Show HN: Lyzr's low-code multi-agent automation framework (github.com)

Lyzr released Lyzr Automata, a low-code multi-agent automation framework (pip install lyzr-automata) that makes it easy to build directed, reusable AI workflows. The framework splits functionality into Models, Agents, Tools, Tasks and Pipelines: Models wrap LLMs (built-in adapters for OpenAI and Perplexity or extend the base AIModel), Agents encapsulate role/persona and memory to steer model behaviour, Tools expose external actions (with Pydantic input/output schemas) or custom functions, and Tasks bind agents, models and tools into discrete work units. Example configs include an OpenAIModel using gpt-4-turbo-preview and a Perplexity pplx-70b-online model; tools and I/O use explicit Pydantic classes so outputs are typed and composable. Why it matters: Lyzr targets practical automation patterns (content creation, publishing, pipelines) with intent-focused agents and typed tool integration, reducing boilerplate when orchestrating multi-agent LLM systems. A LinearSyncPipeline example shows how to chain tasks (e.g., “write LinkedIn post” → “upload post”); the team plans richer async/DAG orchestration and steadier code structure in future releases. The project is experimental and open for contributions—useful for developers prototyping production-ready multi-agent flows with typed tool endpoints and clear agent personas.