Velvet AI Gateway is a powerful tool designed for engineers and developers who are leveraging OpenAI and Anthropic APIs. This innovative product allows you to warehouse every AI request to your PostgreSQL database, providing a comprehensive solution for logging, analyzing, and optimizing your AI usage. With just two lines of code, you can install the proxy and start taking control of your data. Velvet is free to get started, ensuring that you own and manage your data effectively.

How it Works Velvet operates as a seamless gateway between your applications and AI models. It logs every request, including custom metadata, and stores this data in a PostgreSQL database. This enables you to granularly query usage and costs, evaluate models, and generate datasets for further analysis and fine-tuning. The system supports caching and batching, which significantly reduces costs and latency, making it an efficient solution for production environments.

Trusted by Innovative Startups Velvet has garnered trust from various innovative startups who have found it invaluable for their AI operations. For instance, Chirag Mahapatra from an unnamed startup appreciates how Velvet simplifies logging and caching, allowing them to store training sets for fine-tuning their models. Mehdi Djabri highlights Velvet's role in providing a source of truth for interactions between their product copilot and LLMs, aiding in evaluations, cost calculations, and issue resolution. Philip Thomas notes the daily utility of Velvet for monitoring AI features in production, reducing costs through caching, and fine-tuning using logs.

Features that Save Engineering Cycles Velvet offers a range of features that streamline AI operations and save valuable engineering time:

• Warehouse Requests: Log every request with custom metadata to your database.
• Analyze Usage & Costs: Store data as a JSON object for granular querying.
• Cache Requests: Return identical results quickly, reducing costs and latency.
• Batch Jobs: Access extra data for querying each file inside the batch.
• Generate Datasets: Export datasets for batch or fine-tuning workflows.
• AI SQL Editor: Automate SQL queries on large datasets.

Common Use Cases Velvet is versatile and can be used for various purposes, including:

• Analyzing Model Usage: Query logs to understand usage patterns, troubleshoot issues, calculate costs, and evaluate models.
• Optimizing AI Features: Enable caching and batching to reduce costs and latency, and test different prompts and models.
• Preparing Datasets: Generate training sets for fine-tuning and use the batch API for evaluations, classification, or embeddings.
• Switching Between Models: Test and try different models without losing data, reducing platform lock-in by warehousing requests.

Support and Documentation Velvet supports OpenAI and Anthropic endpoints and warehouses requests to PostgreSQL. For custom requirements, the team is open to adding features to their roadmap. Detailed documentation is available, and the team can be contacted for further inquiries or to schedule a call to learn more about how Velvet can benefit your AI operations.

Articles from Velvet Velvet also provides insightful articles that highlight its capabilities, such as logging AI requests, creating fine-tuning datasets, and caching LLM requests to reduce latency and costs. These articles further demonstrate the practical applications and benefits of using Velvet in AI-powered projects.

In summary, Velvet AI Gateway is a comprehensive solution for managing and optimizing AI requests, offering features that are essential for efficient AI operations. Its ease of integration, detailed logging capabilities, and support for caching and batching make it an invaluable tool for any developer or engineer working with AI models.