AutoAgents: Build Agents in Rust

AutoAgents is LiquidOS’ open-source Rust SDK for building reliable agentic systems. It’s designed for performance, safety, and scalability — and ships with the core building blocks you need to take agents from a demo to a deployable service.

What “production-ready” looks like

In production, agents need more than prompts. AutoAgents helps you structure agent systems with:

• Explicit runtimes: build agents as DirectAgent (simple) or actor-based agents for concurrency.
• Provider flexibility: choose only the model backend you need (cloud or local) via Cargo features.
• Tooling boundaries: attach tools through autoagents-toolkit (filesystem, search) and keep execution controlled.
• Composability: wire together the Executor → Agent → Tools/Memory/Providers stack and evolve it over time.

Quick start: build an agent in Rust

1. Create a Rust project:
   • cargo new my_agent && cd my_agent
2. Add dependencies:
   • cargo add autoagents
   • cargo add autoagents-derive
3. Build your agent:
   • Define an agent with #[agent(...)] and AgentHooks
   • Configure an LLM backend using LLMBuilder
   • Run tasks through the BasicAgent executor

AutoAgents supports multiple provider backends (for example: openai, anthropic, openrouter, groq, google, azure_openai, xai, deepseek, ollama). Enable only what you use.

Make it production-ready

Once you have the minimal agent working, the next step is to harden the system around it:

• Choose the right runtime model
  • Use DirectAgent for simpler request/response flows.
  • Use actor-based agents when you need concurrency, orchestration, or multi-agent workloads.
• Add tools deliberately
  • Pull in autoagents-toolkit features (filesystem, search) for standard integrations.
  • Prefer explicit tool surfaces over “magic” behavior; treat tools as part of your API contract.
• Test and iterate using examples
  • The repo includes examples like examples/basic and examples/coding_agent for multi-turn and tool-enabled agents.
• Plan for deployment
  • Keep configuration externalized (keys, model selection, timeouts).
  • Start with clear boundaries (tools, providers, memory), then grow into richer orchestration as needed.

AutoAgents is available today. LiquidOS Platform is coming soon as an agentic operating system with local-first execution for embedded systems, robotics, and local agent workflows, plus tracing, policy controls, evals, and deployment workflows.