feat: add unified provider architecture with Amazon Bedrock support

Refactored LLM provider infrastructure to support sustainable additions of new providers with both embedding and text generation capabilities.

## Major Changes

### Unified Provider Architecture (ADR-015)
- Created `nextcloud_mcp_server/providers/` with unified Provider ABC
- Providers now support optional capabilities (embeddings and/or generation)
- Auto-detection registry with priority: Bedrock → Ollama → Simple
- Backward compatible - existing code continues to work

### New Providers
- **BedrockProvider**: Full Amazon Bedrock integration
  - Embeddings: Titan Embed, Cohere Embed models
  - Generation: Claude, Llama, Titan Text, Mistral models
  - Model-specific request/response handling
  - AWS credential chain integration
- **OllamaProvider**: Migrated with both capabilities support
- **AnthropicProvider**: Moved from test code to production providers
- **SimpleProvider**: Migrated in-memory fallback provider

### Breaking Changes
None - full backward compatibility maintained:
- `embedding.get_embedding_service()` still works
- RAG evaluation tests updated to use unified providers
- All existing tests pass (127 unit tests)

### Testing
- Added 9 comprehensive Bedrock unit tests with mocked boto3
- All existing unit tests pass
- Type checking (ty) and linting (ruff) pass
- Verified backward compatibility

### Documentation
- `docs/ADR-015-unified-provider-architecture.md`: Comprehensive ADR
- `docs/bedrock-setup.md`: AWS setup guide with IAM permissions
- `CLAUDE.md`: Updated with provider architecture section

### Dependencies
- Added `boto3>=1.35.0` to dev dependencies (optional)

## Environment Variables

### Bedrock
- `AWS_REGION`: AWS region (e.g., "us-east-1")
- `BEDROCK_EMBEDDING_MODEL`: Model ID for embeddings
- `BEDROCK_GENERATION_MODEL`: Model ID for generation
- `AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY`: Optional credentials

### Ollama
- `OLLAMA_BASE_URL`: API URL
- `OLLAMA_EMBEDDING_MODEL`: Embedding model (default: "nomic-embed-text")
- `OLLAMA_GENERATION_MODEL`: Generation model

## AWS Bedrock Permissions Required

Minimal IAM policy:
```json
{
  "Effect": "Allow",
  "Action": ["bedrock:InvokeModel"],
  "Resource": ["arn:aws:bedrock:*::foundation-model/*"]
}
```

See `docs/bedrock-setup.md` for detailed setup instructions.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

CLAUDE.md

@@ -61,8 +61,60 @@ This file provides guidance to Claude Code (claude.ai/code) when working with co
 - `nextcloud_mcp_server/server/` - MCP tool/resource definitions
 - `nextcloud_mcp_server/auth/` - OAuth/OIDC authentication
 - `nextcloud_mcp_server/models/` - Pydantic response models
+- `nextcloud_mcp_server/providers/` - Unified LLM provider infrastructure (embeddings + generation)
 - `tests/` - Layered test suite (unit, smoke, integration, load)
 
+### Provider Architecture (ADR-015)
+
+**Unified Provider System** for embeddings and text generation:
+
+**Location:** `nextcloud_mcp_server/providers/`
+- `base.py` - `Provider` ABC with optional capabilities
+- `registry.py` - Auto-detection and factory pattern
+- `ollama.py` - Ollama provider (embeddings + generation)
+- `anthropic.py` - Anthropic provider (generation only)
+- `bedrock.py` - Amazon Bedrock provider (embeddings + generation)
+- `simple.py` - Simple in-memory provider (embeddings only, fallback)
+
+**Usage:**
+```python
+from nextcloud_mcp_server.providers import get_provider
+
+provider = get_provider()  # Auto-detects from environment
+
+# Check capabilities
+if provider.supports_embeddings:
+    embeddings = await provider.embed_batch(texts)
+
+if provider.supports_generation:
+    text = await provider.generate("prompt", max_tokens=500)
+```
+
+**Environment Variables:**
+
+Bedrock:
+- `AWS_REGION` - AWS region (e.g., "us-east-1")
+- `BEDROCK_EMBEDDING_MODEL` - Embedding model ID (e.g., "amazon.titan-embed-text-v2:0")
+- `BEDROCK_GENERATION_MODEL` - Generation model ID (e.g., "anthropic.claude-3-sonnet-20240229-v1:0")
+- `AWS_ACCESS_KEY_ID`, `AWS_SECRET_ACCESS_KEY` - Optional, uses AWS credential chain
+
+Ollama:
+- `OLLAMA_BASE_URL` - API URL (e.g., "http://localhost:11434")
+- `OLLAMA_EMBEDDING_MODEL` - Embedding model (default: "nomic-embed-text")
+- `OLLAMA_GENERATION_MODEL` - Generation model (e.g., "llama3.2:1b")
+- `OLLAMA_VERIFY_SSL` - SSL verification (default: "true")
+
+Simple (fallback, no config needed):
+- `SIMPLE_EMBEDDING_DIMENSION` - Dimension (default: 384)
+
+**Auto-Detection Priority:** Bedrock → Ollama → Simple
+
+**Backward Compatibility:**
+- Old code using `nextcloud_mcp_server.embedding.get_embedding_service()` still works
+- `EmbeddingService` now wraps `get_provider()` internally
+
+**For Details:** See `docs/ADR-015-unified-provider-architecture.md`
+
 ## Development Commands (Quick Reference)
 
