🎉 Sooo, I've been working on something fun, and I’m super excited to share it with you all! 🎉. Meet PopLight – my first-ever Mac menu bar app! It’s a tiny little dot that tells you if you’re actually connected to the internet.

Just added the "choose default response" setting to preferences using this workflow to the FlexiChat app. I like it.

Here's how using VS Code for building iOS apps looks like now end-to-end for me. Edit code in VS Code with Claude Code assist within. CMD+SHIFT+P to run the app (using Sweetpad extension). Simulator launches automatically when the build is complete. Profit.

Alright, canceled Cursor Pro and @alexcodes_ai for now. Going all-in on Claude Code Max for all things code. VS Code + SweetPad Xcode CLI for all things iOS/macOS. Will report how this goes :-)

RT @0xAvious: Overhead at a SF cafe:. "What's Windsurf?"."It's like Cursor… but the founder walked away from generational wealth for his em….

Overheard at a NYC café:. "What's an LLM?". "It's like ChatGPT… anything you can chat with.". It’s easy to laugh or correct. But that's what I love about New York. You forget how deep in the tech bubble you are until you step outside it. Not everyone is thinking about AI,.

and the full list of supported models here:.

You can also read more about how Serverless Inference works here:.

Here's a full guide with all the code examples from this thread:.

You can also pass `include_retrieval_info=True` if you want to get the source metadata as well.

To talk to an Agent, use the same OpenAI SDK. Same method. Just change the `base_url` to your Agent endpoint (don’t forget to add /api/v1 at the end).

You can use the same approach with DigitalOcean Agents too. What are Agents?. Agents are models connected to your own knowledge base - docs, URLs, or structured data. They’re great for internal tools, documentation assistants, or domain-specific bots.

That’ll print out the streaming response - especially helpful if you’re building a chatbot or an assistant-style UI.

Streaming is supported too. Just add `stream=True` and handle the tokens as they arrive.

That’ll return the completion response - same format you’re used to from OpenAI.

Run a chat completion. Call `.chat.completions.create()` exactly like you would with OpenAI SDK.

That should give you a list of all the supported models from OpenAI, Anthropic, Meta, DeepSeek, and Mistral - with more being added regularly.

List available models. You can use `.models.list()` to see all the models DigitalOcean serves.

Use the OpenAI SDK like normal. Everything starts with the client. All you do is change the `base_url` to point to DigitalOcean's Serverless Inference endpoint.

🎉 @DigitalOcean now offers Inference APIs. No GPUs to manage. No models to host. Just one endpoint and one API key. You can run GPT-4o, Claude 3, LLaMA 3, and more - right from the @OpenAI SDK. Here’s what that looks like in practice. 👇

Been chipping away at a Hugo content manager for my personal site. Just ~2 focused hours over the past couple days, and it’s really coming together.