UN-3503 [FEAT] Add NVIDIA Build and OpenRouter OpenAI-compatible LLM and embedding adapters

unstract/sdk1/src/unstract/sdk1/adapters/base1.py

@@ -353,6 +353,9 @@ def validate_model(adapter_metadata: dict[str, "Any"]) -> str:
 _OPENAI_PROVIDER_PREFIX = "openai/"
 _CUSTOM_OPENAI_PROVIDER_PREFIX = "custom_openai/"
 _OPENAI_REASONING_MODEL_PATTERN = re.compile(r"^(o1|o3|o4|gpt-5)(?:[-/]|$)")
+# Keyless gateways still need a non-empty key; the OpenAI SDK rejects a
+# null/blank one before any request reaches the endpoint.
+_NO_AUTH_API_KEY = "no-auth"
 
 
 def _is_openai_reasoning_model(model: str) -> bool:
@@ -494,6 +497,86 @@ def validate_model(adapter_metadata: dict[str, "Any"]) -> str:
         return f"{_CUSTOM_OPENAI_PROVIDER_PREFIX}{model}"
 
 
+# Shared validation for branded adapters that reuse the OpenAI-compatible wire
+# protocol with a fixed default endpoint.
+def _validate_branded_openai_compatible(
+    adapter_metadata: dict[str, "Any"], default_api_base: str
+) -> dict[str, "Any"]:
+    # Endpoint stays overridable so a provider URL change needs no release.
+    api_base = adapter_metadata.get("api_base")
+    if not (isinstance(api_base, str) and api_base.strip()):
+        api_base = default_api_base
+    adapter_metadata = {**adapter_metadata, "api_base": api_base}
+    return OpenAICompatibleLLMParameters.validate(adapter_metadata)
+
+
+_NVIDIA_BUILD_API_BASE = "https://integrate.api.nvidia.com/v1"
+_OPENROUTER_API_BASE = "https://openrouter.ai/api/v1"
+_OPENROUTER_PROVIDER_PREFIX = "openrouter/"
+
+
+class NvidiaBuildLLMParameters(OpenAICompatibleLLMParameters):
+    """OpenAI-compatible adapter for NVIDIA's hosted models (build.nvidia.com)."""
+
+    # Required str so a directly-constructed instance stays valid.
+    api_base: str = _NVIDIA_BUILD_API_BASE
+
+    @staticmethod
+    def validate(adapter_metadata: dict[str, "Any"]) -> dict[str, "Any"]:
+        return _validate_branded_openai_compatible(
+            adapter_metadata, _NVIDIA_BUILD_API_BASE
+        )
+
+
+class OpenRouterLLMParameters(BaseChatCompletionParameters):
+    """Adapter for OpenRouter (openrouter.ai).
+
+    Routed through LiteLLM's native `openrouter/` provider so per-token costs
+    resolve and reasoning params map without provider-specific workarounds.
+    """
+
+    api_key: str
+    api_base: str = _OPENROUTER_API_BASE
+    reasoning_effort: str | None = None
+
+    @staticmethod
+    def validate(adapter_metadata: dict[str, "Any"]) -> dict[str, "Any"]:
+        adapter_metadata = dict(adapter_metadata)
+        api_base = adapter_metadata.get("api_base")
+        if not (isinstance(api_base, str) and api_base.strip()):
+            adapter_metadata["api_base"] = _OPENROUTER_API_BASE
+        adapter_metadata["model"] = OpenRouterLLMParameters.validate_model(
+            adapter_metadata
+        )
+        # Reasoning models reject a non-default temperature; drop it and send
+        # reasoning_effort only when reasoning is enabled. On a re-validation
+        # pass `enable_reasoning` is absent (it isn't serialized), so recover the
+        # state from a surviving reasoning_effort to keep reload idempotent —
+        # unless the user explicitly opted out with `enable_reasoning: false`.
+        enable_reasoning = adapter_metadata.get("enable_reasoning", False)
+        if (
+            not enable_reasoning
+            and "enable_reasoning" not in adapter_metadata
+            and adapter_metadata.get("reasoning_effort") is not None
+        ):
+            enable_reasoning = True
+        adapter_metadata.pop("enable_reasoning", None)
+        if enable_reasoning:
+            adapter_metadata["temperature"] = None
+        else:
+            adapter_metadata.pop("reasoning_effort", None)
+        return OpenRouterLLMParameters(**adapter_metadata).model_dump()
+
+    @staticmethod
+    def validate_model(adapter_metadata: dict[str, "Any"]) -> str:
+        model = str(adapter_metadata.get("model", "")).strip()
+        if not model:
+            raise ValueError("model is required for the OpenRouter adapter.")
+        if model.startswith(_OPENROUTER_PROVIDER_PREFIX):
+            return model
+        return f"{_OPENROUTER_PROVIDER_PREFIX}{model}"
+
+
 class AzureOpenAILLMParameters(BaseChatCompletionParameters):
     """See https://docs.litellm.ai/docs/providers/azure/#completion---using-azure_ad_token-api_base-api_version."""
 
