Custom (OpenAI-compatible endpoint)

The Custom preset lets you connect any OpenAI-compatible API endpoint to Quendian — for either transcription or summarization. This covers self-hosted inference servers, private cloud deployments, and aggregator services not listed as named presets.

Roles: Transcription and/or Summarization (any endpoint that implements the relevant OpenAI API contract)

---

API contracts

Summarization

Quendian sends summarization requests using the standard OpenAI Chat Completions format:

POST /chat/completions
Authorization: Bearer <your-key>
Content-Type: application/json

{
  "model": "<your-model>",
  "messages": [...]
}

Any server that implements the OpenAI-compatible POST /chat/completions endpoint and accepts Bearer token authentication will work.

Compatible summarization servers:

• vLLM — high-throughput inference server
• TGI (Text Generation Inference) — Hugging Face’s server
• LiteLLM — proxy that translates to 100+ provider APIs
• Any other OpenAI-compatible aggregator or self-hosted deployment

Transcription

Quendian sends transcription requests using the standard OpenAI Whisper API format:

POST /audio/transcriptions
Authorization: Bearer <your-key>
Content-Type: multipart/form-data

file: <audio bytes>
model: <your-model>
response_format: verbose_json

Any server that implements the OpenAI-compatible POST /audio/transcriptions endpoint and returns Whisper-shape JSON (segments and word-level timestamps in verbose_json mode) will work.

Compatible transcription servers:

• faster-whisper-server — high-performance self-hosted Whisper
• whisper.cpp — local C++ implementation with an OpenAI-compatible server mode
• LiteLLM — proxy that routes to upstream Whisper providers
• Any other server that implements the OpenAI /audio/transcriptions contract

Note

Most OpenAI-compatible transcription servers (faster-whisper, whisper.cpp, etc.) don’t return speaker labels. Quendian can add them on-device if you opt in via Preferences → Speakers → “Identify speakers on-device”. Local diarization has tradeoffs — see Speakers → Local diarization.

---

Steps

1. Note your endpoint and credentials

You need:

• Base URL — e.g. https://your-server.example.com/v1 or http://localhost:8000/v1
• API key — or any Bearer token your server accepts (leave blank if your server has no auth)
• Model name — the identifier your server expects (e.g. whisper-large-v3 for transcription, llama3 or mistral-7b for summarization)

2. Open Preferences in Quendian

Navigate to Quendian → Preferences → Summarization.

3. Select Custom preset

Choose the Custom preset under the role you want to configure (Transcription or Summarization) and fill in:

Field	Value
Base URL	Your OpenAI-compatible endpoint (include /v1)
API key	Your Bearer token (or leave blank)
Model	Model identifier your server expects

Click Save.

You can use Custom for both roles independently — for example, a self-hosted Whisper server for transcription and a hosted OpenAI-compatible LLM for summarization.

Tip

If you’re using LiteLLM as a proxy, set the Base URL to your LiteLLM server and use whichever model string LiteLLM maps to your target provider.

Caution

Quendian does not validate the endpoint at configuration time. If requests fail, check that your server is reachable from your Mac and that the model name exactly matches what your server expects.