teff.stream

teff.stream

Streaming events for graph execution.

Constitution Principle IX: observability is mandatory. graph.stream() emits :class:StreamEvent objects as a run progresses, so callers can render tokens, progress, and routing decisions before the run finishes — instead of waiting for a fully materialised result from graph.run().

Classes:

Name	Description
StreamEvent	A single event emitted while a graph streams.

StreamEvent dataclass

A single event emitted while a graph streams.

Attributes:

Name	Type	Description
type	str	Event type — run_start, node_start, node_end, node_error, edge, token, llm, tool_call, structured, interrupt, interrupt_resume, checkpoint, or run_end.
node_id	str | None	Graph node id the event belongs to, if any.
node_type	str | None	Node type string, if any.
data	dict[str, Any]	Type-specific payload (token text, error, condition, etc.).

Source code in teff/stream.py

@dataclass
class StreamEvent:
    """A single event emitted while a graph streams.

    Attributes:
        type: Event type — ``run_start``, ``node_start``, ``node_end``,
            ``node_error``, ``edge``, ``token``, ``llm``, ``tool_call``,
            ``structured``, ``interrupt``, ``interrupt_resume``, ``checkpoint``,
            or ``run_end``.
        node_id: Graph node id the event belongs to, if any.
        node_type: Node type string, if any.
        data: Type-specific payload (token text, error, condition, etc.).
    """

    type: str
    node_id: str | None = None
    node_type: str | None = None
    data: dict[str, Any] = field(default_factory=dict)