am an LLM, how can I help you?"]

This can be used as-is with simple chatbots such as HuggingChat, or used as a base to change to suit your webapp, or simply an example of how to expose a suitable API from your existing browser automation to connect the Mindgard CLI.

Building

Requires Java 22+, Maven, and a chromedriver binary in the current working directory.

Build with mvn package

Running

Usage example for HuggingChat

java -jar target/chatbot-api-wrapper-jar-with-dependencies.jar \
 --url "https://example.com/your-chatbot-url" \
 --ready-selector "button.w-full" \
 --input-selector "textarea" \
 --submit-selector "button[type=submit][class~=btn]" \
 --output-selector "div.prose" 

Then interact with

$ curl -vv -XPOST http://localhost:9001/chatbot -d '{"system_prompt":"Please help", "prompt": "Hello LLM, are you there?"}'
["Hello, how can I help?"]

Then point the mindgard CLI at localhost:9001/chatbot . An example compatible mindgard.toml is provided: mindgard test --config mindgard.toml

Options

--parallelism The number of browser/chatbot instances to use at once. Default 5.

--url This is the URL to load in the browser to open the chatbot

--ready-selector A CSS selector of an element to wait for presence of before attempting to interact with the chatbot.

--input-selector A CSS selector for the input to the chatbot, where you would type your messages.

--submit-selector A CSS selector for the submit button to send message to the chatbot.

--output-selector A CSS selector to match the responses from the LLM. The last matching element will be used.

Customisation

This example is flexible enough to support a number of example open source chatbot projects.

For testing custom and more complex applications it is likely you will need to implement your own page interactions in Chatbot.java and rebuild.