Rather than directly calling OpenAI APIs and enforcing monolithic vendor lock-in within your Agent, ContextFlow uses standard abstract Provider bridges.
Providers natively execute with async def arun(). This guarantees that your surrounding State Graph architecture (LangGraph) continues ticking async worker threads concurrently without locking the CPU while fetching network LLM requests.
When constructing complex Graph loops, you can inject MockProvider() to run unit tests and latency traces natively over your pipeline arrays without utilizing a single real API token.
Using standard openai.AsyncClient, the wrapper automatically handles flattening ContextFlow Pydantic Arrays safely into standard to_llm_dict() requirements internally, dropping tiktoken byte identifiers and priority ranks natively prior to the network hop.