feat(serving): add OrcaRouter as a named LLM provider

README.md

@@ -220,6 +220,29 @@ After running, the operator will append the generated results into output_key. F
 ]
 ```
 
+### 🧠 6.1 Named LLM Backends
+
+Besides pointing a serving class at any OpenAI-compatible endpoint, DataFlow ships a few
+pre-configured named backends. For example, `APIOrcaRouterServing` wires the operator
+stack to the [OrcaRouter](https://www.orcarouter.ai) gateway — an OpenAI-compatible AI
+gateway that, like OpenRouter, exposes a provider/model namespace across many models
+through a single endpoint, while also adding adaptive routing, automatic failover,
+zero-markup inference, observability, guardrails, and agent-tool governance behind the
+same endpoint:
+
+```python
+from dataflow.serving import APIOrcaRouterServing
+
+# configure LLM serving with the OrcaRouter gateway
+# api key needs to be set via `export ORCAROUTER_API_KEY=sk-orca-...`
+llm_serving = APIOrcaRouterServing(model_name="orcarouter/auto")
+
+prompted_generator = PromptedGenerator(
+    llm_serving=llm_serving,  # pre-configured LLM backend
+    system_prompt="Please solve this math problem."
+)
+```
+
 <details>
 <summary><h2>🛠️ 7. Pipelines (Click to expand)</h2></summary>
 

dataflow/serving/__init__.py

@@ -1,4 +1,5 @@
 from .api_llm_serving_request import APILLMServing_request
+from .api_orca_router_serving import APIOrcaRouterServing
 from .local_model_llm_serving import LocalModelLLMServing_vllm
 from .local_model_llm_serving import LocalModelLLMServing_sglang
 from .api_vlm_serving_openai import APIVLMServing_openai
@@ -18,6 +19,7 @@
 __all__ = [
     "APIGoogleVertexAIServing",
     "APILLMServing_request",
+    "APIOrcaRouterServing",
     "LocalModelLLMServing_vllm",
     "LocalModelLLMServing_sglang",
     "APIVLMServing_openai",

dataflow/serving/api_orca_router_serving.py

@@ -0,0 +1,55 @@
+from .api_llm_serving_request import APILLMServing_request
+
+
+class APIOrcaRouterServing(APILLMServing_request):
+    """
+    OpenAI-compatible serving class backed by the OrcaRouter gateway.
+
+    OrcaRouter (https://www.orcarouter.ai) is an OpenAI-compatible AI gateway that,
+    like OpenRouter, exposes a provider/model namespace across many models through a
+    single endpoint. On top of that it adds adaptive routing, automatic failover,
+    zero-markup inference, observability, guardrails, and agent-tool governance on
+    the same OpenAI-compatible API.
+
+    This class reuses the request/retry/formatting logic of APILLMServing_request and
+    only wires it to OrcaRouter defaults, so DataFlow users can adopt the gateway
+    without treating it as an anonymous custom base URL.
+    """
+    def __init__(self,
+                 api_url: str = "https://api.orcarouter.ai/v1/chat/completions",
+                 key_name_of_api_key: str = "ORCAROUTER_API_KEY",
+                 model_name: str = "orcarouter/auto",
+                 temperature: float = 0.0,
+                 max_workers: int = 10,
+                 max_retries: int = 5,
+                 connect_timeout: float = 10.0,
+                 read_timeout: float = 120.0,
+                 **configs: dict):
+        """
+        Initialize OrcaRouter serving instance.
+
+        Args:
+            api_url: OrcaRouter OpenAI-compatible chat completions endpoint
+            key_name_of_api_key: Environment variable holding the OrcaRouter API key
+            model_name: OrcaRouter model namespace id (e.g. "orcarouter/auto")
+            temperature: Sampling temperature
+            max_workers: Number of concurrent workers for batch processing
+            max_retries: Number of LLM inference retry chances for each input
+            connect_timeout: Connection timeout in seconds
+            read_timeout: Read timeout in seconds
+            **configs: Additional parameters forwarded to the API payload
+
+        Note:
+            Set the API key via `export ORCAROUTER_API_KEY=sk-orca-...` before use.
+        """
+        super().__init__(
+            api_url=api_url,
+            key_name_of_api_key=key_name_of_api_key,
+            model_name=model_name,
+            temperature=temperature,
+            max_workers=max_workers,
+            max_retries=max_retries,
+            connect_timeout=connect_timeout,
+            read_timeout=read_timeout,
+            **configs,
+        )

test/cpu_only/test_api_orca_router_serving.py

@@ -0,0 +1,45 @@
+import pytest
+from dataflow.serving import APIOrcaRouterServing
+
+
+@pytest.mark.api
+def test_orca_router_serving_defaults_and_request(dummy_server_base_url, monkeypatch):
+    monkeypatch.setenv("ORCAROUTER_API_KEY", "dummy-key")
+
+    api_url = (
+        f"{dummy_server_base_url}/v1/chat/completions"
+        f"?queue=0&ka_interval=0.05&stream=0"
+        f"&body=hello&think="
+    )
+
+    cli = APIOrcaRouterServing(
+        api_url=api_url,
+        model_name="orcarouter/auto",
+        connect_timeout=1.0,
+        read_timeout=3.0,
+        max_retries=1,
+        max_workers=1,
+    )
+
+    assert cli.api_url == api_url
+    assert cli.model_name == "orcarouter/auto"
+    assert cli.api_key == "dummy-key"
+
+    _id, resp = cli._api_chat_with_id(
+        id=0,
+        payload=[{"role": "user", "content": "hi"}],
+        model="orcarouter/auto",
+        is_embedding=False,
+    )
+
+    assert _id == 0
+    assert resp == "hello"
+
+    cli.cleanup()
+
+
+@pytest.mark.api
+def test_orca_router_serving_requires_key(monkeypatch):
+    monkeypatch.delenv("ORCAROUTER_API_KEY", raising=False)
+    with pytest.raises(ValueError):
+        APIOrcaRouterServing()