feat: add Ollama, Azure OpenAI, and generic OpenAI-compatible providers

src/skillspector/providers/__init__.py

@@ -22,10 +22,13 @@
 
 Selection happens via the ``SKILLSPECTOR_PROVIDER`` env var:
 
-    openai           → OpenAIProvider          (api.openai.com)
-    anthropic        → AnthropicProvider       (api.anthropic.com)
-    anthropic_proxy  → AnthropicProxyProvider  (Vertex-style raw-predict proxy)
-    nv_build         → NvBuildProvider         (build.nvidia.com)
+    openai            → OpenAIProvider              (api.openai.com)
+    anthropic         → AnthropicProvider            (api.anthropic.com)
+    anthropic_proxy   → AnthropicProxyProvider       (Vertex-style raw-predict proxy)
+    nv_build          → NvBuildProvider              (build.nvidia.com)
+    ollama            → OllamaProvider               (local Ollama instance)
+    azure_openai      → AzureOpenAIProvider          (Azure OpenAI Service)
+    openai_compatible → OpenAICompatibleProvider     (Groq, Together AI, Mistral, etc.)
 
 When unset, the selector defaults to ``nv_build``.
 """
@@ -44,6 +47,7 @@
     "No LLM API key configured. Set the credential env var for the "
     "active provider, or set OPENAI_API_KEY (and optionally "
     "OPENAI_BASE_URL) to use a standard OpenAI-compatible endpoint. "
+    "For local models, try SKILLSPECTOR_PROVIDER=ollama. "
     "Use --no-llm to skip LLM analysis and run static checks only."
 )
 
@@ -69,6 +73,18 @@ def _select_active_provider() -> LLMProvider:
         from .anthropic_proxy import AnthropicProxyProvider
 
         return AnthropicProxyProvider()
+    if name == "ollama":
+        from .ollama import OllamaProvider
+
+        return OllamaProvider()
+    if name == "azure_openai":
+        from .azure_openai import AzureOpenAIProvider
+
+        return AzureOpenAIProvider()
+    if name == "openai_compatible":
+        from .openai_compatible import OpenAICompatibleProvider
+
+        return OpenAICompatibleProvider()
     if name == "nv_build":
         return NvBuildProvider()
     if name in ("nv_inference", ""):
@@ -83,7 +99,8 @@ def _select_active_provider() -> LLMProvider:
 
     raise ValueError(
         f"Unknown SKILLSPECTOR_PROVIDER: {name!r}. "
-        "Expected one of: openai, anthropic, anthropic_proxy, nv_build (or unset)."
+        "Expected one of: openai, anthropic, anthropic_proxy, ollama, "
+        "azure_openai, openai_compatible, nv_build (or unset)."
     )
 
 

src/skillspector/providers/azure_openai/__init__.py

@@ -0,0 +1,20 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Azure OpenAI provider package (Azure-hosted OpenAI deployments)."""
+
+from .provider import REGISTRY_PATH, AzureOpenAIProvider
+
+__all__ = ["REGISTRY_PATH", "AzureOpenAIProvider"]

src/skillspector/providers/azure_openai/model_registry.yaml

@@ -0,0 +1,23 @@
+# Token-budget metadata for the AzureOpenAIProvider (Azure-hosted OpenAI
+# deployments).  Bundled with the package; consulted whenever the active
+# provider is AzureOpenAIProvider.
+#
+# Format:
+#   models:
+#     "<model-label>":
+#       context_length: <int>          # total context window in tokens (required)
+#       max_output_tokens: <int>       # model's max output cap (optional)
+
+models:
+  "gpt-4o":
+    context_length: 128000
+    max_output_tokens: 16384
+  "gpt-4o-mini":
+    context_length: 128000
+    max_output_tokens: 16384
+  "gpt-4":
+    context_length: 8192
+    max_output_tokens: 4096
+  "gpt-4-turbo":
+    context_length: 128000
+    max_output_tokens: 4096