 ### Testing

docs/ADR-015-unified-provider-architecture.md

@@ -0,0 +1,380 @@
+# ADR-015: Unified Provider Architecture for Embeddings and Text Generation
+
+**Status:** Accepted
+**Date:** 2025-01-16
+**Deciders:** Development Team
+**Related:** ADR-003 (Vector Database), ADR-008 (MCP Sampling), ADR-013 (RAG Evaluation)
+
+## Context
+
+Prior to this refactoring, the codebase had two separate provider systems:
+
+1. **Embedding Providers** (`nextcloud_mcp_server/embedding/`)
+   - Used `EmbeddingProvider` ABC with methods: `embed()`, `embed_batch()`, `get_dimension()`
+   - Had auto-detection via `EmbeddingService._detect_provider()`
+   - Used for semantic search and vector indexing (production)
+
+2. **LLM Providers** (`tests/rag_evaluation/llm_providers.py`)
+   - Used `LLMProvider` Protocol with method: `generate()`
+   - Had separate factory function `create_llm_provider()`
+   - Used only for RAG evaluation tests (not production)
+
+This fragmentation created several problems:
+
+### Problems with Dual Provider Systems
+
+1. **Code Duplication**
+   - Ollama configuration appeared in both `embedding/service.py` and `tests/rag_evaluation/llm_providers.py`
+   - Similar provider detection logic in multiple places
+   - Separate singleton patterns for each system
+
+2. **Limited Extensibility**
+   - Hard-coded provider detection in `EmbeddingService._detect_provider()`
+   - No support for providers that offer both capabilities (like Bedrock)
+   - Adding new providers required modifying multiple files
+
+3. **Inconsistent Patterns**
+   - BM25 provider didn't follow `EmbeddingProvider` ABC
+   - Different method names across providers (`embed` vs `encode`)
+   - ABC vs Protocol for type checking
+
+4. **Difficult Scaling**
+   - Adding Amazon Bedrock (our third provider) would exacerbate all issues
+   - No clear path for future providers (OpenAI, Cohere, etc.)
+
+### Amazon Bedrock Requirements
+
+Bedrock naturally supports **both** embeddings and text generation:
+- **Embeddings**: `amazon.titan-embed-text-v1/v2`, `cohere.embed-*`
+- **Text Generation**: `anthropic.claude-*`, `meta.llama3-*`, `amazon.titan-text-*`
+- **Unified API**: Single `invoke_model()` method via bedrock-runtime
+
+This made it the perfect opportunity to establish a unified provider architecture.
+
+## Decision
+
+We refactored the provider infrastructure to use a **unified Provider ABC** with optional capabilities:
+
+### 1. Unified Provider Interface
+
+**New Structure:**
+```
+nextcloud_mcp_server/providers/
+├── __init__.py
+├── base.py              # Provider ABC with optional capabilities
+├── registry.py          # Auto-detection and factory
+├── ollama.py            # Supports both embedding + generation
+├── anthropic.py         # Generation only
+├── bedrock.py           # Supports both embedding + generation
+└── simple.py            # Embedding only (testing fallback)
+```
+
+**Base Class (`providers/base.py`):**
+```python
+class Provider(ABC):
+    @property
+    @abstractmethod
+    def supports_embeddings(self) -> bool:
+        """Whether this provider supports embedding generation."""
+        pass
+
+    @property
+    @abstractmethod
+    def supports_generation(self) -> bool:
+        """Whether this provider supports text generation."""
+        pass
+
+    @abstractmethod
+    async def embed(self, text: str) -> list[float]:
+        """Generate embedding (raises NotImplementedError if not supported)."""
+        pass
+
+    @abstractmethod
+    async def embed_batch(self, texts: list[str]) -> list[list[float]]:
+        """Generate batch embeddings (raises NotImplementedError if not supported)."""
+        pass
+
+    @abstractmethod
+    def get_dimension(self) -> int:
+        """Get embedding dimension (raises NotImplementedError if not supported)."""
+        pass
+
+    @abstractmethod
+    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
+        """Generate text (raises NotImplementedError if not supported)."""
+        pass
+
+    @abstractmethod
+    async def close(self) -> None:
+        """Close provider and release resources."""
+        pass
+```
+
+### 2. Provider Registry
+
+**Auto-Detection Priority** (`providers/registry.py`):
+```python
+class ProviderRegistry:
+    @staticmethod
+    def create_provider() -> Provider:
+        # 1. Bedrock (AWS_REGION or BEDROCK_*_MODEL)
+        # 2. Ollama (OLLAMA_BASE_URL)
+        # 3. Simple (fallback)
+```
+
+**Environment Variables:**
+
+**Bedrock:**
+- `AWS_REGION`: AWS region (e.g., "us-east-1")
+- `AWS_ACCESS_KEY_ID`: AWS access key (optional, uses credential chain)
+- `AWS_SECRET_ACCESS_KEY`: AWS secret key (optional)
+- `BEDROCK_EMBEDDING_MODEL`: Model ID for embeddings (e.g., "amazon.titan-embed-text-v2:0")
+- `BEDROCK_GENERATION_MODEL`: Model ID for text generation (e.g., "anthropic.claude-3-sonnet-20240229-v1:0")
+
+**Ollama:**
+- `OLLAMA_BASE_URL`: Ollama API base URL (e.g., "http://localhost:11434")
+- `OLLAMA_EMBEDDING_MODEL`: Model for embeddings (default: "nomic-embed-text")
+- `OLLAMA_GENERATION_MODEL`: Model for text generation (e.g., "llama3.2:1b")
+- `OLLAMA_VERIFY_SSL`: Verify SSL certificates (default: "true")
+
+**Simple (no configuration, fallback):**
+- `SIMPLE_EMBEDDING_DIMENSION`: Embedding dimension (default: 384)
+
+### 3. Backward Compatibility
+
+**Old Code Continues to Work:**
+```python
+# Old way (still works)
+from nextcloud_mcp_server.embedding import get_embedding_service
+
+service = get_embedding_service()  # Returns singleton Provider
+embeddings = await service.embed_batch(texts)
+```
+
+**New Way (recommended):**
+```python
+# New way (cleaner)
+from nextcloud_mcp_server.providers import get_provider
+
+provider = get_provider()  # Returns singleton Provider
+embeddings = await provider.embed_batch(texts)
+
+# Can also use generation if provider supports it
+if provider.supports_generation:
+    text = await provider.generate("prompt")
+```
+
+**Migration Path:**
+- `embedding/service.py` now wraps `providers.get_provider()` for compatibility
+- `tests/rag_evaluation/llm_providers.py` now uses unified providers
+- Old imports still work, marked as deprecated in docstrings
+
+### 4. Amazon Bedrock Implementation
+
+**Features:**
+- Supports both embeddings and text generation
+- Model-specific request/response handling for:
+  - Titan Embed (amazon.titan-embed-text-*)
+  - Cohere Embed (cohere.embed-*)
+  - Claude (anthropic.claude-*)
+  - Llama (meta.llama3-*)
+  - Titan Text (amazon.titan-text-*)
+  - Mistral (mistral.*)
+- Uses boto3 bedrock-runtime client
+- Graceful degradation if boto3 not installed
+- Async implementation matching existing patterns
+
+**Model-Specific Handling:**
+```python
+# Bedrock embedding request (Titan)
+{"inputText": text}
+
+# Bedrock generation request (Claude)
+{
+    "anthropic_version": "bedrock-2023-05-31",
+    "max_tokens": max_tokens,
+    "temperature": 0.7,
+    "messages": [{"role": "user", "content": prompt}]
+}
+```
+
+## Consequences
+
+### Positive
+
+1. **Sustainable Provider Additions**
+   - New providers only need to implement `Provider` ABC
+   - Auto-detection via environment variables
+   - No modifications to existing code required
+
+2. **Code Consolidation**
+   - Single provider interface instead of two
+   - Unified configuration pattern
+   - Eliminated duplication
+
+3. **Better Extensibility**
+   - Providers can support one or both capabilities
+   - Clear capability detection via properties
+   - Registry pattern simplifies auto-detection
+
+4. **Improved Testing**
+   - RAG evaluation can use any provider (Ollama, Anthropic, Bedrock)
+   - Comprehensive unit tests for all providers
+   - Mocked boto3 tests for Bedrock
+
+5. **Production-Ready Bedrock Support**
+   - Full embedding and generation support
+   - Multiple model families supported
+   - AWS credential chain integration
+
+### Neutral
+
+1. **Optional Boto3 Dependency**
+   - boto3 is dev dependency only (not required for core functionality)
+   - Bedrock provider gracefully fails if boto3 not installed
+   - Users who want Bedrock must `pip install boto3`
+
+2. **Capability Properties**
+   - All providers must implement capability properties
+   - Methods raise `NotImplementedError` if capability not supported
+   - Clear error messages guide users to alternatives
+
+### Negative
+
+1. **Migration Effort**
+   - Existing code must be migrated to new imports (optional, backward compatible)
+   - Documentation needs updating
+   - Users must learn new environment variables
+
+2. **Increased Complexity**
+   - Provider base class has more methods (embedding + generation)
+   - More environment variables to configure
+   - Capability detection adds runtime checks
+
+## Implementation
+
+### Files Created
+
+**New Provider Infrastructure:**
+- `nextcloud_mcp_server/providers/__init__.py`
+- `nextcloud_mcp_server/providers/base.py`
+- `nextcloud_mcp_server/providers/registry.py`
+- `nextcloud_mcp_server/providers/ollama.py`
+- `nextcloud_mcp_server/providers/anthropic.py`
+- `nextcloud_mcp_server/providers/bedrock.py`
+- `nextcloud_mcp_server/providers/simple.py`
+
+**Tests:**
+- `tests/unit/providers/__init__.py`
+- `tests/unit/providers/test_bedrock.py` (9 unit tests)
+
+**Documentation:**
+- `docs/ADR-015-unified-provider-architecture.md` (this file)
+
+### Files Modified
+
+**Backward Compatibility:**
+- `nextcloud_mcp_server/embedding/service.py` - Now wraps `get_provider()`
+- `tests/rag_evaluation/llm_providers.py` - Uses unified providers
+
+**Dependencies:**
+- `pyproject.toml` - Added `boto3>=1.35.0` to dev dependencies
+
+### Testing Results
+
+**Unit Tests:** 127 passed (including 9 new Bedrock tests)
+**Type Checking:** All checks passed (ty)
+**Linting:** All checks passed (ruff)
+**Backward Compatibility:** Verified - existing embedding tests work
+
+## Alternatives Considered
+
+### Alternative 1: Keep Separate Provider Systems
+
+**Pros:**
+- No refactoring needed
+- Simpler short-term
+
+**Cons:**
+- Bedrock would need to be implemented twice
+- Continued code duplication
+- No long-term scalability
+
+**Decision:** Rejected - technical debt would continue to grow
+
+### Alternative 2: Separate Embedding and Generation Providers
+
+Use composition instead of unified interface:
+```python
+class CombinedProvider:
+    def __init__(self, embedding: EmbeddingProvider, generation: LLMProvider):
+        self.embedding = embedding
+        self.generation = generation
+```
+
+**Pros:**
+- Clearer separation of concerns
+- Simpler individual providers
+
+**Cons:**
+- Bedrock and Ollama naturally do both - artificial separation
+- More complex configuration (two providers to configure)
+- More boilerplate code
+
+**Decision:** Rejected - unified interface better matches provider capabilities
+
+### Alternative 3: Plugin System
+
+Dynamic provider registration via entry points:
+```python
+# setup.py
+entry_points={
+    'nextcloud_mcp.providers': [
+        'ollama = nextcloud_mcp_server.providers.ollama:OllamaProvider',
+        'bedrock = nextcloud_mcp_server.providers.bedrock:BedrockProvider',
+    ]
+}
+```
+
+**Pros:**
+- Most extensible
+- Third-party providers possible
+
+**Cons:**
+- Over-engineered for current needs
+- Added complexity
+- No immediate benefit
+
+**Decision:** Deferred - can add later if needed
+
+## Future Work
+
+1. **Additional Providers**
+   - OpenAI (embeddings + generation)
+   - Cohere (embeddings + generation)
+   - Google Vertex AI
+   - Azure OpenAI
+
+2. **Provider Features**
+   - Streaming generation support
+   - Batch API optimization (when available)
+   - Model-specific optimizations
+   - Cost tracking and metrics
+
+3. **Configuration Improvements**
+   - Provider profiles (development, production)
+   - Model aliasing (e.g., "small", "large")
+   - Fallback provider chains
+
+4. **Testing**
+   - Integration tests with real Bedrock endpoints
+   - Performance benchmarking across providers
+   - Cost comparison analysis
+
+## References
+
+- [boto3 Bedrock Runtime Documentation](https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/bedrock-runtime.html)
+- [Amazon Bedrock User Guide](https://docs.aws.amazon.com/bedrock/latest/userguide/what-is-bedrock.html)
+- ADR-003: Vector Database and Semantic Search
+- ADR-008: MCP Sampling for Semantic Search
+- ADR-013: RAG Evaluation Framework