@@ -1296,6 +1379,8 @@ class OpenAIEmbeddingParameters(BaseEmbeddingParameters):
     api_base: str | None = None
     embed_batch_size: int | None = 10
     dimensions: int | None = None  # For text-embedding-3-* models
+    # Strict endpoints reject the null LiteLLM sends when this is unset.
+    encoding_format: str | None = None
 
     @staticmethod
     def validate(adapter_metadata: dict[str, "Any"]) -> dict[str, "Any"]:
@@ -1311,6 +1396,77 @@ def validate_model(adapter_metadata: dict[str, "Any"]) -> str:
         return model
 
 
+# custom_openai has no embedding support, so these route through LiteLLM's
+# native nvidia_nim provider; its default api_base is pinned for the schema.
+_NVIDIA_NIM_PROVIDER_PREFIX = "nvidia_nim/"
+
+
+class NvidiaBuildEmbeddingParameters(OpenAIEmbeddingParameters):
+    """OpenAI-compatible embeddings via NVIDIA's hosted endpoint (build.nvidia.com)."""
+
+    # Overridable default endpoint.
+    api_base: str = _NVIDIA_BUILD_API_BASE
+
+    @staticmethod
+    def validate(adapter_metadata: dict[str, "Any"]) -> dict[str, "Any"]:
+        adapter_metadata = dict(adapter_metadata)
+        # Endpoint stays overridable so a provider URL change needs no release.
+        api_base = adapter_metadata.get("api_base")
+        if not (isinstance(api_base, str) and api_base.strip()):
+            adapter_metadata["api_base"] = _NVIDIA_BUILD_API_BASE
+        # Strict endpoints reject the null LiteLLM sends; pin a real value.
+        adapter_metadata.setdefault("encoding_format", "float")
+        adapter_metadata["model"] = NvidiaBuildEmbeddingParameters.validate_model(
+            adapter_metadata
+        )
+        return OpenAIEmbeddingParameters(**adapter_metadata).model_dump()
+
+    @staticmethod
+    def validate_model(adapter_metadata: dict[str, "Any"]) -> str:
+        model = str(adapter_metadata.get("model", "")).strip()
+        if not model:
+            raise ValueError("model is required for the NVIDIA Build embedding adapter.")
+        if model.startswith(_NVIDIA_NIM_PROVIDER_PREFIX):
+            return model
+        return f"{_NVIDIA_NIM_PROVIDER_PREFIX}{model}"
+
+
+class OpenAICompatibleEmbeddingParameters(OpenAIEmbeddingParameters):
+    """Embeddings for any OpenAI-compatible server (vLLM, self-hosted, etc.).
+
+    Routes through the `openai/` provider with a user-supplied `api_base`;
+    cost stays unresolved since the endpoint is arbitrary.
+    """
+
+    # Some gateways are keyless; the endpoint is always required.
+    api_key: str | None = None
+    api_base: str
+
+    @staticmethod
+    def validate(adapter_metadata: dict[str, "Any"]) -> dict[str, "Any"]:
+        adapter_metadata = dict(adapter_metadata)
+        api_key = adapter_metadata.get("api_key")
+        if not (isinstance(api_key, str) and api_key.strip()):
+            adapter_metadata["api_key"] = _NO_AUTH_API_KEY
+        # Strict endpoints reject the null LiteLLM sends; pin a real value.
+        adapter_metadata.setdefault("encoding_format", "float")
+        adapter_metadata["model"] = OpenAICompatibleEmbeddingParameters.validate_model(
+            adapter_metadata
+        )
+        return OpenAICompatibleEmbeddingParameters(**adapter_metadata).model_dump()
+
+    @staticmethod
+    def validate_model(adapter_metadata: dict[str, "Any"]) -> str:
+        model = str(adapter_metadata.get("model", "")).strip()
+        if not model:
+            raise ValueError(
+                "model is required for the OpenAI Compatible embedding adapter."
+            )
+        if model.startswith(_OPENAI_PROVIDER_PREFIX):
+            return model
+        return f"{_OPENAI_PROVIDER_PREFIX}{model}"
+
+
 class AzureOpenAIEmbeddingParameters(BaseEmbeddingParameters):
     """See https://docs.litellm.ai/docs/providers/azure."""
 

