{"service":"transcription","version":"1.0.0-staged","mount":"/toll/transcribe","description":"Agent-native audio transcription. Upload audio bytes; the free preflight parses the duration locally and quotes the exact charge before any payment; the paid call transcribes via Groq Whisper and returns text, detected language, billed duration and segment timestamps when the upstream reports them.","trust":"You only pay for delivered answers","endpoints":{"POST /toll/transcribe":"paid — $0.005 per minute of audio, per-second prorated (see accounting_rule); exact amount computed per request from the parsed duration","POST /toll/transcribe/preflight":"free twin — same multipart body; full validation + exact quote + fingerprint, no payment, no transcription","GET /toll/transcribe/health":"free, counts-only","GET /toll/transcribe/catalog":"free, counts-only","GET /toll/transcribe/schema":"free — request/response shapes, error catalogue, accounting rule"},"pricing":{"usd_per_minute":0.005,"per_minute_units":{"usdc_units":5000,"sol_lamports":34000},"accounting_rule":"You pay for the seconds of audio you submit, prorated from $0.005 per minute. Your parsed duration is rounded UP to the next whole second; the charge is that many seconds at $0.005/60 per second, rounded UP to the next whole currency unit (USDC micro-unit or lamport) because currency units are integers. The free preflight states your exact billed seconds and exact charge before you pay. Nothing else is rounded, ever.","note":"No flat per-call price exists for this service; every charge is computed from the parsed duration and quoted exactly by the free preflight."},"formats_accepted":["wav (PCM)","mp3 (Layer III)","ogg-opus","ogg-vorbis","flac","mp4/m4a (audio-only)"],"formats_refused":["webm/matroska — EBML Duration is optional or a placeholder in much real-world WebM, so it cannot back a charge fixed at signature time (DESIGN D3)"],"caps":{"max_body_bytes":26000000,"max_audio_bytes":25000000,"max_duration_seconds":600,"max_charge_usd":0.05,"max_charge_usdc_units":50000,"max_charge_sol_lamports":340000},"models":{"default":"whisper-large-v3","allowlist":["whisper-large-v3"]},"upstream_disclosure":"Audio submitted here is transmitted to Groq for transcription by Whisper; that transmission IS the service. This service makes NO claim about the upstream's own retention or training posture; callers who need that answer should consult Groq's own terms.","privacy":"Audio is processed, not retained — never stored beyond the request. Transcripts are returned to the caller only and never stored. Records carry fingerprint + counts, never audio or transcript content."}