Kitaru

checkpoint

Checkpoint decorator for durable work boundaries.

A checkpoint is a unit of work inside a flow whose outcome is persisted. Successful outputs become artifacts; failures are recorded for retry.

funccheckpoint(func=None, *, retries=0, type=None, runtime=None) -> _CheckpointDefinition | Callable[[Callable[..., Any]], _CheckpointDefinition]

Mark a function as a durable checkpoint.

Can be used as a bare decorator or with arguments:

from kitaru import checkpoint

@checkpoint
def my_step(): ...

@checkpoint(retries=3, type="llm_call")
def my_step(): ...

@checkpoint(runtime="isolated")
def heavy_step(): ...
paramfuncCallable[..., Any] | None
= None

Optional function for bare decorator use.

paramretriesint
= 0

Number of checkpoint-level retries on failure.

paramtypestr | None
= None

Checkpoint type for dashboard visualization.

paramruntimestr | None
= None

Execution runtime for this checkpoint. Accepts "inline" or "isolated" (case-insensitive). When set to "isolated", the checkpoint runs in its own container on remote orchestrators that support it. None (the default) lets the orchestrator decide.

Returns

_CheckpointDefinition | Callable[[Callable[..., Any]], _CheckpointDefinition]

The wrapped checkpoint object or a decorator that returns it.