> ## Documentation Index > Fetch the complete documentation index at: https://docs.audivo.dev/llms.txt > Use this file to discover all available pages before exploring further. # Direct cached fetch of an episode's transcript > Fetches a compatible cached transcript for a canonical episode directly, without submitting a job. When no compatible cache object exists, returns an `EpisodeQuoteHint` — the same quote figures a dry run reports, but *without* `dry_run`, because this request was not a dry run — rather than creating a job. Call `POST /v1/transcripts` to actually create one. **Delivery format.** `format: json` (or omitted) returns `application/json`. `format: text|srt|vtt|md` returns the raw artifact under its own media type (`text/plain`, `application/x-subrip`, `text/vtt`, `text/markdown`), with provenance in the `X-Transcript-Episode-Id`, `X-Transcript-Source`, and `X-Transcript-Timing-Precision` headers and the charge in `X-Credits-Charged`. **The single content-type switch:** a payload over the ~5 MB inline limit returns `200 application/json` with `{ transcript_url, expires_at }` in *every* format. A quote hint — no cached transcript exists to render — is always `application/json`, whatever `format` asked for. ## OpenAPI ````yaml /openapi.yaml get /v1/episodes/{episode_id}/transcript openapi: 3.1.0 info: title: Audivo API version: 0.11.0 description: >- The Audivo API turns a public podcast episode into a clean, timed transcript. Name an episode by an Apple Podcasts link, by its RSS feed URL and GUID, or by the episode id an earlier answer gave you; get back structured, timed text — at once for an episode already transcribed, or as a job you poll for fresh audio. Two authentication schemes, and which an operation takes is part of what it is: `apiKeyBearer` — an `hk_live_` key sent as `Authorization: Bearer` — authenticates the public API, and `dashboardJwt` authenticates the dashboard's own operations. No operation that takes one accepts the other in its place. An API-key caller reads its balance and reservations from any quote response. Deferred on purpose, and absent from this document rather than hidden: webhooks, subscriptions, speaker labels beyond the frozen `diarized: false` shape, custom vocabulary, caller-selected engines, raw-audio URLs and uploads, SDKs, translation, and cross-corpus search. What changed in each version is on the documentation site's changelog. license: name: Proprietary url: https://docs.audivo.dev/legal/terms servers: - url: https://k8mfogcvz4.execute-api.us-east-1.amazonaws.com/prod description: Production API. Append the operation path, including /v1. security: [] tags: - name: Transcripts description: Create, poll, retrieve, and cancel transcription jobs. - name: Catalog description: >- Public podcast show/episode search, charts, detail, and resolution. Every metadata/search/list call in this tag costs 1 credit; no catalog response echoes `credits_charged`, so the charge is stated here rather than per-response. Not yet applied: the discovery operations shipped in 0.2.0 (`GET /v1/search/shows`, `GET /v1/charts`) are metered by request rate only and write no credit-bearing row; the per-call charge lands with the catalog reads that bill. - name: Quotes description: >- Price a selection before spending. A quote resolves every episode in it, records what it resolved under a short TTL, and reserves nothing; the confirm that follows is the operation that spends, and it answers with a job group. - name: Groups description: >- The job group: what one confirm created. A group holds N ≥ 1 members, each a transcription job or a settled cache read; a single-episode `POST /v1/transcripts` is a group of one. A caller reads a group's rollup, lists their own groups newest first, and cancels a group, which cancels every member not yet started and releases each reservation exactly once. - name: Session description: >- Dashboard-JWT session establishment. One operation, called by the dashboard's backend-for-frontend whenever it turns a Cognito sign-in into a Audivo session, and the one place a caller who does not yet have an account is served. - name: Account