Swarm

A Swift framework for building agents and multi-agent workflows.

Chain LLMs, tools, and memory into production workflows -- with compile-time safety, crash recovery, and on-device inference.

Why Swarm?

• Swift concurrency is part of the surface. Swift 6.2 StrictConcurrency is enabled across the package.
• Tools stay type-safe. The @Tool macro generates JSON schemas from Swift structs.
• Workflows can survive crashes. Durable workflow checkpointing lets you resume from an explicit checkpoint ID.
• Cloud and on-device models use the same abstractions. Foundation Models, Anthropic, OpenAI, Ollama, Gemini, MiniMax, OpenRouter, and MLX all fit the same shape.
• It is written in Swift all the way down. AsyncThrowingStream, actors, result builders, and macros are first-class here.

Highlights:

• Compose Workflows Fluently: Build sequential, parallel, and routed workflows with Workflow().step(...).parallel(...).route(...).
• Data Races Are Compile Errors: Swift 6.2 StrictConcurrency on every target. Non-Sendable types crossing actor boundaries won't build.
• Workflows Survive Crashes: Advanced workflows can checkpoint and resume with explicit checkpoint stores and deterministic IDs.
• Semantic Memory -- On-Device: VectorMemory uses Accelerate SIMD for cosine similarity. No cloud API. No network call. Just fast local retrieval.
• Any LLM, Same Code: Foundation Models, Anthropic, OpenAI, Ollama, Gemini, MLX. Swap providers with inferenceProvider: at init.
• Production Resilience: Retry with 7 backoff strategies, circuit breakers, fallback agents, rate limiting, and per-step timeouts.