OrcaRouter Pipe Function for Open WebUI

A Pipe Function that adds OrcaRouter — an OpenAI-compatible LLM router — as a manifold provider in Open WebUI.

Install

In Open WebUI:

1. Admin Panel → Functions → Import from Link
2. Paste:

   https://raw.githubusercontent.com/Continuum-AI-Corp/openwebui-plugin-orcarouter/main/orcarouter.py
   

3. Click Import, then Save.
4. Open the new OrcaRouter entry → ⚙️ Valves → set ORCAROUTER_API_KEY to your sk-orca-... key (get one at orcarouter.ai/console).
5. Toggle the function on.

OrcaRouter models now appear in the chat model selector with the orcarouter. prefix.

Configuration (Valves)

Field	Default	Notes
ORCAROUTER_API_BASE_URL	https://api.orcarouter.ai/v1	Override only if you're testing against a different OrcaRouter deployment.
ORCAROUTER_API_KEY	(empty)	Required. Format: sk-orca-....
NAME_PREFIX	(empty)	Optional prefix prepended to each model name in the selector.
MODEL_CACHE_TTL_SECONDS	600	How long to cache the model list before refetching from /v1/models.
REQUEST_TIMEOUT_SECONDS	300	HTTP timeout. Reasoning models can take minutes to cold-start.

Per-user ORCAROUTER_API_KEY is also supported under the user's User Valves for the function — if set, it overrides the admin-level key for that user's requests.

Models

The function calls GET /v1/models and surfaces chat-capable models in the selector. Image generation, video, embedding, TTS/STT, rerank, codex, and gpt-5-pro family models are filtered out because they use different OrcaRouter endpoints and would fail in /v1/chat/completions.

A few flagship IDs you'll see:

• orcarouter/auto — adaptive router; the strategy is configured per-workspace at orcarouter.ai/console/routing.
• openai/gpt-5.5, anthropic/claude-opus-4.7, google/gemini-3-flash-preview, deepseek/deepseek-v4-pro, grok/grok-4.3, minimax/minimax-m2.7, qwen/qwen3.6-flash, and around 150 others.

Full catalog: https://www.orcarouter.ai/models.

Notes

• The function sends HTTP-Referer: https://openwebui.com/ and X-Title: Open WebUI so OrcaRouter's dashboard can attribute Open WebUI traffic, matching what Open WebUI sends for OpenRouter.
• Reasoning models (e.g. anthropic/claude-opus-4.7, openai/gpt-5 family, deepseek/deepseek-reasoner) reject the temperature parameter. If you see an upstream 400 complaining about temperature, leave it at the default in the chat settings for that model.
• Streaming and non-streaming chat completions are both supported. Tool calling is supported when the underlying upstream model supports it. The orcarouter/auto pool may include models without tool support — if you need tool calling, either pin a specific model or configure the auto router pool from the OrcaRouter console.

Version

0.1.0 — initial release. Future versions may add UI controls for reasoning_effort and extra_body.models / route: "fallback".

License

MIT — see LICENSE.