docs/bedrock-setup.md

@@ -0,0 +1,338 @@
+# Amazon Bedrock Setup Guide
+
+This guide covers how to configure the Nextcloud MCP Server to use Amazon Bedrock for embeddings and text generation.
+
+## Prerequisites
+
+1. **AWS Account** with access to Amazon Bedrock
+2. **boto3 library** installed: `pip install boto3` or `uv sync --group dev`
+3. **Model Access** - Request access to models in AWS Bedrock console
+
+## Required AWS Permissions
+
+### IAM Policy for Bedrock Access
+
+The AWS IAM user or role needs the following permissions:
+
+```json
+{
+  "Version": "2012-10-17",
+  "Statement": [
+    {
+      "Sid": "BedrockInvokeModels",
+      "Effect": "Allow",
+      "Action": [
+        "bedrock:InvokeModel",
+        "bedrock:InvokeModelWithResponseStream"
+      ],
+      "Resource": [
+        "arn:aws:bedrock:*::foundation-model/*"
+      ]
+    }
+  ]
+}
+```
+
+### Minimal Permissions (Production)
+
+For production deployments, restrict to specific models:
+
+```json
+{
+  "Version": "2012-10-17",
+  "Statement": [
+    {
+      "Sid": "BedrockEmbeddings",
+      "Effect": "Allow",
+      "Action": [
+        "bedrock:InvokeModel"
+      ],
+      "Resource": [
+        "arn:aws:bedrock:us-east-1::foundation-model/amazon.titan-embed-text-v2:0"
+      ]
+    },
+    {
+      "Sid": "BedrockGeneration",
+      "Effect": "Allow",
+      "Action": [
+        "bedrock:InvokeModel"
+      ],
+      "Resource": [
+        "arn:aws:bedrock:us-east-1::foundation-model/anthropic.claude-3-sonnet-20240229-v1:0"
+      ]
+    }
+  ]
+}
+```
+
+### Additional Permissions (Optional)
+
+For advanced use cases:
+
+```json
+{
+  "Version": "2012-10-17",
+  "Statement": [
+    {
+      "Sid": "BedrockListModels",
+      "Effect": "Allow",
+      "Action": [
+        "bedrock:ListFoundationModels",
+        "bedrock:GetFoundationModel"
+      ],
+      "Resource": "*"
+    },
+    {
+      "Sid": "BedrockAsyncInvoke",
+      "Effect": "Allow",
+      "Action": [
+        "bedrock:InvokeModelAsync",
+        "bedrock:GetAsyncInvoke",
+        "bedrock:ListAsyncInvokes"
+      ],
+      "Resource": [
+        "arn:aws:bedrock:*::foundation-model/*"
+      ]
+    }
+  ]
+}
+```
+
+## Model Access
+
+Before using Bedrock models, you must request access in the AWS Console:
+
+1. Navigate to **Amazon Bedrock** → **Model access**
+2. Click **Manage model access**
+3. Select models you want to use:
+   - **Embeddings:** Amazon Titan Embed Text, Cohere Embed
+   - **Text Generation:** Anthropic Claude, Meta Llama, Amazon Titan Text
+4. Click **Request model access**
+5. Wait for approval (usually instant for most models)
+
+## Supported Models
+
+### Embedding Models
+
+| Provider | Model ID | Dimensions | Best For |
+|----------|----------|------------|----------|
+| Amazon Titan | `amazon.titan-embed-text-v1` | 1,536 | General purpose |
+| Amazon Titan | `amazon.titan-embed-text-v2:0` | 1,024 | Latest, improved quality |
+| Cohere | `cohere.embed-english-v3` | 1,024 | English text |
+| Cohere | `cohere.embed-multilingual-v3` | 1,024 | Multilingual |
+
+### Text Generation Models
+
+| Provider | Model ID | Context | Best For |
+|----------|----------|---------|----------|
+| Anthropic | `anthropic.claude-3-sonnet-20240229-v1:0` | 200K | Balanced performance |
+| Anthropic | `anthropic.claude-3-haiku-20240307-v1:0` | 200K | Fast, cost-effective |
+| Anthropic | `anthropic.claude-3-opus-20240229-v1:0` | 200K | Highest quality |
+| Meta | `meta.llama3-8b-instruct-v1:0` | 8K | Fast, open-source |
+| Meta | `meta.llama3-70b-instruct-v1:0` | 8K | High quality |
+| Amazon | `amazon.titan-text-express-v1` | 8K | Fast, low cost |
+| Mistral | `mistral.mistral-7b-instruct-v0:2` | 32K | Efficient |
+
+## Configuration
+
+### Environment Variables
+
+**Required:**
+```bash
+AWS_REGION=us-east-1
+```
+
+**Optional (at least one model required):**
+```bash
+# For embeddings
+BEDROCK_EMBEDDING_MODEL=amazon.titan-embed-text-v2:0
+
+# For text generation (RAG evaluation)
+BEDROCK_GENERATION_MODEL=anthropic.claude-3-sonnet-20240229-v1:0
+```
+
+**AWS Credentials (choose one method):**
+
+**Method 1: Environment Variables**
+```bash
+AWS_ACCESS_KEY_ID=AKIAIOSFODNN7EXAMPLE
+AWS_SECRET_ACCESS_KEY=wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY
+```
+
+**Method 2: AWS Credentials File** (`~/.aws/credentials`)
+```ini
+[default]
+aws_access_key_id = AKIAIOSFODNN7EXAMPLE
+aws_secret_access_key = wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY
+```
+
+**Method 3: IAM Role** (when running on AWS EC2/ECS/Lambda)
+- No credentials needed, uses instance/task role automatically
+
+### Docker Configuration
+
+Add to your `docker-compose.yml`:
+
+```yaml
+services:
+  mcp:
+    environment:
+      - AWS_REGION=us-east-1
+      - BEDROCK_EMBEDDING_MODEL=amazon.titan-embed-text-v2:0
+      - BEDROCK_GENERATION_MODEL=anthropic.claude-3-sonnet-20240229-v1:0
+      - AWS_ACCESS_KEY_ID=${AWS_ACCESS_KEY_ID}
+      - AWS_SECRET_ACCESS_KEY=${AWS_SECRET_ACCESS_KEY}
+```
+
+Or use AWS credentials file volume mount:
+
+```yaml
+services:
+  mcp:
+    volumes:
+      - ~/.aws:/root/.aws:ro
+    environment:
+      - AWS_REGION=us-east-1
+      - BEDROCK_EMBEDDING_MODEL=amazon.titan-embed-text-v2:0
+```
+
+## Usage Examples
+
+### Embeddings Only
+
+```bash
+export AWS_REGION=us-east-1
+export BEDROCK_EMBEDDING_MODEL=amazon.titan-embed-text-v2:0
+export AWS_ACCESS_KEY_ID=your-key
+export AWS_SECRET_ACCESS_KEY=your-secret
+
+uv run nextcloud-mcp-server
+```
+
+### Both Embeddings and Generation
+
+```bash
+export AWS_REGION=us-east-1
+export BEDROCK_EMBEDDING_MODEL=amazon.titan-embed-text-v2:0
+export BEDROCK_GENERATION_MODEL=anthropic.claude-3-sonnet-20240229-v1:0
+
+# For RAG evaluation with Bedrock
+export RAG_EVAL_PROVIDER=bedrock
+export RAG_EVAL_BEDROCK_MODEL=anthropic.claude-3-sonnet-20240229-v1:0
+
+uv run python -m tests.rag_evaluation.evaluate
+```
+
+### Programmatic Usage
+
+```python
+from nextcloud_mcp_server.providers import BedrockProvider
+
+# Embeddings only
+provider = BedrockProvider(
+    region_name="us-east-1",
+    embedding_model="amazon.titan-embed-text-v2:0",
+)
+
+embeddings = await provider.embed_batch(["text1", "text2"])
+
+# Both capabilities
+provider = BedrockProvider(
+    region_name="us-east-1",
+    embedding_model="amazon.titan-embed-text-v2:0",
+    generation_model="anthropic.claude-3-sonnet-20240229-v1:0",
+)
+
+# Generate embeddings
+embedding = await provider.embed("query text")
+
+# Generate text
+response = await provider.generate("Write a summary", max_tokens=500)
+```
+
+## Cost Considerations
+
+### Embedding Costs (as of Jan 2025)
+
+| Model | Price per 1K tokens |
+|-------|---------------------|
+| Titan Embed Text v2 | $0.0001 |
+| Cohere Embed English v3 | $0.0001 |
+
+### Generation Costs (as of Jan 2025)
+
+| Model | Input (per 1K tokens) | Output (per 1K tokens) |
+|-------|----------------------|------------------------|
+| Claude 3 Haiku | $0.00025 | $0.00125 |
+| Claude 3 Sonnet | $0.003 | $0.015 |
+| Claude 3 Opus | $0.015 | $0.075 |
+| Llama 3 8B | $0.0003 | $0.0006 |
+| Titan Text Express | $0.0002 | $0.0006 |
+
+**Note:** Prices vary by region. Check [AWS Bedrock Pricing](https://aws.amazon.com/bedrock/pricing/) for current rates.
+
+## Troubleshooting
+
+### Error: "Executable doesn't exist" or boto3 not found
+
+**Solution:**
+```bash
+uv sync --group dev  # Installs boto3
+```
+
+### Error: "AccessDeniedException"
+
+**Causes:**
+1. IAM permissions missing
+2. Model access not requested
+3. Wrong AWS region
+
+**Solution:**
+1. Verify IAM policy includes `bedrock:InvokeModel`
+2. Request model access in Bedrock console
+3. Check model is available in your region
+
+### Error: "ResourceNotFoundException"
+
+**Cause:** Invalid model ID or model not available in region
+
+**Solution:**
+- Verify model ID matches exactly (case-sensitive)
+- Check model availability in your AWS region
+- Use `aws bedrock list-foundation-models` to see available models
+
+### Error: "ThrottlingException"
+
+**Cause:** Rate limit exceeded
+
+**Solution:**
+- Reduce request rate
+- Request quota increase via AWS Support
+- Use batch operations where possible
+
+## Security Best Practices
+
+1. **Use IAM Roles** when running on AWS infrastructure
+2. **Rotate Access Keys** regularly if using IAM users
+3. **Restrict Permissions** to only required models
+4. **Enable CloudTrail** for audit logging
+5. **Use AWS Secrets Manager** for credential management
+6. **Monitor Costs** with AWS Cost Explorer and Budgets
+
+## Regional Availability
+
+Amazon Bedrock is available in:
+- **US East (N. Virginia)**: `us-east-1` ✅ Most models
+- **US West (Oregon)**: `us-west-2` ✅ Most models
+- **Asia Pacific (Singapore)**: `ap-southeast-1`
+- **Asia Pacific (Tokyo)**: `ap-northeast-1`
+- **Europe (Frankfurt)**: `eu-central-1`
+
+**Note:** Model availability varies by region. Check the [AWS Bedrock documentation](https://docs.aws.amazon.com/bedrock/latest/userguide/models-regions.html) for current availability.
+
+## References
+
+- [AWS Bedrock Documentation](https://docs.aws.amazon.com/bedrock/)
+- [AWS Bedrock Pricing](https://aws.amazon.com/bedrock/pricing/)
+- [boto3 Bedrock Runtime API](https://boto3.amazonaws.com/v1/documentation/api/latest/reference/services/bedrock-runtime.html)
+- [Provider Architecture ADR](./ADR-015-unified-provider-architecture.md)

