A Rust SDK for AI Agents
A Rust SDK for AI Agents
Get Started
Works with these providers
Why agents-sdk
Zero-Cost Abstractions
Compile-time checked agent definitions. No reflection, no runtime overhead. Pure Rust performance.
Multi-Model Routing
Route tasks across providers with automatic fallbacks. Cost and latency-aware scheduling built in.
Type-Safe Tools
Derive macros for tool schemas. Full serde integration with automatic validation.
Tracing Integration
First-class tracing support. Export to OpenTelemetry, Jaeger, or any tracing subscriber.
Async Streams
Native async/await with tokio. Stream responses, handle backpressure, cancel anytime.
Memory Safe
Rust's guarantees extend to your agents. No data races, no null pointers, no surprises.
Performance
M3 Max, 64GB RAM · Claude Sonnet 4 · 1000 iterationsArchitecture
Your Application
Where agents-sdk integrates
agents-sdk
RUSTNative Rust core
Native Tools
Built from scratch in Rust
Providers
Swappable LLM backends
How we compare
Releases
View full changelogMulti-model routing & cost tracking
- New Router trait for custom model selection logic
- Built-in cost tracking per provider with budget limits
- Automatic fallback chains when primary model fails
- Gemini 2.0 provider support
Streaming improvements
- Zero-copy streaming with Pin<Box<dyn Stream>>
- Backpressure handling for slow consumers
- Fixed memory leak in long-running streams
Tool system overhaul
- New #[derive(Tool)] macro with automatic schema generation
- Validation via validator crate integration
- Async tool execution with timeout support
- MCP server compatibility layer
Multi-agent orchestration
- Agent-to-agent communication primitives
- Supervisor pattern with delegation
- Shared memory between agents
- Visual workflow builder integration
Quick Start
Get Started
Start building with Agents SDK
Everything you need to build, test, and deploy AI agents with Rust.