Wiring agentic AI into explicit Python DAGs

What it does

Dynamiq is a Python framework that treats LLM calls, agents, document processors, and vector stores as nodes in a directed graph. You wire them together with explicit dependencies, and the Workflow class runs what it can in parallel while respecting the DAG order. It covers the usual suspects: ReAct agents with tool use, multi-agent delegation, and classic RAG indexing pipelines.

The interesting bit

The framework leans heavily into composition: a manager agent can treat other agents as tools, while the underlying engine still sees the whole thing as a node graph with .depends_on() and .inputs() plumbing. That keeps the data flow explicit even when the control flow is agentic.

Key highlights

• Node-based abstraction: LLMs, embedders, splitters, and even agents are nodes wired into a Flow.
• Built-in patterns for parallel and sequential agent execution, plus multi-agent orchestration via manager delegation.
• ReAct agents with tool support, including code interpreters and search.
• RAG pipeline examples covering PDF conversion, chunking, OpenAI embeddings, and Pinecone ingestion.
• Apache 2.0 licensed, Python 3.10+.

Verdict

Worth a look if you want your LLM orchestration expressed as explicit Python objects rather than framework magic. If you are already happy with your current agent framework and don’t need another DAG abstraction, this won’t change your mind.