generated: '2026-07-18'
method: searched
source: https://superpenguin.ai/docs
docs: https://superpenguin.ai/docs
authentication:
  style: api-key
  key_prefix: "sp-"
  env_var: SP_API_KEY
  ref: authentication/carrot-labs-authentication.yml
idempotency:
  supported: true
  mechanism: dedup-key
  scope: batch-ingestion
  detail: >-
    Batch cost-tracking calls (track_openai_batch(), track_anthropic_batch(),
    track_google_batch()) accept idempotent deduplication keys so re-submitting
    the same batch does not double-count spend. Individual wrapped calls are
    logged with per-request identity and cost metadata.
metadata:
  supported: true
  fields:
  - customer_id
  - feature
  - team
  - environment
  - prompt_key
  - prompt_version
  detail: >-
    Optional attribution metadata is attached per request to slice spend by
    customer, feature, team, environment, and prompt version.
capture:
  content_stored_by_default: false
  detail: >-
    By default only cost metadata (tokens, model, latency, tags) is captured;
    prompt and response text are not stored unless the org opts in (Pro/Enterprise)
    to sampled, encrypted, text-only capture.
performance:
  non_streaming_overhead: "<10ms"
  streaming_ttft_impact: "<5ms"
  telemetry: async
pagination:
  documented: false
rate_limiting:
  documented: false
versioning:
  scheme: semver
  detail: SDKs are versioned with semantic versioning (PyPI 0.11.x, npm 0.7.x).
error_envelope:
  documented: false
notes: >-
  SuperPenguin exposes its platform through first-party Python and TypeScript
  SDKs rather than a documented public REST API; no OpenAPI/Swagger specification
  is published, so pagination, rate-limit signaling, and an error envelope are
  not documented at the wire level.