Traces

client.traces — upload trace files (JSON/JSONL, ≤50 MB), retrieve, list, delete.

Upload

Trace files are JSON or JSONL, max 50 MB. The SDK handles presigned-S3 upload automatically (3-step: presigned URL → S3 PUT → trace record creation).

result = client.traces.upload("./my_traces.json")
print(f"Uploaded {len(result.trace_ids)} traces: {result.trace_ids}")

Get one

trace = client.traces.get(trace_id)
# Trace fields: id, organization_id, project_id, created_at, filename,
# data (Dict[str, Any]), input, integration_id

List with filters

response = client.traces.get_many(
 page=1, page_size=50,
 source="...", # filter by source
 judge_id="...", # traces evaluated by this judge
 status="...",
 time_range={"from": "...", "to": "..."},
 search="...",
 sort_by="created_at",
 sort_order="desc",
)
# response.traces,.total_count,.count

Delete

client.traces.delete(trace_id)

List sources

sources = client.traces.get_sources()

Trace shape

The data field is a flexible Dict[str, Any] — Stratix doesn't enforce a specific span schema. For agent traces, common conventions:

inputs (top-level) — request context
outputs (top-level) — final response
spans (list) — child operations (LLM calls, tool calls, retrieval)

For auto-instrumented traces emitted by layerlens.instrument, the spans follow the canonical event types: model.invoke, cost.record, tool.call, policy.violation. See Integrations.

File formats

.json and .jsonl only. For larger datasets, split into chunks ≤50 MB each.

Async

All methods have AsyncStratix equivalents.

Good night