stream-chain

Multi-agent workflow orchestration with streaming is handled by passing the output of one processing step directly as the input to the next. This allows sequential data transformations by invoking a run mode with a sequence of prompts.

Custom chains use the run mode to execute user-defined prompt sequences for complex tasks, while predefined pipelines use the pipeline mode to provide battle-tested workflows with structured steps and predictable behavior for common automation scenarios.

Yes, context flow and memory persistence are supported across sequential steps. Each step receives the full prior outputs and can store state, enabling cross-session context propagation for continuous data processing pipelines.

Multi-agent orchestration supports configurable timeouts to manage step execution limits and includes debugging options. These features help monitor and control context transfer and streaming behavior across diverse agents in a single chain.

Streaming should be used for multi-step code analysis and automation tasks when you need to transfer context sequentially between diverse agents. It enables complex decision-making by allowing each step to process the previous step's output directly.

Sequential context transfer limits workflows to dependent steps where each stage requires the previous output. While memory persistence and configurable timeouts help manage state, this orchestration model is not suited for parallel execution of independent automation tasks.