Merge pull request #135 from stacklok/ollama-support

Ollama provider

docs/cli.md

@@ -61,6 +61,11 @@ codegate serve [OPTIONS]
   - Base URL for Anthropic provider
   - Overrides configuration file and environment variables
 
+- `--ollama-url TEXT`: Ollama provider URL (default: http://localhost:11434)
+  - Optional
+  - Base URL for Ollama provider (/api path is added automatically)
+  - Overrides configuration file and environment variables
+
 ### show-prompts
 
 Display the loaded system prompts:
@@ -120,6 +125,11 @@ Start server with custom vLLM endpoint:
 codegate serve --vllm-url https://vllm.example.com
 ```
 
+Start server with custom Ollama endpoint:
+```bash
+codegate serve --ollama-url http://localhost:11434
+```
+
 Show default system prompts:
 ```bash
 codegate show-prompts

docs/configuration.md

@@ -20,6 +20,7 @@ The configuration system in Codegate is managed through the `Config` class in `c
   - vLLM: "http://localhost:8000"
   - OpenAI: "https://api.openai.com/v1"
   - Anthropic: "https://api.anthropic.com/v1"
+  - Ollama: "http://localhost:11434"
 
 ## Configuration Methods
 
@@ -41,6 +42,7 @@ provider_urls:
   vllm: "https://vllm.example.com"
   openai: "https://api.openai.com/v1"
   anthropic: "https://api.anthropic.com/v1"
+  ollama: "http://localhost:11434"
 ```
 
 ### From Environment Variables
@@ -55,6 +57,7 @@ Environment variables are automatically loaded with these mappings:
 - `CODEGATE_PROVIDER_VLLM_URL`: vLLM provider URL
 - `CODEGATE_PROVIDER_OPENAI_URL`: OpenAI provider URL
 - `CODEGATE_PROVIDER_ANTHROPIC_URL`: Anthropic provider URL
+- `CODEGATE_PROVIDER_OLLAMA_URL`: Ollama provider URL
 
 ```python
 config = Config.from_env()
@@ -72,21 +75,25 @@ Provider URLs can be configured in several ways:
      vllm: "https://vllm.example.com"  # /v1 path is added automatically
      openai: "https://api.openai.com/v1"
      anthropic: "https://api.anthropic.com/v1"
+     ollama: "http://localhost:11434"  # /api path is added automatically
    ```
 
 2. Via Environment Variables:
    ```bash
    export CODEGATE_PROVIDER_VLLM_URL=https://vllm.example.com
    export CODEGATE_PROVIDER_OPENAI_URL=https://api.openai.com/v1
    export CODEGATE_PROVIDER_ANTHROPIC_URL=https://api.anthropic.com/v1
+   export CODEGATE_PROVIDER_OLLAMA_URL=http://localhost:11434
    ```
 
 3. Via CLI Flags:
    ```bash
-   codegate serve --vllm-url https://vllm.example.com
+   codegate serve --vllm-url https://vllm.example.com --ollama-url http://localhost:11434
    ```
 
-Note: For the vLLM provider, the /v1 path is automatically appended to the base URL if not present.
+Note: 
+- For the vLLM provider, the /v1 path is automatically appended to the base URL if not present.
+- For the Ollama provider, the /api path is automatically appended to the base URL if not present.
 
 ### Log Levels
 

docs/development.md

@@ -1,5 +1,3 @@
-# Development Guide
-
 This guide provides comprehensive information for developers working on the Codegate project.
 
 ## Project Overview
@@ -157,6 +155,13 @@ Codegate supports multiple AI providers through a modular provider system.
    - Default URL: https://api.anthropic.com/v1
    - Anthropic Claude API implementation
 
+4. **Ollama Provider**
+   - Default URL: http://localhost:11434
+   - Endpoints:
+     * Native Ollama API: `/ollama/api/chat`
+     * OpenAI-compatible: `/ollama/chat/completions`
+     ```
+
 ### Configuring Providers
 
 Provider URLs can be configured through:
@@ -167,18 +172,20 @@ Provider URLs can be configured through:
      vllm: "https://vllm.example.com"
      openai: "https://api.openai.com/v1"
      anthropic: "https://api.anthropic.com/v1"
+     ollama: "http://localhost:11434"  # /api path added automatically
    ```
 
 2. Environment variables:
    ```bash
    export CODEGATE_PROVIDER_VLLM_URL=https://vllm.example.com
    export CODEGATE_PROVIDER_OPENAI_URL=https://api.openai.com/v1
    export CODEGATE_PROVIDER_ANTHROPIC_URL=https://api.anthropic.com/v1
+   export CODEGATE_PROVIDER_OLLAMA_URL=http://localhost:11434
    ```
 
 3. CLI flags:
    ```bash
-   codegate serve --vllm-url https://vllm.example.com
+   codegate serve --vllm-url https://vllm.example.com --ollama-url http://localhost:11434
    ```
 
 ### Implementing New Providers
@@ -276,4 +283,4 @@ codegate serve --prompts my-prompts.yaml
 codegate serve --vllm-url https://vllm.example.com
 ```
 
