AutoAgents – a Rust-Based Multi-Agent Framework for LLM-Powered Intelligence (github.com)

AutoAgents is a new Rust-based multi-agent framework for building LLM-powered intelligent systems, announced by Liquidos AI. It provides a modular, performance-focused stack (built on tokio and Ractor) that supports cloud, edge and in-browser deployments via native Rust and WebAssembly. Key features include multiple executors (ReAct and basic streaming executors), type-safe structured outputs with JSON schema validation, configurable memory backends (sliding-window today, persistent storage planned), sandboxed WASM tool execution, and composable tools via derive macros. The project is provider-agnostic with out-of-the-box integrations for OpenAI, Anthropic, Ollama and others, plus local model support (Mistral-rs, Burn, ONNX) for privacy- and latency-sensitive use cases. For the AI/ML community this matters because it combines Rust’s safety and performance with multi-agent orchestration, enabling scalable, concurrent agent teams that can run on servers, browsers, or edge devices. The modular crates and swappable components (executors, memory, LLM backends) make it easy to experiment with reasoning patterns, local inference, and hybrid architectures. YAML-driven workflows and a CLI/serve mode speed up prototyping, while type safety and sandboxed WASM tooling reduce runtime surprises and attack surface. Overall AutoAgents targets practical production use—low-latency local inference, secure tool execution, and composable multi-agent workflows—making it a useful foundation for next-gen autonomous AI systems.