unstract/sdk1/src/unstract/sdk1/adapters/embedding1/__init__.py

@@ -4,8 +4,12 @@
 from unstract.sdk1.adapters.embedding1.azure_openai import AzureOpenAIEmbeddingAdapter
 from unstract.sdk1.adapters.embedding1.bedrock import AWSBedrockEmbeddingAdapter
 from unstract.sdk1.adapters.embedding1.gemini import GeminiEmbeddingAdapter
+from unstract.sdk1.adapters.embedding1.nvidia_build import NvidiaBuildEmbeddingAdapter
 from unstract.sdk1.adapters.embedding1.ollama import OllamaEmbeddingAdapter
 from unstract.sdk1.adapters.embedding1.openai import OpenAIEmbeddingAdapter
+from unstract.sdk1.adapters.embedding1.openai_compatible import (
+    OpenAICompatibleEmbeddingAdapter,
+)
 from unstract.sdk1.adapters.embedding1.vertexai import VertexAIEmbeddingAdapter
 from unstract.sdk1.adapters.enums import AdapterTypes
 
@@ -18,7 +22,9 @@
     "AzureOpenAIEmbeddingAdapter",
     "AWSBedrockEmbeddingAdapter",
     "GeminiEmbeddingAdapter",
+    "NvidiaBuildEmbeddingAdapter",
     "OpenAIEmbeddingAdapter",
+    "OpenAICompatibleEmbeddingAdapter",
     "VertexAIEmbeddingAdapter",
     "OllamaEmbeddingAdapter",
 ]

unstract/sdk1/src/unstract/sdk1/adapters/embedding1/nvidia_build.py

@@ -0,0 +1,45 @@
+from typing import Any
+
+from unstract.sdk1.adapters.base1 import BaseAdapter, NvidiaBuildEmbeddingParameters
+from unstract.sdk1.adapters.enums import AdapterTypes
+
+DESCRIPTION = (
+    "Adapter for NVIDIA's OpenAI-compatible embedding models (build.nvidia.com). "
+    "Supply a model name and your NVIDIA API key; the endpoint is preconfigured."
+)
+
+
+class NvidiaBuildEmbeddingAdapter(NvidiaBuildEmbeddingParameters, BaseAdapter):
+    @staticmethod
+    def get_id() -> str:
+        return "nvidiabuild|2afcdf59-5323-4086-97fb-d9a0432e7795"
+
+    @staticmethod
+    def get_metadata() -> dict[str, Any]:
+        return {
+            "name": "NVIDIA Build",
+            "version": "1.0.0",
+            "adapter": NvidiaBuildEmbeddingAdapter,
+            "description": DESCRIPTION,
+            "is_active": True,
+        }
+
+    @staticmethod
+    def get_name() -> str:
+        return "NVIDIA Build"
+
+    @staticmethod
+    def get_description() -> str:
+        return DESCRIPTION
+
+    @staticmethod
+    def get_provider() -> str:
+        return "nvidia_build"
+
+    @staticmethod
+    def get_icon() -> str:
+        return "/icons/adapter-icons/NvidiaBuild.png"
+
+    @staticmethod
+    def get_adapter_type() -> AdapterTypes:
+        return AdapterTypes.EMBEDDING