-See [CLI Documentation](cli.md) for detailed command information.
+See [CLI Documentation](cli.md) for detailed command information.

src/codegate/config.py

@@ -19,6 +19,7 @@
     "openai": "https://api.openai.com/v1",
     "anthropic": "https://api.anthropic.com/v1",
     "vllm": "http://localhost:8000",  # Base URL without /v1 path
+    "ollama": "http://localhost:11434",  # Default Ollama server URL
 }
 
 

src/codegate/pipeline/extract_snippets/extract_snippets.py

@@ -13,6 +13,7 @@
 
 logger = structlog.get_logger("codegate")
 
+
 def ecosystem_from_filepath(filepath: str) -> Optional[str]:
     """
     Determine language from filepath.

src/codegate/providers/__init__.py

@@ -1,5 +1,6 @@
 from codegate.providers.anthropic.provider import AnthropicProvider
 from codegate.providers.base import BaseProvider
+from codegate.providers.ollama.provider import OllamaProvider
 from codegate.providers.openai.provider import OpenAIProvider
 from codegate.providers.registry import ProviderRegistry
 from codegate.providers.vllm.provider import VLLMProvider
@@ -10,4 +11,5 @@
     "OpenAIProvider",
     "AnthropicProvider",
     "VLLMProvider",
+    "OllamaProvider",
 ]

src/codegate/providers/ollama/__init__.py

@@ -0,0 +1,3 @@
+from codegate.providers.ollama.provider import OllamaProvider
+
+__all__ = ["OllamaProvider"]

src/codegate/providers/ollama/adapter.py

@@ -0,0 +1,86 @@
+from typing import Any, Dict
+
+from litellm import ChatCompletionRequest
+
+from codegate.providers.normalizer.base import ModelInputNormalizer, ModelOutputNormalizer
+
+
+class OllamaInputNormalizer(ModelInputNormalizer):
+    def __init__(self):
+        super().__init__()
+
+    def normalize(self, data: Dict) -> ChatCompletionRequest:
+        """
+        Normalize the input data to the format expected by Ollama.
+        """
+        # Make a copy of the data to avoid modifying the original
+        normalized_data = data.copy()
+
+        # Format the model name
+        if "model" in normalized_data:
+            normalized_data["model"] = normalized_data["model"].strip()
+
+        # Convert messages format if needed
+        if "messages" in normalized_data:
+            messages = normalized_data["messages"]
+            converted_messages = []
+            for msg in messages:
+                if isinstance(msg.get("content"), list):
+                    # Convert list format to string
+                    content_parts = []
+                    for part in msg["content"]:
+                        if part.get("type") == "text":
+                            content_parts.append(part["text"])
+                    msg = msg.copy()
+                    msg["content"] = " ".join(content_parts)
+                converted_messages.append(msg)
+            normalized_data["messages"] = converted_messages
+
+        # Ensure the base_url ends with /api if provided
+        if "base_url" in normalized_data:
+            base_url = normalized_data["base_url"].rstrip("/")
+            if not base_url.endswith("/api"):
+                normalized_data["base_url"] = f"{base_url}/api"
+
+        return ChatCompletionRequest(**normalized_data)
+
+    def denormalize(self, data: ChatCompletionRequest) -> Dict:
+        """
+        Convert back to raw format for the API request
+        """
+        return data
+
+
+class OllamaOutputNormalizer(ModelOutputNormalizer):
+    def __init__(self):
+        super().__init__()
+
+    def normalize_streaming(
+        self,
+        model_reply: Any,
+    ) -> Any:
+        """
+        Pass through Ollama response
+        """
+        return model_reply
+
+    def normalize(self, model_reply: Any) -> Any:
+        """
+        Pass through Ollama response
+        """
+        return model_reply
+
+    def denormalize(self, normalized_reply: Any) -> Any:
+        """
+        Pass through Ollama response
+        """
+        return normalized_reply
+
+    def denormalize_streaming(
+        self,
+        normalized_reply: Any,
+    ) -> Any:
+        """
+        Pass through Ollama response
+        """
+        return normalized_reply