nextcloud_mcp_server/embedding/service.py

@@ -1,57 +1,30 @@
-"""Embedding service with provider detection."""
+"""Embedding service with provider detection.
+
+DEPRECATED: This module is maintained for backward compatibility.
+New code should use nextcloud_mcp_server.providers.get_provider() directly.
+"""
 
 import logging
-import os
 
-from .base import EmbeddingProvider
+from nextcloud_mcp_server.providers import get_provider
+
 from .bm25_provider import BM25SparseEmbeddingProvider
-from .ollama_provider import OllamaEmbeddingProvider
-from .simple_provider import SimpleEmbeddingProvider
 
 logger = logging.getLogger(__name__)
 
 
 class EmbeddingService:
-    """Unified embedding service with automatic provider detection."""
+    """
+    Unified embedding service with automatic provider detection.
+
+    DEPRECATED: This class wraps the new unified provider infrastructure
+    for backward compatibility. New code should use
+    nextcloud_mcp_server.providers.get_provider() directly.
+    """
 
     def __init__(self):
         """Initialize embedding service with auto-detected provider."""
-        self.provider = self._detect_provider()
-
-    def _detect_provider(self) -> EmbeddingProvider:
-        """
-        Auto-detect available embedding provider.
-
-        Checks environment variables in order:
-        1. OLLAMA_BASE_URL - Use Ollama provider (production)
-        2. OPENAI_API_KEY - Use OpenAI provider (future)
-        3. Fallback to SimpleEmbeddingProvider (testing/development)
-
-        Returns:
-            Configured embedding provider
-        """
-        # Ollama provider (production)
-        ollama_url = os.getenv("OLLAMA_BASE_URL")
-        if ollama_url:
-            logger.info(f"Using Ollama embedding provider: {ollama_url}")
-            return OllamaEmbeddingProvider(
-                base_url=ollama_url,
-                model=os.getenv("OLLAMA_EMBEDDING_MODEL", "nomic-embed-text"),
-                verify_ssl=os.getenv("OLLAMA_VERIFY_SSL", "true").lower() == "true",
-            )
-
-        # OpenAI provider (future implementation)
-        # openai_key = os.getenv("OPENAI_API_KEY")
-        # if openai_key:
-        #     return OpenAIEmbeddingProvider(api_key=openai_key)
-
-        # Fallback to simple provider for development/testing
-        logger.warning(
-            "No embedding provider configured (OLLAMA_BASE_URL or OPENAI_API_KEY not set). "
-            "Using SimpleEmbeddingProvider for testing/development. "
-            "For production, configure an external embedding service."
-        )
-        return SimpleEmbeddingProvider(dimension=384)
+        self.provider = get_provider()
 
     async def embed(self, text: str) -> list[float]:
         """

nextcloud_mcp_server/providers/__init__.py

@@ -0,0 +1,18 @@
+"""Unified provider infrastructure for embeddings and text generation."""
+
+from .anthropic import AnthropicProvider
+from .base import Provider
+from .bedrock import BedrockProvider
+from .ollama import OllamaProvider
+from .registry import get_provider, reset_provider
+from .simple import SimpleProvider
+
+__all__ = [
+    "Provider",
+    "OllamaProvider",
+    "AnthropicProvider",
+    "SimpleProvider",
+    "BedrockProvider",
+    "get_provider",
+    "reset_provider",
+]

