dynamiq-ai/dynamiq

What it solves

Dynamiq solves the complexity of building AI-powered applications that use large language models (LLMs) and retrieval-augmented generation (RAG). Instead of stitching together many components manually, it provides a unified framework to define, run, and orchestrate AI workflows.

How it works

Dynamiq is a Python framework (requires Python 3.11+) built around a workflow graph. Developers define nodes (such as LLM calls, agents, tools, converters, splitters, embedders, retrievers, writers) and connect them in a directed acyclic graph (DAG) to form a flow. Nodes can run sequentially or in parallel, and outputs can be wired as inputs to other nodes. The framework includes built-in agents (like ReAct), memory for conversation history, and a graph orchestrator that lets you create custom state machines (e.g., for iterative feedback loops). It also has connectors for many LLM providers, vector stores (e.g., Pinecone, Chroma), and document converters.

Who it's for

Developers building LLM-powered applications or agent-based systems in Python. It's suited for teams that want a structured way to assemble workflows—from simple one-shot LLM calls to complex multi-agent orchestration and RAG pipelines—without building all the plumbing from scratch.

Highlights

• Ready-to-use components: agents, LLM nodes, tools, memory, vector store integrations, and document converters.
• Two execution modes: sequential and parallel workflows, plus a graph orchestrator for custom state flows.
• Supports async execution, e.g., for ReAct agents.
• Optional extras for vector stores, file processing, data visualization, SQL, browser automation, sandboxes, and more.
• Eager import toggle (DYNAMIQ_EAGER_IMPORTS=1) to pre-load integrations for faster server startup.
• Includes built-in memory backends for chatbots.
• Provides examples for RAG indexing and retrieval, multi-agent delegation, and email refinement loops.