v26.9.0 ·
Bring any model
A custom OpenAI-compatible AI provider on the desktop app, and a fix for models that refuse the temperature parameter.
🔌 Custom AI provider (desktop)
- Any OpenAI-compatible endpoint. A new "Custom (OpenAI-compatible)" option in AI settings takes any
/v1base URL: OpenRouter, Together, Fireworks, DeepSeek, LM Studio, vLLM, or your own gateway. - The key is optional. Local endpoints need none. Requests go straight from your machine to your endpoint — the same bring-your-own-key posture as every other provider, now for effectively any model.
🩹 Fixed
- Temperature rejections. Some models refuse the
temperatureparameter outright (Anthropic requests with extended thinking, OpenAI reasoning models), and the first AI action failed with a raw provider error. Wurk.Flow now strips the parameter and retries once when a provider rejects it — for every current and future provider, with no per-model compatibility list to go stale.