nextcloud_mcp_server/providers/anthropic.py

@@ -0,0 +1,97 @@
+"""Unified Anthropic provider for text generation."""
+
+import logging
+
+from anthropic import AsyncAnthropic
+
+from .base import Provider
+
+logger = logging.getLogger(__name__)
+
+
+class AnthropicProvider(Provider):
+    """
+    Anthropic provider for text generation.
+
+    Supports Claude models via the Anthropic API.
+    Note: Anthropic doesn't provide embedding models, only text generation.
+    """
+
+    def __init__(self, api_key: str, model: str = "claude-3-5-sonnet-20241022"):
+        """
+        Initialize Anthropic provider.
+
+        Args:
+            api_key: Anthropic API key
+            model: Model name (e.g., "claude-3-5-sonnet-20241022")
+        """
+        self.client = AsyncAnthropic(api_key=api_key)
+        self.model = model
+
+        logger.info(f"Initialized Anthropic provider (model={model})")
+
+    @property
+    def supports_embeddings(self) -> bool:
+        """Whether this provider supports embedding generation."""
+        return False
+
+    @property
+    def supports_generation(self) -> bool:
+        """Whether this provider supports text generation."""
+        return True
+
+    async def embed(self, text: str) -> list[float]:
+        """
+        Generate embedding vector for text.
+
+        Raises:
+            NotImplementedError: Anthropic doesn't provide embedding models
+        """
+        raise NotImplementedError(
+            "Embedding not supported by Anthropic - use Ollama or Bedrock for embeddings"
+        )
+
+    async def embed_batch(self, texts: list[str]) -> list[list[float]]:
+        """
+        Generate embeddings for multiple texts.
+
+        Raises:
+            NotImplementedError: Anthropic doesn't provide embedding models
+        """
+        raise NotImplementedError(
+            "Embedding not supported by Anthropic - use Ollama or Bedrock for embeddings"
+        )
+
+    def get_dimension(self) -> int:
+        """
+        Get embedding dimension.
+
+        Raises:
+            NotImplementedError: Anthropic doesn't provide embedding models
+        """
+        raise NotImplementedError(
+            "Embedding not supported by Anthropic - use Ollama or Bedrock for embeddings"
+        )
+
+    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
+        """
+        Generate text using Anthropic API.
+
+        Args:
+            prompt: The prompt to generate from
+            max_tokens: Maximum tokens to generate
+
+        Returns:
+            Generated text
+        """
+        message = await self.client.messages.create(
+            model=self.model,
+            max_tokens=max_tokens,
+            temperature=0.7,
+            messages=[{"role": "user", "content": prompt}],
+        )
+        return message.content[0].text
+
+    async def close(self) -> None:
+        """Close the client (no-op for Anthropic SDK)."""
+        pass

nextcloud_mcp_server/providers/base.py

@@ -0,0 +1,91 @@
+"""Unified provider interface for embeddings and text generation."""
+
+from abc import ABC, abstractmethod
+
+
+class Provider(ABC):
+    """
+    Unified base class for LLM providers.
+
+    Providers can support embeddings, text generation, or both.
+    Use capability properties to determine what features are available.
+    """
+
+    @property
+    @abstractmethod
+    def supports_embeddings(self) -> bool:
+        """Whether this provider supports embedding generation."""
+        pass
+
+    @property
+    @abstractmethod
+    def supports_generation(self) -> bool:
+        """Whether this provider supports text generation."""
+        pass
+
+    @abstractmethod
+    async def embed(self, text: str) -> list[float]:
+        """
+        Generate embedding vector for text.
+
+        Args:
+            text: Input text to embed
+
+        Returns:
+            Vector embedding as list of floats
+
+        Raises:
+            NotImplementedError: If provider doesn't support embeddings
+        """
+        pass
+
+    @abstractmethod
+    async def embed_batch(self, texts: list[str]) -> list[list[float]]:
+        """
+        Generate embeddings for multiple texts (optimized).
+
+        Args:
+            texts: List of texts to embed
+
+        Returns:
+            List of vector embeddings
+
+        Raises:
+            NotImplementedError: If provider doesn't support embeddings
+        """
+        pass
+
+    @abstractmethod
+    def get_dimension(self) -> int:
+        """
+        Get embedding dimension for this provider.
+
+        Returns:
+            Vector dimension (e.g., 768 for nomic-embed-text)
+
+        Raises:
+            NotImplementedError: If provider doesn't support embeddings
+        """
+        pass
+
+    @abstractmethod
+    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
+        """
+        Generate text from a prompt.
+
+        Args:
+            prompt: The prompt to generate from
+            max_tokens: Maximum tokens to generate
+
+        Returns:
+            Generated text
+
+        Raises:
+            NotImplementedError: If provider doesn't support generation
+        """
+        pass
+
+    @abstractmethod
+    async def close(self) -> None:
+        """Close the provider and release resources."""
+        pass

