Posted on Jun 7, 2024 • Originally published at elbruno.com on Jun 7, 2024

Sample Code using the new OpenAI library for .NET

Hi!

The new OpenAI SDK for NET was officially announced, so today, let’s review the annoucement and show some sample code on how to use it.

The current blog scenarios are

Sample Chat demo
Sample Audio to Text demo
Sample Image analisis demo

Here is a sample output for the chat demo! And yes, with a funny system message and questions about France!

Code Repository: GPT-4o Labs

OpenAI SDK for .NET

Microsoft Build 2024 has unveiled new AI investments for .NET developers, including the first beta release of the official OpenAI .NET library, version 2.0.0-beta.1. This library facilitates smooth integration with OpenAI and Azure OpenAI, complementing existing libraries for Python and TypeScript/JavaScript.

Developed on GitHub, the .NET library will stay current with OpenAI’s latest features, with ongoing work to refine it based on community feedback. The release acknowledges Roger Pincombe’s pioneering work on the initial OpenAI .NET package and encourages continued innovation from community library developers. Participation and collaboration within the community are highly encouraged as the project progresses.

_ Note: Part of the content of this post was generated by Microsoft Copilot, an AI assistant._

Scenario 1: Chat

Time to share the “Hello World” of using GPT models. The following sample is a console application that interacts with the OpenAI API to generate chat responses.

It retrieves an API key from user secrets and specifies the model to be used for the chat.
It initializes a ChatClient with the model and API key.
The system message and user question are defined and added to a list of chat messages.
The chat is then completed using the ChatClient and the response from the chat is retrieved.
Finally, the system prompt, user question, and chat response are displayed in the console.

This is the sample output for this program:

System Prompt: You are a useful assitant that replies using a funny style.

User Question: What is the capital of France?

Response: Well, let me put on my fancy beret and sip some imaginary café au lait while I tell ya – the capital of France is the one and only Paris! Yes, indeed, the city of love, croissants, and an Eiffel Tower that’s basically the world’s largest toothpick! So, if you’re planning to hunt for baguettes or have a romantic escapade, Paris is where the magic happens! ??????

Scenario 2: Audio

The SDK also allow us to work with Whisper and audio files.

The following code is a C# console application that uses the OpenAI API to transcribe audio files.

It sets up the configuration to retrieve the OpenAI API key and specifies the model to be used for transcription.
It creates an AudioClient instance with the model and API key.
The audio file to be transcribed is specified, and transcription options are set, including the response format and granularities for timestamps.
The audio file is then transcribed, and the transcription text is printed to the console.
Additionally, the start and end times for each word and segment in the transcription are also printed to the console.

This is a sample output for this program:

