From e1795624b38981b2529a4f4977875312a55c7d2f Mon Sep 17 00:00:00 2001
From: XiaoHuo888 <sjh00112233@outlook.com>
Date: Wed, 12 Aug 2026 06:29:42 +0800
Subject: [PATCH] docs: document OrcaRouter as a hosted inference option

Adds a named OrcaRouter example alongside the existing OpenRouter
examples in the Model docstring, the SFT distillation teacher-client
snippet, and the MCP-RL scenario-generation docs.

OrcaRouter is an OpenAI-compatible gateway (https://api.orcarouter.ai/v1,
keys start with sk-orca-) that exposes 150+ models behind a single
endpoint, including the orcarouter/auto router model.

Co-Authored-By: Claude <noreply@anthropic.com>
---
 docs/features/mcp-rl.mdx           | 12 ++++++++++++
 docs/fundamentals/sft-training.mdx |  6 ++++++
 src/art/model.py                   | 10 ++++++++++
 3 files changed, 28 insertions(+)

diff --git a/docs/features/mcp-rl.mdx b/docs/features/mcp-rl.mdx
index 2cdf2b930..1fa7be5b0 100644
--- a/docs/features/mcp-rl.mdx
+++ b/docs/features/mcp-rl.mdx
@@ -42,6 +42,18 @@ scenario_collection = await generate_scenarios(
 )
 ```
 
+You can generate scenarios against [OrcaRouter](https://www.orcarouter.ai) the same way — point `generator_base_url` at `https://api.orcarouter.ai/v1` and use an OrcaRouter key (keys start with `sk-orca-`):
+
+```python
+scenario_collection = await generate_scenarios(
+    tools=tools_list,
+    num_scenarios=24,
+    show_preview=True,
+    generator_api_key="your_orcarouter_key",
+    generator_base_url="https://api.orcarouter.ai/v1",
+)
+```
+
 ART automatically generates diverse training scenarios that exercise different aspects of the MCP server: simple single-tool usage, complex multi-step workflows, edge cases and error handling, and creative combinations of available tools.
 
 ### 3. **RULER Evaluation**
diff --git a/docs/fundamentals/sft-training.mdx b/docs/fundamentals/sft-training.mdx
index e49fe475e..5242e156a 100644
--- a/docs/fundamentals/sft-training.mdx
+++ b/docs/fundamentals/sft-training.mdx
@@ -142,6 +142,12 @@ async def main():
         api_key="your-api-key",
         base_url="https://openrouter.ai/api/v1",
     )
+    # Or point the teacher at OrcaRouter:
+    #
+    # teacher_client = AsyncOpenAI(
+    #     api_key=os.getenv("ORCAROUTER_API_KEY"),
+    #     base_url="https://api.orcarouter.ai/v1",
+    # )
     # Small models often produce malformed JSON or miss fields.
     # Distilling from a larger model teaches consistent structured extraction.
     system_prompt = "Extract {name, role, company} as JSON from the text. Return only valid JSON."
diff --git a/src/art/model.py b/src/art/model.py
index 1dd47cbb4..bc6f497b5 100644
--- a/src/art/model.py
+++ b/src/art/model.py
@@ -433,6 +433,16 @@ class Model(
     )
     ``
 
+    Or, if you're pointing at [OrcaRouter](https://www.orcarouter.ai):
+
+    ``python model = art.Model(
+        name="gemini-2.5-pro", project="my-project",
+        inference_api_key=os.getenv("ORCAROUTER_API_KEY"),
+        inference_base_url="https://api.orcarouter.ai/v1",
+        inference_model_name="orcarouter/auto",
+    )
+    ``
+
     For trainable (`art.TrainableModel`) models the inference values will be
     populated automatically by `model.register(api)` so you generally don't need
     to think about them.