Set up RAG with document search

Upload documents to a Library and query them in a chat completion.

• Create a Library and upload a file
• The model retrieves relevant passages from your documents
• Answers are grounded in your content instead of general knowledge

Time to complete: ~10 minutes

• A Mistral API key (see Get your API key if you don't have one yet)
• Python 3.9+ or Node.js 18+ installed
• The Mistral SDK installed (see Install the SDK if you haven't yet)
• A document to upload (PDF, TXT, or DOCX)

A Library is a container for documents that the model can search during conversations.

import os
from mistralai import Mistral

client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])

# Create a library
library = client.files.upload(
    file=open("company-handbook.pdf", "rb"),
    purpose="retrieval",
)
print(f"File uploaded: {library.id}")

import os
from mistralai import Mistral

client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])

# Create a library
library = client.files.upload(
    file=open("company-handbook.pdf", "rb"),
    purpose="retrieval",
)
print(f"File uploaded: {library.id}")

We process and index the document for retrieval. Check the status before querying.

import time

# Wait for the file to be processed
while True:
    file_info = client.files.retrieve(file_id=library.id)
    if file_info.status == "processed":
        print("File ready for retrieval")
        break
    print(f"Status: {file_info.status}... waiting")
    time.sleep(2)

import time

# Wait for the file to be processed
while True:
    file_info = client.files.retrieve(file_id=library.id)
    if file_info.status == "processed":
        print("File ready for retrieval")
        break
    print(f"Status: {file_info.status}... waiting")
    time.sleep(2)

We typically process small files (under 10 pages) in under 30 seconds. Larger documents may take a few minutes.

Ask a question and include the file reference so the model retrieves relevant passages before answering.

response = client.chat.complete(
    model="mistral-medium-latest",
    messages=[
        {
            "role": "user",
            "content": "What is our company's remote work policy?",
        }
    ],
    documents=[{"type": "file", "id": library.id}],
)

print(response.choices[0].message.content)

response = client.chat.complete(
    model="mistral-medium-latest",
    messages=[
        {
            "role": "user",
            "content": "What is our company's remote work policy?",
        }
    ],
    documents=[{"type": "file", "id": library.id}],
)

print(response.choices[0].message.content)

The response should reference information from your uploaded document instead of general knowledge. For example, if you uploaded a company handbook and asked about remote work policy, you should see specific details like:

"According to the handbook, employees may work remotely up to 3 days per week with manager approval. Remote work requests must be submitted through the HR portal..."

Look for:

• Specific details, names, or policies from your document
• Answers grounded in the uploaded content rather than generic advice
• Reduced hallucination compared to the same question without documents

If the response seems generic, confirm the file status is processed and that the documents parameter is included in your request.