OpenAI Configuration¶

Use OpenAI's GPT models for high-quality documentation search.

Prerequisites¶

OpenAI API account
API key from platform.openai.com

Configuration¶

Create a .env file:

# LLM Configuration
LLM_PROVIDER="openai"
LLM_MODEL="gpt-4o"
LLM_API_KEY="sk-your-api-key-here"

# Embedding Configuration
EMBEDDING_PROVIDER="openai"
EMBEDDING_MODEL="text-embedding-3-small"

Full Configuration¶

# LLM Configuration
LLM_PROVIDER="openai"
LLM_MODEL="gpt-4o"
LLM_API_KEY="sk-your-api-key-here"

# Embedding Configuration
EMBEDDING_PROVIDER="openai"
EMBEDDING_MODEL="text-embedding-3-small"
EMBEDDING_DIMENSIONS="1536"

# Optional: Use Azure OpenAI
# LLM_ENDPOINT="https://your-resource.openai.azure.com/"
# AZURE_OPENAI_API_KEY="your-azure-key"

Verify Setup¶

import asyncio
import pygrad as pg


async def test_openai():
    print("Testing OpenAI configuration...")

    # Index a repository
    await pg.add("https://github.com/encode/httpx")
    print("Repository indexed!")

    # Search
    result = await pg.search(
        "https://github.com/encode/httpx",
        "How to make HTTP requests?"
    )
    print(result)


asyncio.run(test_openai())

Cost Optimization¶

Use Smaller Models¶

For development and testing:

LLM_MODEL="gpt-3.5-turbo"
EMBEDDING_MODEL="text-embedding-3-small"

Batch Processing¶

Index repositories in batches to reduce API calls:

# Pygrad handles batching internally
await pg.add("https://github.com/owner/repo")

Cache Results¶

Pygrad caches indexed repositories, so subsequent searches don't require re-indexing.

Azure OpenAI¶

For Azure OpenAI deployments:

LLM_PROVIDER="azure"
LLM_ENDPOINT="https://your-resource.openai.azure.com/"
LLM_MODEL="your-deployment-name"
AZURE_OPENAI_API_KEY="your-azure-key"
AZURE_OPENAI_API_VERSION="2024-02-01"

EMBEDDING_PROVIDER="azure"
EMBEDDING_MODEL="your-embedding-deployment-name"

Troubleshooting¶

Authentication Error¶

AuthenticationError: Incorrect API key provided

Verify your API key:

# Test with curl
curl https://api.openai.com/v1/models \
  -H "Authorization: Bearer sk-your-key"

Rate Limits¶

If you hit rate limits:

Use a smaller model
Add delays between requests
Upgrade your OpenAI plan

Timeout Errors¶

For large repositories, increase timeout:

# Handle timeouts gracefully
import asyncio

try:
    await asyncio.wait_for(pg.add(url), timeout=300)
except asyncio.TimeoutError:
    print("Processing took too long, try a smaller repository")