description: Usage, limits, and supported languages for the calling account. - name: API Keys description: Dashboard-JWT key lifecycle management. - name: Billing description: Dashboard-JWT Stripe Checkout, Customer Portal, invoices, and refunds. - name: Data description: Dashboard-JWT portable export and account deletion. paths: /v1/episodes/{episode_id}/transcript: get: tags: - Transcripts summary: Direct cached fetch of an episode's transcript description: >- Fetches a compatible cached transcript for a canonical episode directly, without submitting a job. When no compatible cache object exists, returns an `EpisodeQuoteHint` — the same quote figures a dry run reports, but *without* `dry_run`, because this request was not a dry run — rather than creating a job. Call `POST /v1/transcripts` to actually create one. **Delivery format.** `format: json` (or omitted) returns `application/json`. `format: text|srt|vtt|md` returns the raw artifact under its own media type (`text/plain`, `application/x-subrip`, `text/vtt`, `text/markdown`), with provenance in the `X-Transcript-Episode-Id`, `X-Transcript-Source`, and `X-Transcript-Timing-Precision` headers and the charge in `X-Credits-Charged`. **The single content-type switch:** a payload over the ~5 MB inline limit returns `200 application/json` with `{ transcript_url, expires_at }` in *every* format. A quote hint — no cached transcript exists to render — is always `application/json`, whatever `format` asked for. operationId: getEpisodeTranscript parameters: - $ref: '#/components/parameters/EpisodeIdPathParam' - $ref: '#/components/parameters/FormatQueryParam' responses: '200': description: A cached transcript read, or a quote hint if none exists yet. headers: X-Request-Id: $ref: '#/components/headers/XRequestId' X-RateLimit-Limit: $ref: '#/components/headers/XRateLimitLimit' X-Transcript-Episode-Id: $ref: '#/components/headers/XTranscriptEpisodeId' X-Transcript-Source: $ref: '#/components/headers/XTranscriptSource' X-Transcript-Timing-Precision: $ref: '#/components/headers/XTranscriptTimingPrecision' X-Credits-Charged: $ref: '#/components/headers/XCreditsCharged' content: application/json: schema: oneOf: - $ref: '#/components/schemas/TranscriptRead' - $ref: '#/components/schemas/EpisodeQuoteHint' - $ref: '#/components/schemas/TranscriptUrlRef' examples: cacheHit: $ref: '#/components/examples/CacheHitTranscript' noCacheYet: $ref: '#/components/examples/NoCacheQuoteHint' oversizedArtifact: $ref: '#/components/examples/OversizedArtifactRef' text/plain: schema: $ref: '#/components/schemas/RawTranscriptArtifact' application/x-subrip: schema: $ref: '#/components/schemas/RawTranscriptArtifact' examples: srt: $ref: '#/components/examples/SrtArtifact' text/vtt: schema: $ref: '#/components/schemas/RawTranscriptArtifact' text/markdown: schema: $ref: '#/components/schemas/RawTranscriptArtifact' '400': $ref: '#/components/responses/BadRequestApiKey' '401': $ref: '#/components/responses/UnauthorizedApiKey' '402': $ref: '#/components/responses/PaymentRequiredApiKey' '404': $ref: '#/components/responses/NotFoundApiKey' '409': $ref: '#/components/responses/ConflictApiKeyAccountState' '422': $ref: '#/components/responses/UnprocessableApiKey' '429': $ref: '#/components/responses/TooManyRequestsApiKey' '500': $ref: '#/components/responses/InternalErrorApiKey' security: - apiKeyBearer: [] components: parameters: EpisodeIdPathParam: name: episode_id in: path required: true schema: $ref: '#/components/schemas/EpisodeId' FormatQueryParam: name: format in: query required: false description: >- The derived output to return once the transcript is available, and with it the response media type: `json` → `application/json`, `text` → `text/plain`, `srt` → `application/x-subrip`, `vtt` → `text/vtt`, `md` → `text/markdown`. An oversized payload is `application/json` in every format — the one documented content-type switch, described on each operation. Delivering content is the whole point of the operations that use this parameter, so omitting it defaults to `json` rather than withholding the transcript (contrast `PollFormatQueryParam`, which has no default). Delivery-only: never