nextcloud_mcp_server/providers/bedrock.py

@@ -0,0 +1,397 @@
+"""Amazon Bedrock provider for embeddings and text generation."""
+
+import json
+import logging
+from typing import Any
+
+try:
+    import boto3
+    from botocore.exceptions import BotoCoreError, ClientError
+
+    BOTO3_AVAILABLE = True
+except ImportError:
+    BOTO3_AVAILABLE = False
+
+from .base import Provider
+
+logger = logging.getLogger(__name__)
+
+
+class BedrockProvider(Provider):
+    """
+    Amazon Bedrock provider supporting both embeddings and text generation.
+
+    Uses AWS Bedrock Runtime API with boto3. Supports various model families:
+    - Embeddings: amazon.titan-embed-text-v1, amazon.titan-embed-text-v2, cohere.embed-*
+    - Text Generation: anthropic.claude-*, meta.llama3-*, amazon.titan-text-*, mistral.*, etc.
+
+    Requires AWS credentials configured via:
+    - Environment variables (AWS_ACCESS_KEY_ID, AWS_SECRET_ACCESS_KEY, AWS_REGION)
+    - AWS credentials file (~/.aws/credentials)
+    - IAM role (when running on AWS)
+    """
+
+    def __init__(
+        self,
+        region_name: str | None = None,
+        embedding_model: str | None = None,
+        generation_model: str | None = None,
+        aws_access_key_id: str | None = None,
+        aws_secret_access_key: str | None = None,
+    ):
+        """
+        Initialize Bedrock provider.
+
+        Args:
+            region_name: AWS region (e.g., "us-east-1"). Defaults to AWS_REGION env var.
+            embedding_model: Model ID for embeddings (e.g., "amazon.titan-embed-text-v2:0").
+                None disables embeddings.
+            generation_model: Model ID for text generation (e.g., "anthropic.claude-3-sonnet-20240229-v1:0").
+                None disables generation.
+            aws_access_key_id: AWS access key (optional, uses default credential chain if not provided)
+            aws_secret_access_key: AWS secret key (optional, uses default credential chain if not provided)
+
+        Raises:
+            ImportError: If boto3 is not installed
+        """
+        if not BOTO3_AVAILABLE:
+            raise ImportError(
+                "boto3 is required for Bedrock provider. Install with: pip install boto3"
+            )
+
+        self.embedding_model = embedding_model
+        self.generation_model = generation_model
+        self._dimension: int | None = None  # Detected dynamically
+
+        # Initialize bedrock-runtime client
+        client_kwargs: dict[str, Any] = {}
+        if region_name:
+            client_kwargs["region_name"] = region_name
+        if aws_access_key_id:
+            client_kwargs["aws_access_key_id"] = aws_access_key_id
+        if aws_secret_access_key:
+            client_kwargs["aws_secret_access_key"] = aws_secret_access_key
+
+        self.client = boto3.client("bedrock-runtime", **client_kwargs)
+
+        logger.info(
+            f"Initialized Bedrock provider in region {region_name or 'default'} "
+            f"(embedding_model={embedding_model}, generation_model={generation_model})"
+        )
+
+    @property
+    def supports_embeddings(self) -> bool:
+        """Whether this provider supports embedding generation."""
+        return self.embedding_model is not None
+
+    @property
+    def supports_generation(self) -> bool:
+        """Whether this provider supports text generation."""
+        return self.generation_model is not None
+
+    def _create_embedding_request(self, text: str) -> dict[str, Any]:
+        """
+        Create model-specific embedding request payload.
+
+        Args:
+            text: Input text to embed
+
+        Returns:
+            Request payload dict for the embedding model
+        """
+        if not self.embedding_model:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        # Titan Embed models
+        if self.embedding_model.startswith("amazon.titan-embed"):
+            return {"inputText": text}
+
+        # Cohere Embed models
+        elif self.embedding_model.startswith("cohere.embed"):
+            return {"texts": [text], "input_type": "search_document"}
+
+        # Unknown model - try Titan format as default
+        else:
+            logger.warning(
+                f"Unknown embedding model format for {self.embedding_model}, "
+                "using Titan format as default"
+            )
+            return {"inputText": text}
+
+    def _parse_embedding_response(self, response: dict[str, Any]) -> list[float]:
+        """
+        Parse model-specific embedding response.
+
+        Args:
+            response: Raw response from Bedrock
+
+        Returns:
+            Embedding vector as list of floats
+        """
+        # Titan Embed models
+        if self.embedding_model and self.embedding_model.startswith(
+            "amazon.titan-embed"
+        ):
+            return response["embedding"]
+
+        # Cohere Embed models
+        elif self.embedding_model and self.embedding_model.startswith("cohere.embed"):
+            return response["embeddings"][0]
+
+        # Unknown model - try Titan format as default
+        else:
+            logger.warning(
+                f"Unknown embedding response format for {self.embedding_model}, "
+                "trying Titan format"
+            )
+            return response.get("embedding", response.get("embeddings", [None])[0])
+
+    async def embed(self, text: str) -> list[float]:
+        """
+        Generate embedding vector for text.
+
+        Args:
+            text: Input text to embed
+
+        Returns:
+            Vector embedding as list of floats
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+            ClientError: If Bedrock API call fails
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        try:
+            request_body = self._create_embedding_request(text)
+
+            response = self.client.invoke_model(
+                modelId=self.embedding_model,
+                body=json.dumps(request_body),
+                accept="application/json",
+                contentType="application/json",
+            )
+
+            response_body = json.loads(response["body"].read())
+            embedding = self._parse_embedding_response(response_body)
+
+            return embedding
+
+        except (BotoCoreError, ClientError) as e:
+            logger.error(f"Bedrock embedding error: {e}")
+            raise
+
+    async def embed_batch(self, texts: list[str]) -> list[list[float]]:
+        """
+        Generate embeddings for multiple texts.
+
+        Note: Current implementation sends requests sequentially.
+        Future optimization could use asyncio for concurrent requests.
+
+        Args:
+            texts: List of texts to embed
+
+        Returns:
+            List of vector embeddings
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+            ClientError: If Bedrock API call fails
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        embeddings = []
+        for text in texts:
+            embedding = await self.embed(text)
+            embeddings.append(embedding)
+        return embeddings
+
+    async def _detect_dimension(self):
+        """
+        Detect embedding dimension by generating a test embedding.
+        """
+        if self._dimension is None and self.supports_embeddings:
+            logger.debug(
+                f"Detecting embedding dimension for model {self.embedding_model}..."
+            )
+            test_embedding = await self.embed("test")
+            self._dimension = len(test_embedding)
+            logger.info(
+                f"Detected embedding dimension: {self._dimension} "
+                f"for model {self.embedding_model}"
+            )
+
+    def get_dimension(self) -> int:
+        """
+        Get embedding dimension.
+
+        Returns:
+            Vector dimension for the configured embedding model
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+            RuntimeError: If dimension not detected yet (call _detect_dimension first)
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        if self._dimension is None:
+            raise RuntimeError(
+                f"Embedding dimension not detected yet for model {self.embedding_model}. "
+                "Call _detect_dimension() first or generate an embedding."
+            )
+        return self._dimension
+
+    def _create_generation_request(
+        self, prompt: str, max_tokens: int
+    ) -> dict[str, Any]:
+        """
+        Create model-specific text generation request payload.
+
+        Args:
+            prompt: The prompt to generate from
+            max_tokens: Maximum tokens to generate
+
+        Returns:
+            Request payload dict for the generation model
+        """
+        if not self.generation_model:
+            raise NotImplementedError(
+                "Text generation not supported - no generation_model configured"
+            )
+
+        # Anthropic Claude models
+        if self.generation_model.startswith("anthropic.claude"):
+            return {
+                "anthropic_version": "bedrock-2023-05-31",
+                "max_tokens": max_tokens,
+                "temperature": 0.7,
+                "messages": [{"role": "user", "content": prompt}],
+            }
+
+        # Meta Llama models
+        elif self.generation_model.startswith("meta.llama"):
+            return {"prompt": prompt, "max_gen_len": max_tokens, "temperature": 0.7}
+
+        # Amazon Titan Text models
+        elif self.generation_model.startswith("amazon.titan-text"):
+            return {
+                "inputText": prompt,
+                "textGenerationConfig": {
+                    "maxTokenCount": max_tokens,
+                    "temperature": 0.7,
+                },
+            }
+
+        # Mistral models
+        elif self.generation_model.startswith("mistral"):
+            return {"prompt": prompt, "max_tokens": max_tokens, "temperature": 0.7}
+
+        # Unknown model - try Claude format as default
+        else:
+            logger.warning(
+                f"Unknown generation model format for {self.generation_model}, "
+                "using Claude format as default"
+            )
+            return {
+                "anthropic_version": "bedrock-2023-05-31",
+                "max_tokens": max_tokens,
+                "temperature": 0.7,
+                "messages": [{"role": "user", "content": prompt}],
+            }
+
+    def _parse_generation_response(self, response: dict[str, Any]) -> str:
+        """
+        Parse model-specific text generation response.
+
+        Args:
+            response: Raw response from Bedrock
+
+        Returns:
+            Generated text
+        """
+        # Anthropic Claude models
+        if self.generation_model and self.generation_model.startswith(
+            "anthropic.claude"
+        ):
+            return response["content"][0]["text"]
+
+        # Meta Llama models
+        elif self.generation_model and self.generation_model.startswith("meta.llama"):
+            return response["generation"]
+
+        # Amazon Titan Text models
+        elif self.generation_model and self.generation_model.startswith(
+            "amazon.titan-text"
+        ):
+            return response["results"][0]["outputText"]
+
+        # Mistral models
+        elif self.generation_model and self.generation_model.startswith("mistral"):
+            return response["outputs"][0]["text"]
+
+        # Unknown model - try common response fields
+        else:
+            logger.warning(
+                f"Unknown generation response format for {self.generation_model}, "
+                "trying common fields"
+            )
+            # Try common response field names
+            for field in ["text", "generation", "outputText", "completion"]:
+                if field in response:
+                    return response[field]
+            # Last resort: return JSON string
+            return json.dumps(response)
+
+    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
+        """
+        Generate text from a prompt.
+
+        Args:
+            prompt: The prompt to generate from
+            max_tokens: Maximum tokens to generate
+
+        Returns:
+            Generated text
+
+        Raises:
+            NotImplementedError: If generation not enabled (no generation_model)
+            ClientError: If Bedrock API call fails
+        """
+        if not self.supports_generation:
+            raise NotImplementedError(
+                "Text generation not supported - no generation_model configured"
+            )
+
+        try:
+            request_body = self._create_generation_request(prompt, max_tokens)
+
+            response = self.client.invoke_model(
+                modelId=self.generation_model,
+                body=json.dumps(request_body),
+                accept="application/json",
+                contentType="application/json",
+            )
+
+            response_body = json.loads(response["body"].read())
+            text = self._parse_generation_response(response_body)
+
+            return text
+
+        except (BotoCoreError, ClientError) as e:
+            logger.error(f"Bedrock generation error: {e}")
+            raise
+
+    async def close(self) -> None:
+        """Close the client (no-op for boto3 clients)."""
+        pass