src/skillspector/providers/azure_openai/provider.py

@@ -0,0 +1,92 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Azure OpenAI provider — enterprise Azure-hosted OpenAI deployments.
+
+Uses ``AzureChatOpenAI`` from ``langchain_openai`` which handles Azure's
+deployment-based routing and ``api-version`` query parameter natively.
+
+Required env vars:
+    AZURE_OPENAI_ENDPOINT    — Azure resource endpoint
+    AZURE_OPENAI_API_KEY     — Azure API key
+
+Optional env vars:
+    AZURE_OPENAI_DEPLOYMENT  — deployment name (defaults to model label)
+    AZURE_OPENAI_API_VERSION — API version (defaults to ``2024-06-01``)
+"""
+
+from __future__ import annotations
+
+import os
+from pathlib import Path
+
+from langchain_core.language_models.chat_models import BaseChatModel
+from langchain_openai import AzureChatOpenAI
+from pydantic import SecretStr
+
+from skillspector.providers import registry
+
+REGISTRY_PATH = str(Path(__file__).with_name("model_registry.yaml"))
+
+
+class AzureOpenAIProvider:
+    """Azure OpenAI credentials + bundled-YAML metadata provider."""
+
+    DEFAULT_MODEL = "gpt-4o"
+    SLOT_DEFAULTS: dict[str, str] = {}
+
+    def resolve_credentials(self) -> tuple[str, str | None] | None:
+        """Return ``(api_key, endpoint)`` from Azure OpenAI env vars."""
+        api_key = os.environ.get("AZURE_OPENAI_API_KEY", "").strip()
+        endpoint = os.environ.get("AZURE_OPENAI_ENDPOINT", "").strip()
+        if not api_key or not endpoint:
+            return None
+        return api_key, endpoint
+
+    def create_chat_model(
+        self,
+        model: str,
+        *,
+        max_tokens: int,
+        timeout: float | None = 120,
+    ) -> BaseChatModel | None:
+        """Create ``AzureChatOpenAI`` using Azure-specific credentials."""
+        creds = self.resolve_credentials()
+        if creds is None:
+            return None
+
+        api_key, endpoint = creds
+        deployment = os.environ.get("AZURE_OPENAI_DEPLOYMENT", "").strip() or model
+        api_version = os.environ.get("AZURE_OPENAI_API_VERSION", "").strip() or "2024-06-01"
+
+        return AzureChatOpenAI(
+            azure_endpoint=endpoint,
+            azure_deployment=deployment,
+            api_key=SecretStr(api_key),
+            api_version=api_version,
+            max_tokens=max_tokens,
+            timeout=timeout,
+        )
+
+    def get_context_length(self, model: str) -> int | None:
+        return registry.lookup_context_length(REGISTRY_PATH, model)
+
+    def get_max_output_tokens(self, model: str) -> int | None:
+        return registry.lookup_max_output_tokens(REGISTRY_PATH, model)
+
+    def resolve_model(self, slot: str = "default") -> str:
+        """Resolve model: ``SKILLSPECTOR_MODEL`` env > slot default > ``DEFAULT_MODEL``."""
+        user_input = os.environ.get("SKILLSPECTOR_MODEL", "").strip()
+        return user_input or self.SLOT_DEFAULTS.get(slot, "") or self.DEFAULT_MODEL

src/skillspector/providers/ollama/__init__.py

@@ -0,0 +1,20 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Ollama provider package (local/self-hosted OpenAI-compatible endpoint)."""
+
+from .provider import OLLAMA_DEFAULT_BASE_URL, REGISTRY_PATH, OllamaProvider
+
+__all__ = ["OLLAMA_DEFAULT_BASE_URL", "REGISTRY_PATH", "OllamaProvider"]

src/skillspector/providers/ollama/model_registry.yaml

@@ -0,0 +1,29 @@
+# Token-budget metadata for the OllamaProvider (local/self-hosted LLMs).
+# Bundled with the package; consulted whenever the active provider is
+# OllamaProvider.
+#
+# Format:
+#   models:
+#     "<model-label>":
+#       context_length: <int>          # total context window in tokens (required)
+#       max_output_tokens: <int>       # model's max output cap (optional)
+
+models:
+  "llama3.1:8b":
+    context_length: 131072
+    max_output_tokens: 4096
+  "llama3.1:70b":
+    context_length: 131072
+    max_output_tokens: 4096
+  "mistral:7b":
+    context_length: 32768
+    max_output_tokens: 4096
+  "qwen2.5:7b":
+    context_length: 131072
+    max_output_tokens: 8192
+  "gemma2:9b":
+    context_length: 8192
+    max_output_tokens: 4096
+  "deepseek-v3:latest":
+    context_length: 65536
+    max_output_tokens: 8192

