Hacker News comment for the launch of our AI gateway. Will go along with Product Hunt launch. Goal is to drive people to our demo project so they can see how it works.
—
Show HN: Warehouse OpenAI requests in your own DB
Today we’re launching Velvet, an AI gateway for warehousing OpenAI and Anthropic requests to your PostgreSQL instance. Our first product was an AI SQL editor. It worked really well, but we had no data around what happened between our app and OpenAI. We started warehousing every request, giving us control over the development process.
We didn’t think of the warehousing tool as a product until one of our customers (Find AI) asked to use it. We warehoused over millions of requests for them in the first week, logging 1,500 requests per second during their launch.
We’re announcing the product today, but are already warehousing over 3 million requests per week for customers. We didn’t expect that so many teams would want a trusted caching layer and analysis about opaque endpoints like OpenAI’s Batch API.
We support OpenAI and Anthropic endpoints. Data is formatted as JSON and logged to your own PostgreSQL instance. You can include queryable metadata in the header - like user ID, org ID, model ID, and version ID.
The queries you are really impressive - here are some questions I've used:
*Today we're launching this. We originally started building [x], and realized our biggest problem was [y]. So, we built [z].
Today we're launching, but already doing [impressive metric]. What we're seeing that we didn't expect is that [y].
If you warehouse your data today, you'll start to find interesting uses for the data soon. Set up Velvet in 2 lines of code and preserve the valuable AI requests you've made.*
—
Hey HN! We’re Emma and Chris from Velvet. We built an AI gateway to warehouse OpenAI and Anthropic requests to your database. Engineers use Velvet logs to analyze usage and costs, evaluate models, and generate datasets.
Our first product was an AI SQL editor. It worked really well, but we had no insights into what happened between our app and OpenAI. We wanted to query logs directly from our database, so we started warehousing requests. It completely changed the way we implement AI features - giving us full control over the process.
One of our customers (Find AI) asked to use the warehousing tool, and we quickly shipped it to production ahead of their launch. They started logging millions of requests per week - using Velvet logs to test new models, calculate costs (including opaque Batch API usage), and to query datasets (for fine-tuning, classification, embeddings).
We support OpenAI and Anthropic endpoints. Data is formatted as JSON and logged to your own PostgreSQL instance. You can include queryable metadata in the header - like user ID, org ID, model ID, and version ID.