Steve Sanderson

You can extend Copilot CLI with any UI you want - here’s how! youtu.be/HcjUnrS41II

Shipped a new build of the Copilot SDK that allows agents to link up to the same session. Unlocks a whole ecosystem of client to client agent interactions.

Something I didn’t expect to see - an updated release of knockout.js: github.com/knockout/knock…

Follow the @github Copilot *actual devs* no marketing here, in one list! -> x.com/i/lists/202926…

youtu.be/L1w6wBxhpgE?si… (from last month, but many parts are still true!)

@JamesMontemagno Yes, no change to auth.

@stevensanderson Does auth work the same if they are logged in on their local machine?

As of 5 minutes ago, Copilot SDK's Node/.NET/Python packages now auto-bundle Copilot CLI in your build output. It means apps built on Copilot SDK don't have to worry about deploying Copilot CLI manually, and version mismatches are no longer possible. github.com/github/copilot…

@AdaTheDev Hasn’t been announced but don’t be surprised if it’s soon

@stevensanderson Any idea when Copilot CLI is likely to go GA Steve?

@Richie_Dalziel It’s a self-contained native binary dependency, so you don’t need to add a node runtime.

@stevensanderson How would this work in something like app services ? I assume we still need a node container or is that not the case ?

Just published: new AI Chat Web App template Been wondering how to get started with .NET and AI? Well here’s one easy option! It takes in knowledge from your own data sources, gives citations, and runs AI inference on GitHub Models (free) by default. devblogs.microsoft.com/dotnet/announc…

@davidfowl @prdevlin1 @1_amaechi @danroth27 You can deploy to a Linux host if you also set up Nginx or another web server to serve those files. Blazor WebAssembly is just static files so it can’t serve itself.

@prdevlin1 @1_amaechi Oh maybe some app service limitation? I don't have the background. cc @stevensanderson @danroth27 might know more

C#

Do you know that .net9 now has a basic form of escape analysis to avoid heap allocations in some cases? Here is a simple case: allocating a new object, getting a property or calling a method and not storing the object. The current logic is limited, but it's huge!!

@nonlinear_james @EvanChaki @ollama That’s not how I understand it. They bill it as an audio-to-audio model, so it doesn’t have to pass through any transcription step.

@stevensanderson @EvanChaki @ollama I'm giving it a go... Realtime api from openai is nothing more than Whisper streaming (there's research papers on how this works) that then connects into standard chat streaming, and then back out with cancellation. So it should be able to reproduce that with Semantic Kernel.

OpenAI and Azure OpenAI recently released “realtime” APIs. Let’s experiment! Can we build a new kind of voice-based UI experience? youtu.be/H4BEF-VSDXQ?si…

@nonlinear_james @EvanChaki @ollama Haven’t tried it, but would be interested if you do. Traditional speech recognition might not be fast enough to feel natural (the point of the realtime APIs is to enable this responsiveness) but maybe that’s wrong and it would work.

@EvanChaki @stevensanderson @ollama That’s awesome! Any chance of a sample with ollama instead of OpenAI? I know it would require more work but should be able to use the streaming text response with whisper and then do cancellation to mimic the realtime functionality in the OpenAI client.

Seems that neverssl.com is now served with… you guessed it… SSL?

@cecilphillip @condrong Slightly. Hyperlight is a much stronger form of isolation as it’s backed by hardware virtualization features.

@stevensanderson @condrong Is this related to the dotnetIsolator stuff you showed off a while back?

Hyperlight, a library for micro VMs, has just been open-sourced: github.com/hyperlight-dev… I worked with getting .NET to run on wasm inside this a while back. It’s still bleeding edge, but with work you can start up lightweight hardware-isolated tasks incredibly fast.

@kzu @john_kattenhorn Call CompleteAsync<T>, optionally passing useNativeStructuredOutput:true

@stevensanderson @john_kattenhorn How do you do structured outputs I with OpenAI models using these abstractions? See openai.com/index/introduc…

Introducing Microsoft.AI.Extensions - a library to streamline use of AI in .NET apps. Glad to see this go live today! The library is something I and a few smart people have been working on in recent months. devblogs.microsoft.com/dotnet/introdu…

🚀 Exciting news! @dotnet and #SemanticKernel are thrilled to announce a streamlined AI integration with Microsoft.Extensions.AI, now available in Preview! 🎉 #dotnet #ai devblogs.microsoft.com/dotnet/introdu…