SapphireGuard 

Core abstractions and harness loop for structured LLM agents in .NET. An agent = model + harness; this package is the harness.

Built-in implementations for SapphireGuard.ModelHarness: sensors, tracing, tool registry, skills, HITL, and the standard DI wiring.

Anthropic Claude adapter for SapphireGuard.ModelHarness. Wires ClaudeModelClient into the harness via the IModelClient port.

Azure OpenAI / Azure AI Foundry adapter for SapphireGuard.ModelHarness. Supports API key and DefaultAzureCredential authentication.

Polly-backed retry and circuit breaker decorator for SapphireGuard.ModelHarness. Wraps any IModelClient with resilience policies.

Ollama adapter for SapphireGuard.ModelHarness via OllamaSharp. Run the harness against locally-hosted models.

File-based checkpoint and resume for SapphireGuard.ModelHarness. Save agent state to disk and resume after a crash or restart.

Domain types, evaluation orchestrator, deterministic evaluators, and file-system adapters for AI skill/agent evaluation in .NET. No MAF or MEAI dependency — bring your own runner.

Drop-in ASP.NET Core live dashboard for SapphireGuard.ModelHarness. MapHarnessDashboard() serves a real-time agent console — runs, per-turn telemetry, per-event drill-in, and the run result — over Server-Sent... More information

Domain types, evaluation orchestrator, deterministic evaluators, and file-system adapters for AI skill/agent evaluation in .NET. No MAF or MEAI dependency — bring your own runner.