Building a Library-First AI Agent SDK in Rust

Описание и применение

Создан Rust‑SDK для AI‑агентов, позволяющий встраивать агент в сервисы, уменьшать потребление памяти и расширять функциональность через динамические библиотеки.

Задача

Need an embeddable, performant Rust agent runtime instead of a TypeScript/Node CLI that hits memory limits on low‑spec servers.

Результат

The Rust SDK enables embedding an AI agent into services, reduces memory usage, and allows plugin extensions via stable C ABI.

Рабочий процесс

• 1. Design library‑first architecture with crates: rpi‑ai, rpi‑agent, rpi‑tools, rpi‑harness, rpi‑cli.
• 2. Implement minimal agent example using FauxProvider (no API key).
• 3. Define custom tool (AddTool) and integrate via AgentBuilder.
• 4. Bridge async Provider to sync StreamFn for compatibility.
• 5. Run example with cargo run and observe streaming output.
• 6. Replace FauxProvider with real provider (Anthropic/OpenAI) for production use.
• 7. Extend with native Rust cdylib plugins via rpi‑plugin‑sdk ABI.
• 8. Install extensions with rpi install <name> and use in CLI or embedded runtime.

Доказательства

I was running an agent on a low-spec server and kept hitting memory limits. So I ported the core runtime to Rust.

Источник

rpi is split into independent crates. Each layer can be used on its own: rpi‑ai, rpi‑agent, rpi‑tools, rpi‑harness, rpi‑cli.

Источник

Run it with: cargo run

Источник