affects the cache key or the produced transcript content. schema: $ref: '#/components/schemas/TranscriptFormat' default: json headers: XRequestId: description: >- Echoed on every response, success or error. A refusal generated at the edge (401, 429, and the unmetered-key 500) carries API Gateway's own hyphenated request id after the `req_` prefix; a response from a function carries the same id without hyphens. schema: type: string pattern: ^req_[A-Za-z0-9-]+$ maxLength: 100 XRateLimitLimit: description: >- The caller's requests-per-minute ceiling for the current tier. Present on every response for a resolved caller, success and error alike; absent from `401 unauthenticated`, which has no account to report a ceiling for, and from the edge's own `429`, which is emitted before any function runs. schema: type: integer minimum: 0 XTranscriptEpisodeId: description: >- The canonical `episode_id` of the delivered transcript, so a raw artifact is not provenance-free. Absent on responses that carry no transcript. schema: $ref: '#/components/schemas/EpisodeId' XTranscriptSource: description: >- The delivered transcript's `source` — the pinned ASR engine, or `publisher` for passthrough. Absent on responses that carry no transcript. schema: $ref: '#/components/schemas/TranscriptSource' XTranscriptTimingPrecision: description: >- The delivered transcript's `timing_precision` (`word | segment | none`). every transcript exposes it, so it travels with every delivery format — a raw SRT body reports the same precision the canonical JSON would. Absent on responses that carry no transcript. schema: $ref: '#/components/schemas/TimingPrecision' XCreditsCharged: description: >- Credits charged for this read, mirroring the `credits_charged` field of the JSON envelope so a raw-artifact delivery does not silently drop the billing figure. Absent on responses that carry no charge. schema: $ref: '#/components/schemas/CreditAmount' RetryAfter: description: >- Seconds to wait before retrying. Sent with the edge's `rate_limited` throttle refusal. Not sent with `concurrency_limited`, which clears when one of the account's own jobs reaches a terminal state — a moment the API cannot put a number on. schema: type: integer minimum: 0 schemas: TranscriptRead: type: object additionalProperties: false required: - format - is_cached - credits_charged description: >- A synchronous JSON transcript read: a cache hit or publisher passthrough (`POST /v1/transcripts`), a direct cached fetch (`GET /v1/episodes/{episode_id}/transcript`), or the re-delivery of a read that was already paid for (`GET /v1/reads/{read_id}`). `credits_charged` is what *this* request charged, which is why it is `0` on the last of the three and the cached-read price for this account on the other two. `format` is always `json` here — a `text|srt|vtt|md` read of the same transcript is delivered as a raw body with `X-Credits-Charged` carrying this envelope's `credits_charged`. See the comment above for why this shape is self-contained rather than composed via `allOf`. properties: format: const: json is_cached: type: boolean credits_charged: $ref: '#/components/schemas/CreditAmount' transcript: $ref: '#/components/schemas/CanonicalTranscript' transcript_url: type: string format: uri maxLength: 2048 expires_at: type: string format: date-time oneOf: - required: - transcript - required: - transcript_url - expires_at EpisodeQuoteHint: type: object additionalProperties: false required: - episode_id - is_cached - estimated_credits - quote_ceiling_credits - quote_basis - estimated_seconds description: >- The "no compatible cache object exists yet" response of `GET /v1/episodes/{episode_id}/transcript`: the same quote figures as `TranscriptQuote`, minus `dry_run`. It is a separate schema rather than a reuse of `TranscriptQuote` because that schema pins `dry_run: const true`, and this request was not a dry run — a response asserting otherwise would be a lie the generated SDK and conformance suite would both propagate. Like a dry run it creates no job and no reservation; call `POST /v1/transcripts` for that. The two shapes stay disjoint (`additionalProperties: false` on both, one