nextcloud_mcp_server/providers/ollama.py

@@ -0,0 +1,221 @@
+"""Unified Ollama provider for embeddings and text generation."""
+
+import logging
+
+import httpx
+
+from .base import Provider
+
+logger = logging.getLogger(__name__)
+
+
+class OllamaProvider(Provider):
+    """
+    Ollama provider supporting both embeddings and text generation.
+
+    Supports TLS, SSL verification, and automatic model loading.
+    """
+
+    def __init__(
+        self,
+        base_url: str,
+        embedding_model: str | None = None,
+        generation_model: str | None = None,
+        verify_ssl: bool = True,
+        timeout: httpx.Timeout | None = None,
+    ):
+        """
+        Initialize Ollama provider.
+
+        Args:
+            base_url: Ollama API base URL (e.g., https://ollama.internal.example.com:443)
+            embedding_model: Model for embeddings (e.g., "nomic-embed-text"). None disables embeddings.
+            generation_model: Model for text generation (e.g., "llama3.2:1b"). None disables generation.
+            verify_ssl: Verify SSL certificates (default: True)
+            timeout: HTTP timeout configuration
+        """
+        self.base_url = base_url.rstrip("/")
+        self.embedding_model = embedding_model
+        self.generation_model = generation_model
+        self.verify_ssl = verify_ssl
+
+        if timeout is None:
+            timeout = httpx.Timeout(timeout=120, connect=5)
+
+        self.client = httpx.AsyncClient(verify=verify_ssl, timeout=timeout)
+        self._dimension: int | None = None  # Detected dynamically for embeddings
+
+        logger.info(
+            f"Initialized Ollama provider: {base_url} "
+            f"(embedding_model={embedding_model}, generation_model={generation_model}, "
+            f"verify_ssl={verify_ssl})"
+        )
+
+        # Pre-check and auto-load models
+        if embedding_model:
+            self._check_model_is_loaded(embedding_model, autoload=True)
+        if generation_model:
+            self._check_model_is_loaded(generation_model, autoload=True)
+
+    @property
+    def supports_embeddings(self) -> bool:
+        """Whether this provider supports embedding generation."""
+        return self.embedding_model is not None
+
+    @property
+    def supports_generation(self) -> bool:
+        """Whether this provider supports text generation."""
+        return self.generation_model is not None
+
+    async def embed(self, text: str) -> list[float]:
+        """
+        Generate embedding vector for text.
+
+        Args:
+            text: Input text to embed
+
+        Returns:
+            Vector embedding as list of floats
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        response = await self.client.post(
+            f"{self.base_url}/api/embeddings",
+            json={"model": self.embedding_model, "prompt": text},
+        )
+        response.raise_for_status()
+        return response.json()["embedding"]
+
+    async def embed_batch(self, texts: list[str]) -> list[list[float]]:
+        """
+        Generate embeddings for multiple texts (batched requests).
+
+        Note: Ollama doesn't have native batch API, so we send requests sequentially.
+
+        Args:
+            texts: List of texts to embed
+
+        Returns:
+            List of vector embeddings
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        embeddings = []
+        for text in texts:
+            embedding = await self.embed(text)
+            embeddings.append(embedding)
+        return embeddings
+
+    async def _detect_dimension(self):
+        """
+        Detect embedding dimension by generating a test embedding.
+
+        This method queries the model to determine the actual dimension
+        instead of relying on hardcoded values.
+        """
+        if self._dimension is None and self.supports_embeddings:
+            logger.debug(
+                f"Detecting embedding dimension for model {self.embedding_model}..."
+            )
+            test_embedding = await self.embed("test")
+            self._dimension = len(test_embedding)
+            logger.info(
+                f"Detected embedding dimension: {self._dimension} "
+                f"for model {self.embedding_model}"
+            )
+
+    def get_dimension(self) -> int:
+        """
+        Get embedding dimension.
+
+        Returns:
+            Vector dimension for the configured embedding model
+
+        Raises:
+            NotImplementedError: If embeddings not enabled (no embedding_model)
+            RuntimeError: If dimension not detected yet (call _detect_dimension first)
+        """
+        if not self.supports_embeddings:
+            raise NotImplementedError(
+                "Embedding not supported - no embedding_model configured"
+            )
+
+        if self._dimension is None:
+            raise RuntimeError(
+                f"Embedding dimension not detected yet for model {self.embedding_model}. "
+                "Call _detect_dimension() first or generate an embedding."
+            )
+        return self._dimension
+
+    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
+        """
+        Generate text from a prompt.
+
+        Args:
+            prompt: The prompt to generate from
+            max_tokens: Maximum tokens to generate
+
+        Returns:
+            Generated text
+
+        Raises:
+            NotImplementedError: If generation not enabled (no generation_model)
+        """
+        if not self.supports_generation:
+            raise NotImplementedError(
+                "Text generation not supported - no generation_model configured"
+            )
+
+        response = await self.client.post(
+            f"{self.base_url}/api/generate",
+            json={
+                "model": self.generation_model,
+                "prompt": prompt,
+                "stream": False,
+                "options": {
+                    "num_predict": max_tokens,
+                    "temperature": 0.7,
+                },
+            },
+        )
+        response.raise_for_status()
+        data = response.json()
+        return data["response"]
+
+    def _check_model_is_loaded(self, model: str, autoload: bool = True):
+        """
+        Check if model is loaded in Ollama, optionally auto-loading it.
+
+        Args:
+            model: Model name to check
+            autoload: Whether to automatically pull the model if not loaded
+        """
+        response = httpx.get(f"{self.base_url}/api/tags")
+        response.raise_for_status()
+
+        models = [m["name"] for m in response.json().get("models", [])]
+        logger.info("Ollama has following models pre-loaded: %s", models)
+
+        if (model not in models) and autoload:
+            logger.warning(
+                "Model '%s' not yet available in ollama, attempting to pull now...",
+                model,
+            )
+            response = httpx.post(f"{self.base_url}/api/pull", json={"model": model})
+            response.raise_for_status()
+
+    async def close(self) -> None:
+        """Close HTTP client."""
+        await self.client.aclose()

nextcloud_mcp_server/providers/registry.py

@@ -0,0 +1,126 @@
+"""Provider registry and factory for auto-detection and instantiation."""
+
+import logging
+import os
+
+from .base import Provider
+from .bedrock import BedrockProvider
+from .ollama import OllamaProvider
+from .simple import SimpleProvider
+
+logger = logging.getLogger(__name__)
+
+
+class ProviderRegistry:
+    """
+    Registry for provider auto-detection and instantiation.
+
+    Checks environment variables in priority order and creates appropriate provider:
+    1. Bedrock (AWS_REGION + BEDROCK_*_MODEL)
+    2. Ollama (OLLAMA_BASE_URL)
+    3. Simple (fallback for testing/development)
+    """
+
+    @staticmethod
+    def create_provider() -> Provider:
+        """
+        Auto-detect and create provider based on environment variables.
+
+        Priority order:
+        1. Bedrock - if AWS_REGION or BEDROCK_EMBEDDING_MODEL is set
+        2. Ollama - if OLLAMA_BASE_URL is set
+        3. Simple - fallback for testing/development
+
+        Returns:
+            Provider instance
+
+        Environment Variables:
+            Bedrock:
+                - AWS_REGION: AWS region (e.g., "us-east-1")
+                - AWS_ACCESS_KEY_ID: AWS access key (optional, uses credential chain)
+                - AWS_SECRET_ACCESS_KEY: AWS secret key (optional)
+                - BEDROCK_EMBEDDING_MODEL: Model ID for embeddings (e.g., "amazon.titan-embed-text-v2:0")
+                - BEDROCK_GENERATION_MODEL: Model ID for text generation (e.g., "anthropic.claude-3-sonnet-20240229-v1:0")
+
+            Ollama:
+                - OLLAMA_BASE_URL: Ollama API base URL (e.g., "http://localhost:11434")
+                - OLLAMA_EMBEDDING_MODEL: Model for embeddings (default: "nomic-embed-text")
+                - OLLAMA_GENERATION_MODEL: Model for text generation (e.g., "llama3.2:1b")
+                - OLLAMA_VERIFY_SSL: Verify SSL certificates (default: "true")
+
+            Simple (no configuration needed, fallback):
+                - SIMPLE_EMBEDDING_DIMENSION: Embedding dimension (default: 384)
+        """
+        # 1. Check for Bedrock
+        aws_region = os.getenv("AWS_REGION")
+        bedrock_embedding_model = os.getenv("BEDROCK_EMBEDDING_MODEL")
+        bedrock_generation_model = os.getenv("BEDROCK_GENERATION_MODEL")
+
+        if aws_region or bedrock_embedding_model or bedrock_generation_model:
+            logger.info(
+                f"Using Bedrock provider: region={aws_region}, "
+                f"embedding_model={bedrock_embedding_model}, "
+                f"generation_model={bedrock_generation_model}"
+            )
+            return BedrockProvider(
+                region_name=aws_region,
+                embedding_model=bedrock_embedding_model,
+                generation_model=bedrock_generation_model,
+                aws_access_key_id=os.getenv("AWS_ACCESS_KEY_ID"),
+                aws_secret_access_key=os.getenv("AWS_SECRET_ACCESS_KEY"),
+            )
+
+        # 2. Check for Ollama
+        ollama_url = os.getenv("OLLAMA_BASE_URL")
+        if ollama_url:
+            embedding_model = os.getenv("OLLAMA_EMBEDDING_MODEL", "nomic-embed-text")
+            generation_model = os.getenv("OLLAMA_GENERATION_MODEL")
+            verify_ssl = os.getenv("OLLAMA_VERIFY_SSL", "true").lower() == "true"
+
+            logger.info(
+                f"Using Ollama provider: {ollama_url}, "
+                f"embedding_model={embedding_model}, "
+                f"generation_model={generation_model}"
+            )
+            return OllamaProvider(
+                base_url=ollama_url,
+                embedding_model=embedding_model,
+                generation_model=generation_model,
+                verify_ssl=verify_ssl,
+            )
+
+        # 3. Fallback to Simple provider for development/testing
+        dimension = int(os.getenv("SIMPLE_EMBEDDING_DIMENSION", "384"))
+        logger.warning(
+            "No provider configured (AWS_REGION, OLLAMA_BASE_URL not set). "
+            "Using SimpleProvider for testing/development. "
+            "For production, configure Bedrock or Ollama."
+        )
+        return SimpleProvider(dimension=dimension)
+
+
+# Singleton instance
+_provider: Provider | None = None
+
+
+def get_provider() -> Provider:
+    """
+    Get singleton provider instance.
+
+    Returns:
+        Global Provider instance (auto-detected on first call)
+    """
+    global _provider
+    if _provider is None:
+        _provider = ProviderRegistry.create_provider()
+    return _provider
+
+
+def reset_provider():
+    """
+    Reset singleton provider instance.
+
+    Useful for testing or reconfiguration.
+    """
+    global _provider
+    _provider = None

