dynamiq

Why we included this project

If you're assembling multi-step LLM workflows and want to avoid hand-wiring every call, this Python framework is worth a look. It treats agents, tools, and model calls as nodes you can connect into sequential or parallel flows, so a RAG pipeline or a ReAct-style agent with a code interpreter becomes configuration rather than plumbing. The framework takes care of the boring parts: passing outputs between dependent steps and limiting how many loops an agent can run. The examples show common patterns quickly, from a single translation call to two agents sharing a pipeline, which makes it a practical starting point before you commit to a heavier platform. It suits developers comfortable with Python who manage their own API keys, and the workflow abstraction also offers a way to standardize internal AI wiring.