// the find
dynamiq-ai/dynamiq
Dynamiq is an orchestration framework for agentic AI and LLM applications
Dynamiq is a Python framework for building agentic LLM workflows: RAG pipelines, single and multi-agent orchestration, and a graph-based state machine for things like feedback loops. It's aimed at teams who want an opinionated DAG/workflow layer on top of LLM calls instead of wiring OpenAI/LangChain primitives by hand.
The integration catalog is genuinely wide for a 1k-star project — a dozen+ LLM providers, most major vector stores (Pinecone, Qdrant, Weaviate, pgvector, OpenSearch, Milvus), document converters, and even audio STT/TTS backends. The explicit `.depends_on()` / `InputTransformer` wiring between nodes makes data flow traceable instead of implicit, which is a real problem in some agent frameworks. The Graph Orchestrator with conditional edges and a real `END` state is a legitimate primitive for iterative agent loops (human-in-the-loop feedback, retries), not just linear chains.
Every README example hardcodes API keys as literal strings passed into connection objects — no mention of env vars or secrets config in the primary onboarding doc, which is the kind of thing people copy-paste straight into committed code. The module list (knowledge_graphs, detectors, evaluations, artifacts, checkpoints, cache, audio) is huge for the contributor count implied by 143 forks, so expect uneven depth and maintenance across the less-central pieces. The CLI ships subcommands for org/project/deployment/service/trigger that strongly suggest a hosted platform behind the open-source client — worth checking how much of the 'framework' actually requires their backend before building on it standalone.