Use Inference APIs with Open WebUI

One connection gives Open WebUI chat models, speech-to-text and text-to-speech from Inference APIs — useful when the local Whisper or TTS engine is too slow on a CPU host.

Settings

Setting	Value
Base URL	`https://api.inferenceapis.com/v1` (also `/openai/v1`)
API key	From API Keys; send as `Authorization: Bearer …`
Chat model ids	`openai/gpt-oss-120b`, `deepseek-ai/DeepSeek-V4-Flash`, `zai-org/GLM-5.3-Flash`, `meta-llama/Llama-3.3-70B-Instruct-Turbo` — all models
Audio model ids	`openai/whisper-large-v3` (transcription), `hexgrad/Kokoro-82M` (speech)

Which model

Chat: any model from the list. Speech-to-text: check accuracy for your language first. Prices and measured speed: chat models.

Configuration

Chat: environment variables

OPENAI_API_BASE_URL=https://api.inferenceapis.com/v1
OPENAI_API_KEY=$INFERENCE_API_KEY
# or Admin Panel → Settings → Connections → OpenAI API → add https://api.inferenceapis.com/v1

Speech-to-text (replaces local Whisper)

AUDIO_STT_ENGINE=openai
AUDIO_STT_OPENAI_API_BASE_URL=https://api.inferenceapis.com/v1
AUDIO_STT_OPENAI_API_KEY=$INFERENCE_API_KEY
AUDIO_STT_MODEL=openai/whisper-large-v3

Text-to-speech

AUDIO_TTS_ENGINE=openai
AUDIO_TTS_OPENAI_API_BASE_URL=https://api.inferenceapis.com/v1
AUDIO_TTS_OPENAI_API_KEY=$INFERENCE_API_KEY
AUDIO_TTS_MODEL=hexgrad/Kokoro-82M
AUDIO_TTS_VOICE=af_heart

Verify

Admin Panel → Settings → Connections → click the verify icon next to the connection; the model list should populate.

How this guide was checked

Configuration only These are Open WebUI's documented environment variables. We verified the three endpoints it calls (chat, transcription, speech) and that OpenAI voice names are accepted, but did not run Open WebUI itself. If a step does not match what you see, tell us.

Gotchas

Open WebUI lists every id from /v1/models including aliases; hide the ones you do not want under Admin → Settings → Models.
The same variables can be set in the Admin Panel → Settings → Audio page instead of the environment.

If something fails

401 — key missing or wrong · 402 insufficient_balance — add credits on Billing · 404 model_not_found — check the id against the model list (aliases such as gpt-oss-120b work too) · 503 model_unavailable — the model is being enabled. Full details in the API docs.