 Transcription: Estás escuchando No Tiene Nombre, un podcast sobre tecnología que es bastante probable que sea escrito por una inteligencia artificial, por ejemplo, con chat GPT. El host, autor, editor y encargado de los efectos visuales del podcast es Bruno Capuano. Puedes contactarlo en Twitter en arroba elbruno o en las redes sociales también buscando por elbruno. Mi nombre es Elena, de Italia YOS invito al episodio de hoy. Subtítulos por la comunidad de Amara.org Words: Estás : 700 - 1420 escuchando : 1420 - 2060 No : 2060 - 2420 Tiene : 2420 - 2580 Nombre : 2580 - 3000 un : 3340 - 3500 podcast : 3500 - 3780 sobre : 3780 - 4100 tecnología : 4100 - 4720 que : 4720 - 4960 es : 4960 - 5120 bastante : 5120 - 5560 probable : 5560 - 6020 que : 6020 - 6260 sea : 6260 - 6480 escrito : 6480 - 6820 por : 6820 - 7100 una : 7100 - 7460 inteligencia : 7460 - 7820 artificial : 7820 - 8380 por : 8820 - 9020 ejemplo : 9020 - 9300 con : 9680 - 9880 chat : 9880 - 10060 GPT : 10060 - 11340 El : 11360 - 11760 host : 11760 - 12100 autor : 12660 - 12660 editor : 13140 - 13140 y : 13140 - 13340 encargado : 13340 - 13740 de : 13740 - 13920 los : 13920 - 14300 efectos : 14300 - 14420 visuales : 14420 - 14860 del : 14860 - 15020 podcast : 15020 - 15400 es : 15400 - 15760 Bruno : 15760 - 15920 Capuano : 15920 - 16400 Puedes : 17640 - 17700 contactarlo : 17700 - 18320 en : 18320 - 18500 Twitter : 18500 - 18740 en : 18740 - 18920 arroba : 18920 - 19160 elbruno : 19160 - 19540 o : 19540 - 19960 en : 19960 - 20080 las : 20080 - 20360 redes : 20360 - 20360 sociales : 20360 - 20800 también : 20800 - 21180 buscando : 21180 - 21560 por : 21560 - 21960 elbruno : 21960 - 22340 Mi : 23480 - 23560 nombre : 23560 - 23840 es : 23840 - 24280 Elena : 24280 - 24440 de : 24780 - 24940 Italia : 24940 - 25360 YOS : 25360 - 26180 invito : 26180 - 26660 al : 26660 - 26880 episodio : 26880 - 27340 de : 27340 - 27720 hoy : 27720 - 27720 Subtítulos : 30000 - 31540 por : 31540 - 31540 la : 31540 - 31540 comunidad : 31540 - 31540 de : 31540 - 31540 Amara : 31540 - 31540 org : 31540 - 31540 Segments: Estás escuchando No Tiene Nombre, un podcast sobre tecnología que es bastante probable : 700 - 6480 que sea escrito por una inteligencia artificial, por ejemplo, con chat GPT. : 6480 - 12660 El host, autor, editor y encargado de los efectos visuales del podcast es Bruno Capuano. : 13140 - 18500 Puedes contactarlo en Twitter en arroba elbruno o en las redes sociales también buscando por elbruno. : 18500 - 26180 Mi nombre es Elena, de Italia YOS invito al episodio de hoy. : 26180 - 27720 Subtítulos por la comunidad de Amara.org : 30000 - 31540

Scenario 3: Using Vision with GPT4o.

And the final scenario will be on how to use vision capabilities with this new SDK. The following program project is a C# console application that uses the OpenAI API to interact with an AI assistant.

It starts by reading the OpenAI API key from the user secrets and initializing the OpenAI client with this key.
It uploads an image file named foggyday.png from the imgs directory to the OpenAI server for the purpose of vision tasks.
After the image is uploaded, it creates an AI assistant with a specific instruction to reply in a funny style.
It starts a new thread with the assistant, sending an initial message asking the assistant to describe the uploaded image.
The program then enters a loop, listening for updates from the assistant.
When a new message is received from the assistant, it is printed to the console.
The program continues to listen for updates until it is manually stopped.

This is a sample output for this program, and below is the image used to test:

 --- Run started! --- Well, well, well! What do we have here? Is that paradise? A spa? Or just someone's backyard on a dreamy, misty morning? There's a pristine blue pool just begging for a cannonball, surrounded by neat paving stones that scream, "We mean business!" Looming over the scene is a terrific tree that looks like it's auditioning for the role of "Majestic Sentinel." In the distance, the fog is putting up a great show, hiding whatever secrets lie beyond this yard-maybe a herd of unicorns or a neighborhood of lawn gnomes planning their next move. The whole scene feels like it's on the set of the next great mystery movie... or maybe just an ad for pool cleaning services. Either way, sign me up for a dip! D:\sk\gpt4ol-sk-csharp\src\OAINETSDK_lab03\bin\Debug\net8.0\OAINETSDK_lab03.exe (process 3668) exited with code 0. To automatically close the console when debugging stops, enable Tools->Options->Debugging->Automatically close the console when debugging stops. Press any key to close this window . . .