Show HN: Radkit – Build reliable AI agents with Rust (github.com)

Radkit is a new Rust SDK for building reliable, production-grade AI agent systems that puts developer control and type safety front and center. It ships native support for an Agent-to-Agent (A2A) communication standard, a unified LLM interface that abstracts Anthropic, OpenAI, Gemini, Grok and DeepSeek, and optional runtime features (HTTP server, dev UI, tracing) enabled via Cargo features. Developers keep low-level access when needed but get high-level primitives — Thread, Event, Content — for structured conversation state, plus optional server runtime and an interactive native UI. Technically, Radkit emphasizes correctness with Rust’s type system and JSON Schema-based structured outputs. LlmFunction<T> produces type-safe deserialized responses for predictable downstream logic, while LlmWorker<T> adds automatic tool calling, multi-turn loops, and stateful execution using #[tool] macros. The SDK integrates with tokio, serde, schemars, and supports environment/key configuration for each provider. Implications: unified multi-provider APIs make swapping LLMs easier, structured outputs and tool automation reduce brittle parsing and orchestration code, and A2A-first design enables scalable multi-agent architectures. It’s aimed at teams needing deterministic, auditable agent behavior and safe, composable tooling in Rust-based production systems.