--- title: "Noundry Platform Updates: TOON, AI SDK, and Sentinel - George Rios" description: "A look at the latest Noundry platform releases including TOON serialization, the AI SDK with chain prompting, the hosted AI Gateway, and Sentinel APM." url: https://georgerios.com/articles/noundry-2026-platform-updates author: George Rios --- [ All articles](https://georgerios.com/articles) Noundry # Noundry Platform Updates: TOON, AI SDK, and Sentinel The latest releases bring 40% token savings for LLM applications, a complete AI orchestration SDK, and unified observability. Published 2026-01-23 Reading time 4 min Topic Noundry Author George Rios The past few months have been busy. We've shipped several major updates to the Noundry platform that I'm genuinely excited about. Here's a breakdown of what's new and why it matters. ## TOON Serialization: 40% Fewer Tokens If you're building applications that send data to LLMs, you're probably bleeding money on unnecessary tokens. JSON is verbose. Every key is quoted, every nested object adds brackets, and the structure eats up tokens that should be going toward your actual content. TOON (Token-Oriented Object Notation) is a new serialization format designed specifically for LLM data transfer. The results speak for themselves: - **\~40% token reduction** compared to equivalent JSON payloads - Maintains full data fidelity (no lossy compression) - LLMs parse it just as easily as JSON Here's what it looks like in practice: ``` // Convert any object to TOON var toonData = myObject.ToToon(); // Or configure with options var options = new ToonOptions { Delimiter = "|", IncludeTypeHints = false }; var toonData = myDataTable.ToToon(options); ``` We've added extension methods for DataTable, DataSet, and all standard collections. If you're sending structured data to Claude, GPT, or any other LLM, TOON can cut your API costs significantly. ## Noundry.AIGW Hosted API The AI Gateway has been popular since launch, but running your own gateway requires infrastructure management. Some teams just want to call an endpoint. Enter the hosted API at `api.noundry.ai`. It's a BYOK (Bring Your Own Key) service: - You supply your own API keys for OpenAI, Anthropic, or Google - We relay requests with zero markup on provider pricing - Unified API across all providers (no code changes to switch models) - Built-in retry logic, rate limiting, and failover ``` // Same code, any provider var client = new NoundryAI("https://api.noundry.ai"); var response = await client.ChatAsync(new ChatRequest { Provider = "anthropic", // or "openai", "google" Model = "claude-3-opus", Messages = messages }, headers: new { x_api_key = yourAnthropicKey }); ``` The self-hosted version remains available for teams that need it. The hosted API is for everyone else who just wants AI integration without infrastructure headaches. ## Noundry AI SDK: The C# Answer to Vercel's AI SDK The AI Gateway was always step one. Step two is giving C# developers a complete AI orchestration toolkit that rivals what the TypeScript ecosystem takes for granted. The Noundry AI SDK (`Noundry.AIG.Client` on NuGet) provides a single, consistent C# interface across OpenAI, Anthropic Claude, Google Gemini, Groq, Ollama, and any OpenAI-compatible endpoint. Switch providers by changing a model string. That's it. But the real differentiator is **chain prompting** \--the ability to orchestrate multi-step AI workflows where each step's output feeds the next. We built [chainprompting.com](https://chainprompting.com) to showcase the pattern and the SDK in action: - **Multi-provider fallback.** Define a priority chain--OpenAI, then Anthropic, then Gemini--and the SDK handles failover automatically. No try-catch spaghetti. - **ChainPromptBuilder.** Compose complex workflows by linking LLM calls in sequence, mixing different models at each step for optimal cost and quality. - **Real-time streaming.** SSE-based token-by-token delivery via `IAsyncEnumerable` across all providers. - **Production-grade.** Thread-safe with `HttpClientFactory`, connection pooling, ASP.NET Core DI integration, and built-in analytics for tokens, latency, and costs. ``` // Chain prompting: generate → refine → translate var chain = new ChainPromptBuilder() .WithDefaultModel("openai/gpt-4o") .AddStep("Generate", _ => new PromptBuilder() .AddUserMessage("Generate a product description for the Noundry AI SDK.")) .AddStep("Refine", previousOutput => new PromptBuilder() .WithModel("anthropic/claude-sonnet-4") .AddUserMessage($"Refine this for a technical audience:\n{previousOutput}")) .AddStep("Translate", previousOutput => new PromptBuilder() .WithModel("openai/gpt-4o-mini") .AddUserMessage($"Translate to Spanish:\n{previousOutput}")); var result = await chain.ExecuteAsync(aigClient); if (result.Success) Console.WriteLine(result.FinalOutput); ``` The SDK is free and open-source. If you're a .NET developer building AI-powered applications, this is the toolkit we wish existed three years ago. Explore the patterns at [chainprompting.com](https://chainprompting.com) and grab the package from NuGet. ## Noundry.Sentinel: Unified Observability We've merged several monitoring tools into a single, self-hosted APM solution. Sentinel provides: - **Metrics collection** with customizable dashboards - **Exception tracking** with stack traces and context - **Log aggregation** with full-text search - **Distributed tracing** across services The goal was simple: you shouldn't need Datadog, Sentry, and ELK stack for a typical application. Sentinel handles all three concerns with a single deployment. For existing Logably users: Logably has been deprecated and merged into Sentinel. Migration is straightforward, and the feature set is strictly better. ## NounJS Hits v0.9.25 NounJS continues to mature with the latest release. At \~32KB bundled, it's larger than minimal libraries, but that's by design. NounJS includes a lot more out of the box: - **Utility functions** built-in, so you're not hunting for lodash alternatives - **UI as a first-class citizen** , not an afterthought - **Lower cognitive load** with a simpler mental model The philosophy is straightforward: when React, Vue, or Svelte feel like overkill, NounJS gives you everything you need without the complexity. It's not about being the smallest. It's about being the simplest to reason about. ## What's Next The roadmap for Q1 2026 includes: - Noundry.Authnz reaching GA with full OAuth 2.0 and OIDC support - Tuxedo 1.0 with the migration library (Bowtie) fully integrated - Additional UI components bringing the Razor library past 70 components Check the [full changelog](https://noundry.com/changelog) for version-by-version details, or reach out if you have questions about any of these updates. _Using Noundry in production? I'd love to hear about your experience. [Drop me a line](mailto:george@georgerios.com) or connect on the [Noundry community](https://noundry.com)._ ## References [ Back to articles](https://georgerios.com/articles) [Book a call ](https://cal.com/georgerios/30min) --- Contact: george@georgerios.com · Book a call: https://cal.com/georgerios/30min · Services summary for agents: https://georgerios.com/llms.txt