Agno now checkpoints runs at the tool-batch level and exposes a unified /continue endpoint that handles both regenerating a run and forking it, along with support for forking sessions. Together they let you branch off a run at a known-good point rather than starting over, which is the kind of control that makes long, expensive agent runs practical to operate.
See the checkpointing cookbooks for examples.