Overview
The chat endpoint allows you to ask natural language questions about your uploaded documents. It uses a RAG (Retrieval-Augmented Generation) system to find relevant information from your documents and generate accurate, context-aware responses.
Authentication
This endpoint requires authentication using Laravel Sanctum. Include your API token in the request.
The chat endpoint only searches through documents that belong to the authenticated user.
How RAG works
The RAG system processes your query through the following steps:
- Query embedding - Your question is converted into a vector embedding using the same embedding model used for document chunks
- Vector search - MongoDB’s vector search finds the 3 most relevant document chunks (from 100 candidates) that match your query
- Context retrieval - The content from matching chunks is assembled into context, separated by delimiters
- LLM generation - Your question and the retrieved context are sent to an LLM (via OpenRouter) to generate a natural language answer
The system uses MongoDB Atlas Vector Search with cosine similarity to find the most relevant document chunks.
Request
Body parameters
The question you want to ask about your documents. Maximum 1000 characters.
Response
The AI-generated answer based on your documents. This is a natural language response that synthesizes information from the retrieved document chunks.
Error responses
The endpoint handles several error scenarios gracefully:
If no relevant documents are found, you’ll receive: "I couldn't find any relevant information in your documents to answer that."
Validation errors
Status code: 422
{
"message": "The query field is required. (and 1 more error)",
"errors": {
"query": [
"The query field is required."
]
}
}
Authentication errors
Status code: 401
{
"message": "Unauthenticated."
}
AI service errors
If the embedding service fails or the LLM API encounters an error, you’ll receive a descriptive error message in the response field:
"I'm sorry, I couldn't process your request at the moment." - Embedding generation failed"Error communicating with AI service." - LLM API request failed"An unexpected error occurred." - Unexpected exception during processing
Examples
curl -X POST https://api.filebright.com/api/chat \
-H "Authorization: Bearer YOUR_TOKEN" \
-H "Content-Type: application/json" \
-H "Accept: application/json" \
-d '{
"query": "What are the main features of the product?"
}'
Response example
{
"response": "Based on your documents, the main features of the product include: advanced document processing with OCR capabilities, real-time collaboration tools, automated workflow management, and comprehensive analytics dashboards. The product also offers integration with popular cloud storage services and supports over 50 file formats."
}
Vector search configuration
The RAG system uses MongoDB Atlas Vector Search with the following configuration:
- Index name:
vector_index
- Vector field:
embedding
- Number of candidates: 100
- Results returned: 3 most relevant chunks
- Filter: User ID (ensures you only search your own documents)
The system retrieves the top 3 most relevant chunks from your documents. These chunks are concatenated with \n\n---\n\n delimiters to provide context to the LLM.