Replay and resume

Artifact helpers change saved state, the native Agent API can continue a declared-resumable transcript, and replay either reuses recorded work or executes again.

Pause, load, and mark running

artifact_state.py

1from opentine import Run
2
3run.pause("checkpoint.tine")
4loaded = Run.load("checkpoint.tine")
5running = Run.resume("checkpoint.tine")
6
7# Run.resume changes artifact state; it does not invoke a model.

Continuation still requires compatible code, tools, model configuration, credentials, and a resumable manifest.

Continue a native run

native_resume.py

1from opentine import Agent, Run
2from opentine.models.anthropic import Anthropic
3
4saved = Run.load("checkpoint.tine")
5agent = Agent(model=Anthropic())
6continued = agent.resume_sync(saved, from_step=saved.steps[-1].id,
7                              prompt="Continue with the revised constraint.")
8continued.save("continued.tine")

Replay modes

Terminal

tine replay result.tine --mode cache --save replayed.tine
tine replay result.tine --inspect
tine replay result.tine --inspect --from-step 3
tine replay result.tine --mode rerun --harness codex --save rerun.tine

Cache replay makes no live calls and uses a deterministic fork identity, so the same cached replay remains idempotent. Rerun executes a fresh run with a new run ID and may differ. The CLI requires an explicit harness for rerun; native rerun is available through Agent.replay().

See the v0.4 upgrade guide before relying on predicted fork or replay IDs.