requiring `dry_run` and the other forbidding it), so a `oneOf` between them is unambiguous. properties: episode_id: $ref: '#/components/schemas/EpisodeId' is_cached: type: boolean estimated_credits: $ref: '#/components/schemas/CreditAmount' quote_ceiling_credits: $ref: '#/components/schemas/CreditAmount' description: estimate + 25%, rounded up. quote_basis: $ref: '#/components/schemas/QuoteBasis' estimated_seconds: $ref: '#/components/schemas/EstimatedProcessingSeconds' TranscriptUrlRef: type: object additionalProperties: false required: - transcript_url - expires_at description: >- The oversized-payload response — the single content-type switch in this API. A transcript above the ~5 MB inline limit is never sent inline in any format; instead the response is `200 application/json` with a presigned URL (24 h) and its expiry, even when `format` asked for `text`, `srt`, `vtt`, or `md`. For `format=json` the media type does not change, and the same two fields appear inside the JSON envelope (`TranscriptRead.transcript_url` / `TranscriptArtifact.transcript_url`) in place of the inline transcript — the behavior is identical in every format; only the raw formats also change media type. properties: transcript_url: type: string format: uri maxLength: 2048 expires_at: type: string format: date-time RawTranscriptArtifact: type: string maxLength: 5000000 description: >- A derived transcript artifact delivered as a raw body under its own media type: `text/plain` (`format=text`), `application/x-subrip` (`srt`), `text/vtt` (`vtt`), or `text/markdown` (`md`). Provenance travels in the `X-Transcript-Episode-Id`, `X-Transcript-Source`, and `X-Transcript-Timing-Precision` response headers, because a raw body has no field to carry it. Bodies above the ~5 MB inline limit are not sent here at all: that response switches to `application/json` carrying a `TranscriptUrlRef`. EpisodeId: type: string pattern: ^ep_[a-z2-7]{16}$ description: Canonical episode identifier. TranscriptFormat: type: string enum: - json - text - srt - vtt - md description: >- Delivery-only: never affects the cache key or the produced transcript content. Every format is derived at read time from the one canonical transcript. Response media type per format: `json` → `application/json`, `text` → `text/plain`, `srt` → `application/x-subrip`, `vtt` → `text/vtt`, `md` → `text/markdown`. The one exception is an oversized payload, which is `application/json` in every format (see `TranscriptUrlRef`). TranscriptSource: type: string enum: - qwen3-asr - whisper-large-v3-turbo - whisper-large-v3 - publisher description: >- The ASR engine that produced this transcript, or `publisher` for passthrough. TimingPrecision: type: string enum: - word - segment - none description: >- How precisely segments/words are timed. ASR output targets `word`. Publisher-sourced transcripts report the precision the source actually provides and never claim word timing they do not have: plain text is `none`, segment-timed sources are `segment`, and only genuine word-timed sources are `word`. CreditAmount: type: integer minimum: 0 maximum: 9007199254740991 description: >- A non-negative JavaScript-safe integer credit amount. The upper bound matches `Number.MAX_SAFE_INTEGER`, which is enforced by the domain ledger so arithmetic and JSON round-trips cannot silently lose cents worth of credit precision. CanonicalTranscript: type: object additionalProperties: false required: - episode_id - show_id - language - duration_sec - source - source_revision - model_version - pipeline_version - timing_precision - diarized - warnings - segments - created_at properties: episode_id: $ref: '#/components/schemas/EpisodeId' show_id: $ref: '#/components/schemas/ShowId' language: $ref: '#/components/schemas/Language' description: Resolved BCP-47 language code. duration_sec: type: number minimum: 0 maximum: 129600 source: $ref: '#/components/schemas/TranscriptSource' source_revision: type: string minLength: 1 maxLength: 150 description: >- e.g. `sha256:...