Speech-to-text for AI agents and vibe coders

The prompt to paste

Works in Cursor, Claude Code, Codex, Windsurf, Lovable, Replit, Bolt and v0. It points the agent at the machine-readable instructions so it uses the right model ids and parameters.

Add speech-to-text to this app with the KolWrite API.
Read https://console.kolwrite.com/agent-instructions.md first.
- Key: process.env.KOLWRITE_API_KEY, sent as the x-api-key header (server side only).
- POST https://app.kolwrite.com/transcribe {"audio_url", "model": "oracle", "language": "auto", "output_format": "json"}
- Local files: POST /upload first, then use fileUrl as audio_url.
- Poll GET /transcribe/{jobId} every 5 s until COMPLETED or FAILED.
- Show result.segments (start, end, speaker, text).

Pin it in your project

Add this line to AGENTS.md, CLAUDE.md, .cursorrules or your tool's project rules, so every future session uses the same reference:

For speech-to-text use the KolWrite API; read https://console.kolwrite.com/agent-instructions.md and https://console.kolwrite.com/llms-full.txt before writing code.

What the agent will build

1. Read the file URL (or upload a local file with POST /upload and take fileUrl).
2. Start a job with POST /transcribe and keep the jobId.
3. Poll GET /transcribe/{jobId} until COMPLETED, or receive a webhook.
4. Render result.segments: text, start and end times, speaker labels.

The smallest working request (voice note to plain text):

curl https://app.kolwrite.com/transcribe \
  -H "x-api-key: $KOLWRITE_API_KEY" -H "Content-Type: application/json" \
  -d '{"audio_url": "https://example.com/voice-note.m4a", "model": "oracle_mini", "output_format": "txt"}'

Where the API key comes from

A person creates the account at console.kolwrite.com/auth, buys credits and generates a key on the API Keys page. Agents cannot sign up on their own. Put the key in an environment variable (KOLWRITE_API_KEY) or your platform's secret store, and call the API only from server code or a serverless function, never from the browser.

What it costs

Metered per second, no minimum. Full rate card: /llms/pricing.md.

Machine-readable references

• llms.txt: which call and model to use for which job.
• llms-full.txt: the complete reference in one Markdown file.
• openapi.json: OpenAPI 3.1 for code generators and tool definitions.
• agent-instructions.md: the prompt, model ids and gotchas.
• Every page on this site also has a Markdown version: add .md to its address.

Questions and answers