Before you start

1

Create or sign in to OpenRouter

Visit openrouter.ai and sign in. OpenRouter is a separate service with its own terms and privacy policy.

2

Add credits and set a limit

Add a small amount of credit, then set a spending limit that fits your use. SayCopy itself is free, but the models you choose may charge for usage through OpenRouter.

3

Create an API key

Open OpenRouter Keys, create a key for SayCopy, and copy it. A separate key makes it easy to monitor or revoke access later.

4

Save it in SayCopy

In the SayCopy app, open Settings, paste your key into the OpenRouter token field, and tap Save token. The key is stored in the platform's secure storage on your device.

5

Choose your preferred models

In Settings, use Browse transcription models for speech-to-text and Browse translation models for translation and cleanup. These are separate because the two jobs use different kinds of models. If you are unsure, tap Use recommended defaults. The recommended transcription model is openai/whisper-large-v3-turbo, chosen for speed; High accuracy selects openai/whisper-large-v3.

Your preferred transcription model is used when you select a specific language: English, Spanish, Arabic, French, German, Italian, Portuguese, Chinese, Japanese, Korean, or Hindi. Each model shows how many of these it supports. Auto-detect uses a separate Whisper Large V3 Turbo engine, so choosing Auto does not replace your preferred model. The source recording language you choose is saved and restored when you reopen SayCopy.

6

Transcribe or translate

Open Transcribe and speak to get a transcript you can copy or share. Open Translate to type, paste, or speak a phrase, choose From and To languages, listen to the result, and tap Save to keep it with tags. Tap Saved to open your saved translations in a bottom sheet.

Why are there two model choices?

Transcription models listen to audio and return text. Translation and cleanup models receive that text and improve or translate it. A model may support one job without supporting the other, so SayCopy keeps the choices separate.

How does Auto-detect work?

Auto-detect uses Whisper Large V3 Turbo to identify the spoken language and preserve it in the transcript. Your preferred model stays saved and is used whenever you select a language explicitly.

What does Light cleanup do?

When enabled on the Transcribe tab, the active text model improves punctuation, spacing, capitalization, and small artifacts without translating the transcript. SayCopy bounds this polishing step and keeps the original transcription if cleanup cannot finish. Turning it off returns the raw speech-to-text result directly.

What does it cost?

SayCopy does not add a model subscription. OpenRouter charges your account according to the models you select and shows usage in its activity dashboard. Start with a small balance and spending limit until you understand your normal usage.

Troubleshooting

Auto-detect and my preferred model show different names

This is expected. Auto-detect has a dedicated engine, while the preferred model is used for recordings where you select a specific language. See the language and model guide for examples.

The token is not accepted

Copy the complete key again from OpenRouter, make sure no spaces were added, and confirm the key has not been disabled or limited to different models.

A selected model does not run

Return to Settings and select the model from the appropriate in-app OpenRouter list. Availability can change, and an account or key-level model restriction can also prevent routing.

A request says payment is required

Check your OpenRouter balance and key spending limit. SayCopy cannot add credits or change limits on your behalf.

Keep your key private

Treat an API key like a password. Do not post it, email it, or share screenshots that show it. If you think it has been exposed, revoke it in OpenRouter and create a new one.

Need help? Use the SayCopy contact form without including your API key.

Error reporting

Updated SayCopy builds send error diagnostics to Sentry automatically, including handled failures. You do not need a Sentry account. Application reports exclude your API key, audio, transcripts, translations, and clipboard contents. See the privacy policy for details.

If an October 5 preview returns HTTP 400 with Auto-detect, choose a specific recording language and the Recommended transcription model until you install the updated preview. MAI-Transcribe 1.5 rejects SayCopy’s M4A recordings; updated Auto-detect uses Whisper Large V3 Turbo.