` — the asset revision this transcript was produced from. model_version: type: string minLength: 1 maxLength: 100 description: Provider-pinned model version, or `publisher` for passthrough. pipeline_version: type: string minLength: 1 maxLength: 60 timing_precision: $ref: '#/components/schemas/TimingPrecision' diarized: const: false description: Always `false` in the MVP; diarization is post-MVP. warnings: type: array maxItems: 1000 items: $ref: '#/components/schemas/TranscriptWarning' segments: type: array maxItems: 200000 items: $ref: '#/components/schemas/TranscriptSegment' created_at: type: string format: date-time QuoteBasis: type: string enum: - feed_metadata - probed description: >- `feed_metadata` when the estimate is derived from feed-published duration; `probed` when a bounded enclosure probe measured it. EstimatedProcessingSeconds: type: integer minimum: 0 maximum: 72000 description: >- Estimated **wall-clock** seconds until the job reaches a terminal state — intended for choosing a poll interval. This is *not* audio duration: audio duration has its own field (`duration_sec` on the transcript, `duration_sec` on `EpisodeSummary`), and conflating the two silently misprices everything a caller derives from it. The committed examples only make sense under this reading: `estimated_seconds: 420` alongside `estimated_credits: 268` cannot be audio seconds, because 7 audio minutes bills 7 credits at `CREDITS_PER_AUDIO_MINUTE`, while 268 credits is about 4.5 hours of audio. The bound is derived from that meaning rather than copied from the audio cap: twice the published 10-hour `max_audio_duration_seconds` ceiling, an end-to-end envelope that still holds at a pathological 1× realtime factor. The real hard per-job timeout is duration-scaled and much shorter. Error: type: object additionalProperties: false required: - error properties: error: $ref: '#/components/schemas/ErrorDetail' ShowId: type: string pattern: ^sh_[a-z2-7]{16}$ description: Canonical show identifier. Language: type: string pattern: ^[A-Za-z]{2,3}(-[A-Za-z0-9]{1,8})*$ maxLength: 35 description: >- A BCP-47 language tag (general syntax check, not full IANA subtag validation). TranscriptWarning: type: object additionalProperties: false required: - segment - type - detail description: >- A quality warning from the pipeline's own checks, attached to the segment it concerns. Presented as a heuristic, not an accuracy guarantee; the transcript is delivered and charged as usual. `type` is an open string rather than a closed enum; the documentation's output-formats page lists the values in use. properties: segment: type: integer minimum: 0 maximum: 1000000 type: type: string minLength: 1 maxLength: 60 detail: type: string minLength: 1 maxLength: 500 TranscriptSegment: type: object additionalProperties: false required: - id - start - end - speaker - text properties: id: type: integer minimum: 0 maximum: 1000000 start: type: number minimum: 0 maximum: 129600 description: Start time in seconds. end: type: number minimum: 0 maximum: 129600 description: End time in seconds. speaker: enum: - null description: >- Always `null` in the MVP — diarization is post-MVP. A future diarized-speaker field is a deliberate, reviewed type change, not a silent widening. text: type: string maxLength: 20000 words: type: array maxItems: 20000 items: $ref: '#/components/schemas/TranscriptWord' description: Present when `timing_precision` is `word`; absent otherwise. ErrorDetail: type: object additionalProperties: false required: - type - code - message - doc_url - request_id - retryable properties: type: $ref: '#/components/schemas/ErrorType' code: $ref: '#/components/schemas/ErrorCode' message: type: string minLength: 1 maxLength: 1000 doc_url: type: string format: uri maxLength: 300 request_id: $ref: '#/components/schemas/RequestId' retryable: type: boolean oneOf: - required: - code - type properties: code: enum: - invalid_request - invalid_url - source_not_supported - unsafe_source type: const: invalid_request - required: - code - type properties: code: enum: - duration_exceeded - size_exceeded - unsupported_codec - unsupported_language - credits_not_refundable type: const: unprocessable_input - required: - code - type properties: code: const: unauthenticated type: const: unauthenticated - required: - code - type properties: code: enum: - idempotency_conflict - job_not_completed - quote_expired - quote_mismatch - quote_unverified - expected_total_mismatch - request_in_progress - account_suspended - account_closed - account_not_found - api_key_limit_reached - tier_unchanged type: const: conflict - required: - code - type properties: code: enum: - feed_dead - episode_not_found - show_not_found - job_not_found - quote_not_found - group_not_found - api_key_not_found - credit_lot_not_found type: const: not_found - required: - code - type properties: code: const: content_blocked type: const: content_blocked - required: - code - type properties: code: const: payment_required type: const: payment_required - required: - code - type properties: code: enum: - rate_limited - concurrency_limited type: const: rate_limited - required: - code - type properties: code: enum: - engine_unavailable - processing_failed - discovery_unavailable - internal_error type: const: unavailable - required: - code - type properties: code: const: nothing_to_quote type: const: unprocessable_input TranscriptWord: type: object additionalProperties: false required: - w - s - e properties: w: type: string minLength: 1 maxLength: 200 description: The word text. s: type: number minimum: 0 maximum: 129600 description: Start time in seconds. e: type: number minimum: 0 maximum: 129600 description: End time in seconds. ErrorType: type: string enum: - invalid_request - unprocessable_input - unauthenticated - conflict - not_found - content_blocked - payment_required - rate_limited - unavailable description: >- The family an error belongs to, for handling; `code` is the precise reason. `invalid_request`: something in the request is wrong — fix the integration. `unprocessable_input`: the request is well-formed but the episode or resource is outside published support. `unauthenticated`: missing, malformed, revoked or expired credential. `payment_required`: the account's credits cannot cover the ceiling. `conflict`: the resource's or the account's state refuses this request as sent — a lapsed quote, a reused idempotency key, a held account. `not_found`: the job, quote, group or key the request named does not exist for this account. `rate_limited`: over a plan limit; wait and retry. `unavailable`: a failure on our side or at a provider; retry, and quote `request_id` if it persists. ErrorCode: type: string enum: - invalid_request - invalid_url - unauthenticated - idempotency_conflict - job_not_completed - source_not_supported - feed_dead - episode_not_found - show_not_found - unsafe_source - unsupported_codec - unsupported_language - duration_exceeded - size_exceeded - content_blocked - payment_required - rate_limited - concurrency_limited - job_not_found - api_key_not_found - api_key_limit_reached - credit_lot_not_found - credits_not_refundable - engine_unavailable - processing_failed - discovery_unavailable - nothing_to_quote - internal_error - quote_not_found - group_not_found - quote_expired - quote_mismatch - quote_unverified - expected_total_mismatch - request_in_progress - account_suspended - account_closed - account_not_found - tier_unchanged description: >- Every code the API returns, and the HTTP status it comes with. `invalid_request`, `invalid_url` → 400. `unauthenticated` → 401. `payment_required` → 402. `job_not_found`, `quote_not_found`, `group_not_found`, `api_key_not_found`, `credit_lot_not_found` → 404. `idempotency_conflict`, `request_in_progress`, `job_not_completed`, `quote_expired`, `quote_mismatch`, `quote_unverified`, `expected_total_mismatch`, `account_suspended`, `account_closed`, `account_not_found`, `api_key_limit_reached`, `tier_unchanged` → 409. `source_not_supported`, `feed_dead`, `episode_not_found`, `show_not_found`, `unsafe_source`, `unsupported_codec`, `unsupported_language`, `duration_exceeded`, `size_exceeded`, `nothing_to_quote`, `credits_not_refundable` → 422. `content_blocked` → 451. `rate_limited`, `concurrency_limited` → 429. `internal_error`, `processing_failed` → 500. `discovery_unavailable`, `engine_unavailable` → 503. `processing_failed` also appears inside a failed job's `error` field