The CoWork Transcription configuration controls speech-to-text for voice and video calls in i-net CoWork. Transcription can be disabled, automatically enabled for every call, or controlled by participants during a call. This page explains which settings administrators need and what must be considered when choosing a provider.

Configure the options in the Configuration application under CoWork Calls → Communication → CoWork Transcription. Only the settings required for the selected provider are displayed.

• Transcript Activation: Select how transcription is made available.
  • Off - Transcription is disabled.
  • Always On - Transcription starts automatically with every call.
  • Manual - Participants can start and stop transcription during a call.
  • Default value: Off
• Provider: Select the service that creates the transcript.
  • OpenAI - Uses an OpenAI account.
  • OVHcloud - Uses an OVHcloud account with documented privacy options relevant to European customers.
  • Gladia - Uses a Gladia account with documented privacy options relevant to European customers.
  • Mistral AI - Uses a Mistral AI account with documented privacy options relevant to European customers.
  • Custom - Uses a compatible service specified by the administrator.
  • Vosk - Processes speech locally on the CoWork Calls server.
  • Default value: OpenAI
• Language: Enter a two-letter language code such as de or en. Leave the field empty to let the provider detect the language automatically.
• Max. Chunk Duration: Sets the maximum duration of a continuous speech segment. The default value of 30 seconds is suitable for most installations.
  • Default value: 30 seconds
• Generate Summary at End of Call: Creates a summary after the complete transcript is available. This feature always uses OpenAI, independently of the selected transcription provider.
  • Default value: activated

Note: Documented privacy options are not a legal certification or a guarantee by CoWork Calls. The administrator must still review the contract, processing region, retention settings, sub-processors, and the legal basis for transcription.

The OpenAI section is displayed when OpenAI is selected or call summaries are enabled.

• OpenAI API Key: Enter the API key of the OpenAI account. The key is also required for summaries.
• Transcription Model: Enter the model that should create the transcript.
  • Default value: whisper-1
• Transcription Prompt: Optionally provide names, product terms, or other context that can improve recognition.

Additional recipient: When summaries are enabled, the complete transcript is sent to OpenAI even if another provider created it.

• OVHcloud Access Token: Enter the access token for the OVHcloud service.
• OVHcloud Model: Enter the transcription model available for the account.
  • Default value: whisper-large-v3

OVHcloud documents processing in France and states that submitted content is not stored or shared for this service. Review the current service terms and data processing agreement before activation. Further information is available in the OVHcloud audio documentation and the privacy information for AI Endpoints.

• Gladia API Key: Enter the API key of the Gladia account. No model selection is required.

Audio is sent to Gladia for transcription. CoWork Calls requests removal of the uploaded audio and transcript after processing. Gladia offers separate EU and US infrastructure and a data processing agreement. Verify that the account uses the intended region and review the listed sub-processors. Further information is available in the Gladia Compliance Hub and the Gladia data processing agreement.

• Mistral AI API Key: Enter the API key of the Mistral AI account.
• Mistral AI Model: Enter the transcription model available for the account.
  • Default value: voxtral-mini-latest

Mistral documents European data centers by default and states that API data is not used for model training. Review the data processing agreement and the privacy settings of the organization. Enable Zero Data Retention when it is available and required for the installation. Further information is available in the Mistral privacy controls and the Mistral data processing agreement.

Use this option for a self-hosted or external service that offers an OpenAI-compatible transcription interface. Obtain the required values from the administrator of that service.

• Transcription URL: Enter the complete transcription URL provided by the service. Use HTTPS for external services.
• Bearer Token: Enter the access token if the service requires one. Leave the field empty for services without authentication.
• Model: Enter a model supported by the service.
  • Default value: whisper-1

An unencrypted HTTP URL should only be used for a trusted service in the local network and produces a configuration warning. A self-hosted service can keep audio inside the operator's network. Speaches is one example of a compatible self-hosted service. Follow its current installation instructions outside CoWork Calls.

For a custom provider, the operator must assess the service, transport security, storage, processing region, and contractual requirements.

Vosk processes speech locally on the CoWork Calls server and does not require an external account.

• Vosk Model Path: Enter the path to a downloaded and unpacked Vosk language model. The directory must be readable by the server.

Suitable language models are available from the Vosk model list.

Before enabling transcription:

1. Define and document the legal basis for recording and transcription.
2. Inform call participants and obtain consent where required.
3. Review the provider's data processing agreement, sub-processors, processing region, and retention settings.
4. Protect API keys and access tokens and only grant the permissions required for transcription.
5. Remember that call summaries send the complete transcript to OpenAI.
6. Use non-production audio when testing a new provider.

1. Select Always On or Manual under Transcript Activation.
2. Select the provider and enter the account information shown in its section.
3. Leave Language empty for automatic detection or enter the desired language code.
4. Decide whether call summaries should be created. Configure the OpenAI API key when summaries are enabled.
5. Select Test Transcription. The test uses a bundled audio sample and also checks settings that have not yet been saved. API keys and access tokens are never displayed in the result.
6. Save the configuration after the test succeeds.