Recorder
Anthropic API
Wrap each messages.create call. Cache reads and writes are counted correctly.
- Copy
harness_recorder.pyinto your project. - Wrap the shared function that calls the model, as below.
Wrap a call
from harness_recorder import Recorder, usage_from
run = Recorder("support-triage", "2.1.0").start_run(ticket_id)
with run.agent("draft_reply", prompt_ref="prompts/drafter@6", model="anthropic/claude-sonnet-5") as step:
resp = client.messages.create(model="claude-sonnet-5", max_tokens=1200, messages=messages)
step.model_call(**usage_from(resp))
run.finish("completed")Where the log lands
Wherever your service logs go, or a folder.
How it reaches TorqueSight
You download the logs with your own access.