OuroborosAI.Core 2.9.6

What is Ouroboros?

Ouroboros makes it easy to chain and transform API calls with OpenAI. You get:

• Clean chaining: Fluent interface for feeding the output of one API call into the input of another.
  • Easily capture the output of one call, save it to a variable, and then use it as input downchain.
• Template engine: Store your prompts as markdown right in your project, with a corresponding class for fields.
  • Store your prompts as markdown right in your project.
  • You can use a class with the same name for field injection using {{ mustache syntax }}.
  • You can also inject other templates using the same syntax.
• Transform model results into code: using our Hermetic Codex and Proteus Convert (both parts of this project).
  • Detect a numbered list and auto-transform it into a List<String> or List<NumberedListItem>
  • Convert results directly into classes or enums.
  • Smart, recoverable detection of errors / unmappable output.
• Exponential backoff (retry): Always on, for transient outages via <a href="https://github.com/App-vNext/Polly">Polly</a> on all calls.
• Simple Chat API: You can also do regular Chat Completions calls, which gives you Retry.

How do I get started?

Check out the Getting Started Guide (this needs updating).

Where do I get it?

First, <a href="http://docs.nuget.org/docs/start-here/installing-nuget">install NuGet</a>. Then you can install Ouroboros from the package manager console:

PM> Install-Package OuroborosAI.Core

Limits and Possible Contributions

Ouroboros is production-ready, but it does have limits. If you would like those limits to go away, get involved!

• Only supports OpenAI API calls. We built this on top of Betalgo, so it should also be possible to support OpenAI on Azure.
• Chaining only supports the Chat Completions API, meaning GPT 4.1, GPT 4o, etc.
• You can't modify our retry policy, although you can turn it off.
• We could use some help implementing Logging, support for other providers, and the new Responses API.