Skip to main content
The Chat API enables real-time interaction with AI assistants specialized in LaTeX document editing. It streams responses and supports tool calling for file editing operations.

​
Endpoint

POST /api/chat

​
Request parameters

​
messages
array
required
Array of message objects representing the conversation history. Each message should contain:
  • role (string): Either “user” or “assistant”
  • content (string): The message content
​
model
string
The AI model to use for generating responses. Supported values:
  • gpt-4.1-mini (default) - OpenAI GPT-4.1 Mini
  • gemini-2.5-flash - Google Gemini 2.5 Flash
If not specified or an unsupported value is provided, defaults to gpt-4.1-mini.

​
Response

Returns a streaming response with AI-generated content. The maximum streaming duration is 30 seconds. The response uses the AI SDK’s data stream format, which includes:
  • Text content chunks
  • Tool calls (for the editFile tool)
  • Tool results

​
System behavior

The AI assistant operates with a specialized system prompt that defines its role as an expert LaTeX developer. Key behaviors include:
  • Iterative editing: Works back and forth with users on LaTeX documents
  • File editing: Uses the editFile tool to modify LaTeX files
  • User-focused: Prioritizes immediate questions and needs
  • Mathematical expressions: Supports GitHub Flavored Markdown with LaTeX math syntax
  • Language adaptation: Responds in the same language as the user’s message
The system prompt is loaded from docs/system-prompt.md at runtime.

​
Available tools

The assistant has access to the following tool:

​
editFile

Edits the current LaTeX file with new content. Parameters:
  • newFile (string, required): The complete new file content that will replace the current file

​
Example request

curl -X POST https://your-domain.com/api/chat \
  -H "Content-Type: application/json" \
  -d '{
    "messages": [
      {
        "role": "user",
        "content": "Add a table of contents to my document"
      }
    ],
    "model": "gpt-4.1-mini"
  }'

​
Error handling

The endpoint returns standard HTTP error codes:
  • 400 - Bad request (invalid JSON or missing required fields)
  • 500 - Internal server error (model unavailable, system prompt not found)
Errors may occur during streaming and will be included in the stream response.

Build docs developers (and LLMs) love

Get started for freeTalk to us