src/skillspector/providers/ollama/provider.py

@@ -0,0 +1,80 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Ollama provider — local/self-hosted LLM via OpenAI-compatible API.
+
+Ollama serves an OpenAI-compatible endpoint at ``http://localhost:11434/v1``
+by default.  No API key is required; ``ChatOpenAI`` needs a non-empty
+placeholder string.
+
+Override the endpoint via ``OLLAMA_BASE_URL`` for remote Ollama instances.
+"""
+
+from __future__ import annotations
+
+import os
+from pathlib import Path
+
+from langchain_core.language_models.chat_models import BaseChatModel
+
+from skillspector.providers import registry
+from skillspector.providers.chat_models import create_openai_compatible_chat_model
+
+OLLAMA_DEFAULT_BASE_URL = "http://localhost:11434/v1"
+
+REGISTRY_PATH = str(Path(__file__).with_name("model_registry.yaml"))
+
+
+class OllamaProvider:
+    """Ollama credentials + bundled-YAML metadata provider."""
+
+    DEFAULT_MODEL = "llama3.1:8b"
+    SLOT_DEFAULTS: dict[str, str] = {}
+
+    def resolve_credentials(self) -> tuple[str, str | None] | None:
+        """Return ``(api_key, base_url)`` for the Ollama endpoint.
+
+        Ollama does not require an API key but ``ChatOpenAI`` needs a
+        non-empty string — we supply the literal ``"ollama"``.
+        Always returns credentials (Ollama is assumed available when selected).
+        """
+        base_url = os.environ.get("OLLAMA_BASE_URL", "").strip() or OLLAMA_DEFAULT_BASE_URL
+        return "ollama", base_url
+
+    def create_chat_model(
+        self,
+        model: str,
+        *,
+        max_tokens: int,
+        timeout: float | None = 120,
+    ) -> BaseChatModel | None:
+        """Create ``ChatOpenAI`` pointing at the Ollama endpoint."""
+        return create_openai_compatible_chat_model(
+            model=model,
+            credentials=self.resolve_credentials(),
+            max_tokens=max_tokens,
+            timeout=timeout,
+        )
+
+    def get_context_length(self, model: str) -> int | None:
+        return registry.lookup_context_length(REGISTRY_PATH, model)
+
+    def get_max_output_tokens(self, model: str) -> int | None:
+        return registry.lookup_max_output_tokens(REGISTRY_PATH, model)
+
+    def resolve_model(self, slot: str = "default") -> str:
+        """Resolve model: ``SKILLSPECTOR_MODEL`` env > slot default > ``DEFAULT_MODEL``."""
+        user_input = os.environ.get("SKILLSPECTOR_MODEL", "").strip()
+        return user_input or self.SLOT_DEFAULTS.get(slot, "") or self.DEFAULT_MODEL

src/skillspector/providers/openai_compatible/__init__.py

@@ -0,0 +1,20 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+# http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+"""Generic OpenAI-compatible provider package (Groq, Together AI, Mistral, etc.)."""
+
+from .provider import REGISTRY_PATH, OpenAICompatibleProvider
+
+__all__ = ["REGISTRY_PATH", "OpenAICompatibleProvider"]

src/skillspector/providers/openai_compatible/model_registry.yaml

@@ -0,0 +1,43 @@
+# Token-budget metadata for the OpenAICompatibleProvider (Groq, Together AI,
+# Mistral, DeepInfra, Fireworks, LiteLLM, and other OpenAI-compatible
+# endpoints).  Bundled with the package; consulted whenever the active
+# provider is OpenAICompatibleProvider.
+#
+# Format:
+#   models:
+#     "<model-label>":
+#       context_length: <int>          # total context window in tokens (required)
+#       max_output_tokens: <int>       # model's max output cap (optional)
+
+models:
+  # Groq
+  "llama-3.1-70b-versatile":
+    context_length: 131072
+    max_output_tokens: 8192
+  "llama-3.1-8b-instant":
+    context_length: 131072
+    max_output_tokens: 8192
+  "mixtral-8x7b-32768":
+    context_length: 32768
+    max_output_tokens: 4096
+
+  # Together AI
+  "meta-llama/Meta-Llama-3.1-70B-Instruct-Turbo":
+    context_length: 131072
+    max_output_tokens: 4096
+  "meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo":
+    context_length: 131072
+    max_output_tokens: 4096
+
+  # Mistral
+  "mistral-large-latest":
+    context_length: 131072
+    max_output_tokens: 8192
+  "mistral-small-latest":
+    context_length: 131072
+    max_output_tokens: 8192
+
+  # DeepInfra / generic
+  "deepseek-ai/DeepSeek-V3":
+    context_length: 65536
+    max_output_tokens: 8192