Traces Overview - Prime Intellect Docs

Prime Traces is in closed beta.

Prime Traces stores the traces from your training runs, evaluations and inference jobs, and lets you list, fetch, search and delete them.

Trace: one rollout or request, with its messages, model calls, score and metadata.
Episode: a group of traces produced together, such as a multi-agent rollout.
Run: every trace that shares a run_id, with totals like trace count and average reward.

You need an API key with the Prime Traces permission. The SDK and CLI read it from PRIME_API_KEY or prime login. Set PRIME_TEAM_ID to work with a team’s traces.

Python SDK

pip install prime-traces

from prime_traces import TracesClient

client = TracesClient()

client.upload_file("traces.jsonl", context={"source": "my-eval"})

page = client.list(run_id="run_9f3k2m", reward_min=0.9)
for summary in page.items:
    print(summary.trace_id, summary.score.reward)

raw = client.get_raw(page.items[0].trace_id)
matches = client.search("Traceback", run_id="run_9f3k2m")

CLI

prime traces upload traces.jsonl
prime traces list --run-id run_9f3k2m
prime traces get <trace_id> --raw
prime traces search "Traceback" --run-id run_9f3k2m