API Reference

The RealtimeVoiceKIT API gives you speech-to-text, realtime streaming transcription, voice agents, and speech intelligence over plain HTTPS and WebSockets. This page covers the conventions every endpoint shares. The pages in the sidebar document each endpoint in full, with request parameters, response fields, and runnable examples.

All REST endpoints live under a single base URL. WebSocket endpoints use the same host over wss://.

https://api.realtimevoicekit.com/v2

A machine-readable schema of every endpoint is available at GET /v2/openapi.json. It is the same contract this reference documents, filtered to the public surface.

Authenticate with an API key in the Authorization header using the Bearer scheme. Keys start with rtvk_ and are created in your workspace under Settings, API keys. Every endpoint requires a key except GET /v2/models and GET /v2/pricing, which are public.

curl https://api.realtimevoicekit.com/v2/credits \
  -H "Authorization: Bearer rtvk_your_api_key"

A missing, malformed, or revoked key returns 401 with the error code unauthorized.

• Request bodies are JSON. Send Content-Type: application/json on POST requests with a body.
• Responses are JSON, with two exceptions: subtitle exports return plain text (application/x-subrip for SRT, text/vtt for VTT), and media files are uploaded with a direct PUT to a signed storage URL using the file's own content type.
• Timestamps in responses are ISO 8601 strings in UTC. Word and segment timings (start, end) are integer milliseconds from the start of the audio.

Errors return a JSON envelope with a human-readable message and a stable machine-readable code. Branch on code, not on the message text.

{
  "error": {
    "message": "Your free API credit is used up. Add a payment method to continue.",
    "code": "v2_billing_required"
  }
}

Requests whose body fails schema validation (for example a POST /v2/transcripts with neither audio_url nor upload_id) return 422 with the framework's standard validation shape: a detail array describing each failing field. All other errors use the envelope above.

List endpoints (GET /v2/transcripts and GET /v2/usage) use offset pagination and return newest records first.

Results are wrapped in a data array. To page, increase offset by limit until fewer than limit rows come back.

Every account starts with 10 USD of free API credit. Usage is metered per second against the per-hour rate of the model you use and drawn from that credit first. When the credit is exhausted, requests that would incur cost return 402 with code v2_billing_required until you start a pay-as-you-go subscription through POST /v2/billing/checkout.

Check your balance at any time with GET /v2/credits, and inspect per-request charges with GET /v2/usage.

Uploads

Transcripts

Realtime

Voice agent

Speech intelligence

Account and billing