unstract/sdk1/src/unstract/sdk1/adapters/embedding1/openai_compatible.py

@@ -0,0 +1,49 @@
+from typing import Any
+
+from unstract.sdk1.adapters.base1 import (
+    BaseAdapter,
+    OpenAICompatibleEmbeddingParameters,
+)
+from unstract.sdk1.adapters.enums import AdapterTypes
+
+DESCRIPTION = (
+    "Embedding adapter for servers that implement the OpenAI Embeddings API "
+    "(vLLM, self-hosted gateways, and third-party providers). "
+    "Use OpenAI for the official OpenAI service."
+)
+
+
+class OpenAICompatibleEmbeddingAdapter(OpenAICompatibleEmbeddingParameters, BaseAdapter):
+    @staticmethod
+    def get_id() -> str:
+        return "openaicompatible|65573de7-2ea5-4631-bb49-492717972455"
+
+    @staticmethod
+    def get_metadata() -> dict[str, Any]:
+        return {
+            "name": "OpenAI Compatible",
+            "version": "1.0.0",
+            "adapter": OpenAICompatibleEmbeddingAdapter,
+            "description": DESCRIPTION,
+            "is_active": True,
+        }
+
+    @staticmethod
+    def get_name() -> str:
+        return "OpenAI Compatible"
+
+    @staticmethod
+    def get_description() -> str:
+        return DESCRIPTION
+
+    @staticmethod
+    def get_provider() -> str:
+        return "custom_openai"
+
+    @staticmethod
+    def get_icon() -> str:
+        return "/icons/adapter-icons/OpenAICompatible.png"
+
+    @staticmethod
+    def get_adapter_type() -> AdapterTypes:
+        return AdapterTypes.EMBEDDING

unstract/sdk1/src/unstract/sdk1/adapters/embedding1/static/azure.json

@@ -54,13 +54,6 @@
       "title": "Dimensions",
       "description": "Output embedding dimensions. Only supported by text-embedding-3-* models. Leave empty for default dimensions."
     },
-    "embed_batch_size": {
-      "type": "number",
-      "minimum": 0,
-      "multipleOf": 1,
-      "title": "Embedding Batch Size",
-      "default": 5
-    },
     "max_retries": {
       "type": "number",
       "minimum": 0,

unstract/sdk1/src/unstract/sdk1/adapters/embedding1/static/custom_openai.json

@@ -0,0 +1,53 @@
+{
+  "title": "OpenAI Compatible Embedding",
+  "type": "object",
+  "required": [
+    "adapter_name",
+    "model",
+    "api_base"
+  ],
+  "properties": {
+    "adapter_name": {
+      "type": "string",
+      "title": "Name",
+      "default": "",
+      "description": "Provide a unique name for this adapter instance. Example: compatible-emb-1"
+    },
+    "model": {
+      "type": "string",
+      "title": "Model",
+      "description": "The embedding model name expected by your OpenAI-compatible endpoint. Examples: text-embedding-3-small, BAAI/bge-m3, nomic-embed-text"
+    },
+    "api_key": {
+      "type": [
+        "string",
+        "null"
+      ],
+      "title": "API Key",
+      "format": "password",
+      "description": "API key for your OpenAI-compatible endpoint. Leave empty if the endpoint does not require one."
+    },
+    "api_base": {
+      "type": "string",
+      "format": "uri",
+      "title": "API Base",
+      "description": "Base URL for the OpenAI-compatible embeddings endpoint. Examples: https://gateway.example.com/v1, https://llm.example.net/openai/v1"
+    },
+    "max_retries": {
+      "type": "number",
+      "minimum": 0,
+      "multipleOf": 1,
+      "title": "Max Retries",
+      "default": 3,
+      "description": "The maximum number of times to retry a request if it fails."
+    },
+    "timeout": {
+      "type": "number",
+      "minimum": 0,
+      "multipleOf": 1,
+      "title": "Timeout",
+      "default": 240,
+      "description": "Timeout in seconds"
+    }
+  }
+}

unstract/sdk1/src/unstract/sdk1/adapters/embedding1/static/gemini.json

@@ -25,13 +25,6 @@
       "default": "",
       "format": "password"
     },
-    "embed_batch_size": {
-      "type": "number",
-      "minimum": 1,
-      "multipleOf": 1,
-      "title": "Embed Batch Size",
-      "description": "Number of texts to embed in a single batch. Leave empty to use the system default."
-    },
     "timeout": {
       "type": "number",
       "minimum": 0,