> ## Documentation Index > Fetch the complete documentation index at: https://docs.trulayer.ai/llms.txt > Use this file to discover all available pages before exploring further. # Stream the spans of a single trace as CSV or JSONL (sync export) > Cursor-streamed synchronous export of every span belonging to the trace identified by `trace_id`. `trace_id` is **required** — the batch "all traces" export path was removed because it was too expensive to support at hypertable scale. Callers needing trace- level analytics should use `GET /v1/traces` for paginated listing and call this endpoint per trace. Rows are written directly to the response body — never buffered — and capped per plan (Starter: 100, Pro/Team/Business: 5000, Enterprise: unlimited). Hitting the cap sets `X-TruLayer-Truncated: true`; the payload is complete up to that row. CSV columns (in order): `trace_id, span_id, name, type, model, created_at, duration_ms, prompt_tokens, completion_tokens, cost_usd, trace_input, trace_output, span_input, span_output, error`. `span_id` is the unique identifier for the span (renamed from `id`). `trace_input` is the parent trace's input text joined onto every span row. `span_input` is the span's own input text. `trace_output` and `span_output` are the parent trace's and span's output payloads respectively, so a single download carries the full input/output context for both the trace and each span. JSONL emits one JSON object per line with the same keys. The Content-Disposition filename is `spans--YYYY-MM-DD.{csv,jsonl}`. ## OpenAPI ````yaml /api-reference/openapi.yaml get /v1/traces/export openapi: 3.1.0 info: title: TruLayer API version: 0.1.0 description: | Trace ingestion, evaluation, query, and feedback API for TruLayer AI. ### Authentication Two auth schemes coexist: SDK API keys (`Authorization: Bearer tl_...`) and Clerk session JWTs (dashboard). Routes marked "requires Clerk auth" in their summary are unreachable via API key. ### Roles (TRU-234) Dashboard requests carry one of three organization roles: `owner`, `member`, `viewer`. Endpoints enforce role allowlists, not rank comparisons. See `docs/security.md` for the full role-permission matrix. - **owner** — full control including billing, member management, destructive deletes on failure-rules/model-routes/webhooks/eval-rules, control-loop execution, and DLQ resolution. - **member** — read + write (ingest, feedback, evals, API keys, projects, failure-rules, model-routes, webhooks). - **viewer** — read-only dashboard access plus compliance read access to `/v1/audit-log` and `/v1/dlq`. servers: - url: https://api.trulayer.ai description: Production - url: http://localhost:8080 description: Local development security: - BearerAuth: [] tags: - name: health - name: ingest - name: traces - name: metrics - name: feedback - name: evals - name: eval-rules - name: control - name: apikeys - name: model-routes - name: datasets - name: search - name: anomaly - name: webhooks - name: failure-rules - name: eval-runs - name: ci - name: otlp - name: billing - name: deprecations - name: audit - name: failures - name: projects - name: compliance - name: dsr - name: policies - name: members paths: /v1/traces/export: get: tags: - traces - export summary: Stream the spans of a single trace as CSV or JSONL (sync export) description: | Cursor-streamed synchronous export of every span belonging to the trace identified by `trace_id`. `trace_id` is **required** — the batch "all traces" export path was removed because it was too expensive to support at hypertable scale. Callers needing trace- level analytics should use `GET /v1/traces` for paginated listing and call this endpoint per trace. Rows are written directly to the response body — never buffered — and capped per plan (Starter: 100, Pro/Team/Business: 5000, Enterprise: unlimited). Hitting the cap sets `X-TruLayer-Truncated: true`; the payload is complete up to that row. CSV columns (in order): `trace_id, span_id, name, type, model, created_at, duration_ms, prompt_tokens, completion_tokens, cost_usd, trace_input, trace_output, span_input, span_output, error`. `span_id` is the unique identifier for the span (renamed from `id`). `trace_input` is the parent trace's input text joined onto every span row. `span_input` is the span's own input text. `trace_output` and `span_output` are the parent trace's and span's output payloads respectively, so a single download carries the full input/output context for both the trace and each span. JSONL emits one JSON object per line with the same keys. The Content-Disposition filename is `spans--YYYY-MM-DD.{csv,jsonl}`. operationId: exportTraces parameters: - name: trace_id in: query required: true description: >- Trace to export spans for. Required — the endpoint returns 400 if absent. schema: type: string format: uuid - name: format in: query schema: type: string enum: - csv - jsonl default: csv - name: project_id in: query schema: type: string format: uuid responses: '200': description: Streamed CSV or JSONL payload. headers: X-TruLayer-Truncated: description: >- Present and `true` when the per-plan row cap was hit and more rows exist. schema: type: boolean Content-Disposition: description: >- Suggests a filename of `spans--YYYY-MM-DD.{csv,jsonl}`. schema: type: string content: text/csv: schema: type: string application/x-ndjson: schema: type: string '400': $ref: '#/components/responses/BadRequest' '401': $ref: '#/components/responses/Unauthorized' components: responses: BadRequest: description: Bad request content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' Unauthorized: description: Missing or invalid credentials content: application/json: schema: $ref: '#/components/schemas/ErrorResponse' schemas: ErrorResponse: type: object required: - error properties: error: type: string securitySchemes: BearerAuth: type: http scheme: bearer description: SDK API key (`tl_...`) or Clerk session JWT ````