Skip to content
TomoriBot

KoboldCPP Transcription

KoboldCPP has Whisper-based STT support, but endpoint shape can vary by build. TomoriBot’s Phase 4 adapter expects OpenAI-compatible POST /v1/audio/transcriptions.

Setup

Section titled “Setup”

Start KoboldCPP with Whisper/STT enabled and confirm your build exposes:

  • POST /v1/audio/transcriptions
  • GET /v1/models or GET /models

Keep KoboldCPP running while TomoriBot is using it. If your build only exposes /api/extra/transcribe or another custom shape, use a wrapper until TomoriBot has a dedicated adapter.

Register in TomoriBot

Section titled “Register in TomoriBot”

Run /provider custom-endpoint add:

  • capability: transcription
  • api_style: openai-compatible-transcription
  • endpoint_url: your KoboldCPP server root

In the modal:

  • Transcription Model: the model name your server reports
  • Transcription Language: optional language hint, such as en or ja

Registration makes the endpoint active immediately. Use /model transcription later only when switching endpoints.

Use Transcripts

Section titled “Use Transcripts”

After registration, TomoriBot transcribes audio attachments in the background and adds the text to chat context. Use /speech transcripts only if you also want transcripts posted visibly in chat.