OpenAI released the Responses API yesterday. 24 hours later I have reimplemented the API and it's running on Ragwalla, my service.

RT @thecyrusjanssen: An Iranian man left this comment on my YouTube channel. It's the single best explanation I've ever heard on the future….

Everything you need to know about India and Pakistan in a single sentence.

Correction: I removed it from the AppStore —- it’s only available from our website to download.

AI Files supports the OpenAI API and Ragwalla, my implementation of the OpenAI Files and Vector Stores APIs in addition to Assistants and Responses. Here's the Ragwalla blog post on AI Files too:

So I wrote a simple Mac app in Swift that allows me to configure multiple API keys, stores them securely in KeyChain Access, allows bulk operations and supports keyboard commands.

But that really can't be the right answer. Now I have to manage API keys, copy vector store IDs, include the dependencies, etc. It's a huge pain.

I spend an unreasonable amount of time working with the @OpenAI Files and Vector Store API. But the OpenAI Dashboard does not support bulk management of files whatsoever. Want to delete lots of files? Get clicking or write some Python.

Ragwalla supports 5,000,000 vectors per vector store and you can use many vector stores together!.

If you're interested in using the Responses API with true choice with respect to completion model and vector embeddings at very large scale -- you should try Ragwalla. It's at

With Ragwalla you won't be limited by a maximum of 10,000 documents that OpenAI supports in a given vector store.

Now you can use the Responses API with completion models such as Google's Gemini, Deepseek, Llama, Mistral, Grok, etc. And you can set the embedding model that best fits your use cases such as those from @JinaAI_ , @OpenAI , @cohere , @GoogleAI , etc.

I had been reimplementing the Assistants API on Ragwalla, so the plumbing was already in place for streaming, function calling, vector search, tons of embedding models and support for more than 200+ completion models through @OpenRouterAI.

PDF parsing costs comparison between Google Gemini, Anthropic Claude Sonnet, and Mistral OCR: . Google's Gemini 2.0 Flash: 1,000 pages * $0.00025 = $0.25​. Mistral OCR: 1,000 pages * $0.001 = $1.00​. Anthropic's Claude 3.7 Sonnet: 1,000 pages * $0.009 = $9.00.

Good luck getting anything other than an HTTP 500 response from using PDF and OCR functionality with the @mistralai API.

RT @sunbains: DeepSeek doesn’t seem like a side project by some quants. Writing a correct distributed file system alone is a multi year p….

The Economist is not messin’ around today.

RT @PaulieBlade: @AlexandruC4 Everyone knows BRICS stands for Belgium, Rhode Island, Canada and Spain.

RT @BostonSatire: Report: Imagine if this was all snow, New England dads posit

Apparently, it's scheduled maintenance. LOL.