rather than as a live status. Each code has a heading on the documentation site's errors page, which is what `doc_url` links to. RequestId: type: string pattern: ^req_[A-Za-z0-9-]+$ maxLength: 100 examples: CacheHitTranscript: summary: Cache hit (charges 1 credit) value: format: json is_cached: true credits_charged: 1 transcript: episode_id: ep_a3urk6fzq2mvbxyc show_id: sh_k4n7qzvw2mtxyabc language: en duration_sec: 3841 source: qwen3-asr source_revision: sha256:8f14e45fceea167a5a36dedd4bea2543deca1c7c8b3c7e2b model_version: qwen3-asr-1.7b pipeline_version: 2026.08-vad4-fw1.2 timing_precision: word diarized: false warnings: - segment: 41 type: repetition_loop detail: phrase repeated 14x segments: - id: 0 start: 0 end: 6.48 speaker: null text: Welcome back to the show… words: - w: Welcome s: 0 e: 0.31 created_at: '2026-08-20T09:00:00Z' NoCacheQuoteHint: summary: Nothing cached yet — quote hint, and no `dry_run` claim value: episode_id: ep_hbrpoigf3cbfnobm is_cached: false estimated_credits: 268 quote_ceiling_credits: 335 quote_basis: probed estimated_seconds: 420 OversizedArtifactRef: summary: Oversized payload — the single content-type switch description: >- A transcript above the ~5 MB inline limit is returned as `200 application/json` carrying a presigned URL and its expiry, in *every* `format` — so a caller that asked for `srt` receives JSON here rather than `application/x-subrip`. value: transcript_url: https://cdn.hark.dev/transcripts/ep_f4tcqk2zmvbxyna7.srt?sig=abc expires_at: '2026-08-21T09:14:22Z' SrtArtifact: summary: '`?format=srt` on a completed transcript — a raw SRT body' description: >- Served as `application/x-subrip`, not as a JSON-wrapped string. Provenance travels alongside it in `X-Transcript-Episode-Id`, `X-Transcript-Source`, and `X-Transcript-Timing-Precision`. value: | 1 00:00:00,000 --> 00:00:06,480 Welcome back to the show… 2 00:00:06,480 --> 00:00:12,500 Today we are talking about podcasts. responses: BadRequestApiKey: description: >- Invalid request shape or an unrecognized pointer. Codes: `invalid_request`, `invalid_url`. headers: X-Request-Id: $ref: '#/components/headers/XRequestId' X-RateLimit-Limit: $ref: '#/components/headers/XRateLimitLimit' content: application/json: schema: $ref: '#/components/schemas/Error' UnauthorizedApiKey: description: >- Code `unauthenticated` — the API key is missing, malformed, revoked, or does not exist. Every operation in this document is authenticated, so this is the most common error path a client will hit; it is declared on all of them rather than left implicit. The response never distinguishes "no such key" from "revoked key". Unlike every other API-key response it carries no `X-RateLimit-Limit`: that is a per-account figure, and an unauthenticated request has no resolved account to report one for. headers: X-Request-Id: $ref: '#/components/headers/XRequestId' content: application/json: schema: $ref: '#/components/schemas/Error' PaymentRequiredApiKey: description: >- Code `payment_required` — available account credits cannot cover the quote ceiling. headers: X-Request-Id: $ref: '#/components/headers/XRequestId' X-RateLimit-Limit: $ref: '#/components/headers/XRateLimitLimit' content: application/json: schema: $ref: '#/components/schemas/Error' NotFoundApiKey: description: >- Codes: `feed_dead`, `episode_not_found`, `show_not_found`, or `job_not_found` (the job does not exist or belongs to another account — cross-account access is indistinguishable from nonexistence). headers: X-Request-Id: $ref: '#/components/headers/XRequestId' X-RateLimit-Limit: $ref: '#/components/headers/XRateLimitLimit' content: application/json: schema: $ref: '#/components/schemas/Error' ConflictApiKeyAccountState: description: >- The account behind the key may not spend, so the charge this operation would make was refused and nothing was written. Codes: `account_suspended` (a billing or abuse hold — settle it or contact support), `account_closed` (the account is being deleted, or is deleted), `account_not_found` (no account row behind this credential; sign in to the dashboard, which creates it). None is retryable: the same request meets the same account. This operation's only `409`, and it appears here because the read *charges* on a cache hit. A delivery this account has already paid for is not refused — `GET /v1/reads/{read_id}` and the job poll take no money and answer a suspended account exactly as they answer an active one. headers: X-Request-Id: $ref: '#/components/headers/XRequestId' X-RateLimit-Limit: $ref: '#/components/headers/XRateLimitLimit' content: application/json: schema: $ref: '#/components/schemas/Error' UnprocessableApiKey: description: >- The request was well-formed but cannot be fulfilled. Codes: `source_not_supported`, `feed_dead`, `episode_not_found`, `show_not_found`, `unsafe_source`, `unsupported_codec`, `unsupported_language`, `duration_exceeded`, `size_exceeded`. These carry three different `error.type` values, not two: `invalid_request` for `source_not_supported`/`unsafe_source` (fix your integration), `unprocessable_input` for `unsupported_codec`/`unsupported_language`/`duration_exceeded`/ `size_exceeded` (this episode is outside published launch support), and `not_found` for `feed_dead`/`episode_not_found`/`show_not_found` (resolution failed with a typed catalog reason — the same three codes `NotFoundApiKey` carries at 404 for operations where the same failure is a lookup miss instead of an unprocessable request). A client branching on `type` must handle all three, not assume every code documented here is one of two things. headers: X-Request-Id: $ref: '#/components/headers/XRequestId' X-RateLimit-Limit: $ref: '#/components/headers/XRateLimitLimit' content: application/json: schema: $ref: '#/components/schemas/Error' TooManyRequestsApiKey: description: >- Codes `rate_limited` or `concurrency_limited`, from two different limits and two different places. `rate_limited` is the per-account request rate, refused at the edge before the operation runs: it carries `Retry-After` and no `X-RateLimit-Limit`, because nothing that knows the caller's tier has run yet. `concurrency_limited` is the cap on jobs open at once, refused by the operation itself: it carries `X-RateLimit-Limit` like every other answered response, and no `Retry-After`, because it clears when one of the account's own jobs finishes rather than after a fixed wait. headers: X-Request-Id: $ref: '#/components/headers/XRequestId' X-RateLimit-Limit: $ref: '#/components/headers/XRateLimitLimit' Retry-After: $ref: '#/components/headers/RetryAfter' content: application/json: schema: $ref: '#/components/schemas/Error' InternalErrorApiKey: description: >- Code `internal_error` — a failure on our side that no other code describes. The message is fixed text and says nothing about the cause; `request_id` is what identifies it in our logs. Retryable. headers: X-Request-Id: $ref: '#/components/headers/XRequestId' content: application/json: schema: $ref: '#/components/schemas/Error' securitySchemes: apiKeyBearer: type: http scheme: bearer bearerFormat: hk_live_... or hk_test_... description: >- `Authorization: Bearer hk_live_...` for live keys or `Authorization: Bearer hk_test_...` for test-mode keys. `hk_test_` keys resolve real public catalog metadata but return deterministic committed fixtures, never call inference, and never mutate live credits. This is the *only* transport for the credential: the `x-api-key` alias once documented was removed in 0.2.0, because the edge authorizer reads `Authorization` as its single identity source and a request on any other header is refused before it is authenticated. Never accepted on a `dashboardJwt` operation, and there are no exceptions. `GET /v1/usage` and `GET /v1/limits` briefly declared both schemes (0.8.0); that was withdrawn in 0.8.1 because no deployed route could honor it — both operations are served by the control-plane API, whose authorizer verifies a Cognito token and refuses an `hk_live_` credential on shape, and the customer API does not route either path. An API-key holder reads its balance and reservation from `QuoteResponse`, which carries `balance_credits` and `reserved_credits` on every quote. Every operation in this document takes one scheme or the other and refuses the wrong one as `unauthenticated`. ````