nextcloud_mcp_server/providers/simple.py

@@ -0,0 +1,149 @@
+"""Simple in-process embedding provider for testing.
+
+This provider uses a basic TF-IDF-like approach with feature hashing to generate
+deterministic embeddings without requiring external services. Suitable for testing
+but not for production use.
+"""
+
+import hashlib
+import math
+import re
+from collections import Counter
+
+from .base import Provider
+
+
+class SimpleProvider(Provider):
+    """Simple deterministic embedding provider using feature hashing.
+
+    This implementation:
+    - Tokenizes text into words
+    - Uses feature hashing to map words to fixed-size vectors
+    - Applies TF-IDF-like weighting
+    - Normalizes vectors to unit length
+
+    Not suitable for production but good for testing semantic search infrastructure.
+    Only supports embeddings, not text generation.
+    """
+
+    def __init__(self, dimension: int = 384):
+        """Initialize simple embedding provider.
+
+        Args:
+            dimension: Embedding dimension (default: 384)
+        """
+        self.dimension = dimension
+
+    @property
+    def supports_embeddings(self) -> bool:
+        """Whether this provider supports embedding generation."""
+        return True
+
+    @property
+    def supports_generation(self) -> bool:
+        """Whether this provider supports text generation."""
+        return False
+
+    def _tokenize(self, text: str) -> list[str]:
+        """Tokenize text into lowercase words.
+
+        Args:
+            text: Input text
+
+        Returns:
+            List of lowercase word tokens
+        """
+        # Simple word tokenization
+        text = text.lower()
+        words = re.findall(r"\b\w+\b", text)
+        return words
+
+    def _hash_word(self, word: str) -> int:
+        """Hash word to dimension index.
+
+        Args:
+            word: Word to hash
+
+        Returns:
+            Index in range [0, dimension)
+        """
+        hash_bytes = hashlib.md5(word.encode()).digest()
+        hash_int = int.from_bytes(hash_bytes[:4], byteorder="big")
+        return hash_int % self.dimension
+
+    def _embed_single(self, text: str) -> list[float]:
+        """Generate embedding for single text.
+
+        Args:
+            text: Input text
+
+        Returns:
+            Normalized embedding vector
+        """
+        tokens = self._tokenize(text)
+        if not tokens:
+            return [0.0] * self.dimension
+
+        # Count term frequencies
+        term_freq = Counter(tokens)
+
+        # Initialize vector
+        vector = [0.0] * self.dimension
+
+        # Apply TF weighting with feature hashing
+        for word, count in term_freq.items():
+            idx = self._hash_word(word)
+            # Simple TF weighting: log(1 + count)
+            vector[idx] += math.log1p(count)
+
+        # Normalize to unit length
+        norm = math.sqrt(sum(x * x for x in vector))
+        if norm > 0:
+            vector = [x / norm for x in vector]
+
+        return vector
+
+    async def embed(self, text: str) -> list[float]:
+        """Generate embedding vector for text.
+
+        Args:
+            text: Input text to embed
+
+        Returns:
+            Vector embedding as list of floats
+        """
+        return self._embed_single(text)
+
+    async def embed_batch(self, texts: list[str]) -> list[list[float]]:
+        """Generate embeddings for multiple texts.
+
+        Args:
+            texts: List of texts to embed
+
+        Returns:
+            List of vector embeddings
+        """
+        return [self._embed_single(text) for text in texts]
+
+    def get_dimension(self) -> int:
+        """Get embedding dimension.
+
+        Returns:
+            Vector dimension
+        """
+        return self.dimension
+
+    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
+        """
+        Generate text from a prompt.
+
+        Raises:
+            NotImplementedError: Simple provider doesn't support text generation
+        """
+        raise NotImplementedError(
+            "Text generation not supported by Simple provider - use Ollama, Anthropic, or Bedrock"
+        )
+
+    async def close(self) -> None:
+        """Close the provider (no-op for simple provider)."""
+        pass

pyproject.toml

@@ -104,6 +104,7 @@ module-root = ""
 [dependency-groups]
 dev = [
     "anthropic>=0.42.0",  # For RAG evaluation with Anthropic LLMs
+    "boto3>=1.35.0",  # For Amazon Bedrock provider (optional)
     "commitizen>=4.8.2",
     "datasets>=3.3.0",  # For BeIR nfcorpus dataset loading
     "ipython>=9.2.0",

tests/rag_evaluation/llm_providers.py

@@ -1,99 +1,20 @@
 """LLM provider abstraction for RAG evaluation.
 
-Supports Ollama (local) and Anthropic (cloud) providers for both ground truth
+DEPRECATED: This module is maintained for backward compatibility with RAG evaluation tests.
+New code should use nextcloud_mcp_server.providers directly.
+
+Supports Ollama (local), Anthropic (cloud), and Bedrock (AWS) providers for both ground truth
 generation and evaluation.
 """
 
 import os
-from typing import Protocol
 
-import httpx
-from anthropic import AsyncAnthropic
-
-
-class LLMProvider(Protocol):
-    """Protocol for LLM providers."""
-
-    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
-        """Generate text from a prompt.
-
-        Args:
-            prompt: The prompt to generate from
-            max_tokens: Maximum tokens to generate
-
-        Returns:
-            Generated text
-        """
-        ...
-
-    async def close(self) -> None:
-        """Close the provider and release resources."""
-        ...
-
-
-class OllamaProvider:
-    """Ollama provider for local LLM inference."""
-
-    def __init__(self, base_url: str, model: str):
-        """Initialize Ollama provider.
-
-        Args:
-            base_url: Ollama API base URL (e.g., http://localhost:11434)
-            model: Model name (e.g., llama3.1:8b)
-        """
-        self.base_url = base_url.rstrip("/")
-        self.model = model
-        self.client = httpx.AsyncClient(timeout=600.0)  # 10 min timeout for generation
-
-    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
-        """Generate text using Ollama API."""
-        response = await self.client.post(
-            f"{self.base_url}/api/generate",
-            json={
-                "model": self.model,
-                "prompt": prompt,
-                "stream": False,
-                "options": {
-                    "num_predict": max_tokens,
-                    "temperature": 0.7,
-                },
-            },
+from nextcloud_mcp_server.providers import (
+    AnthropicProvider,
+    BedrockProvider,
+    OllamaProvider,
+    Provider,
 )
-        response.raise_for_status()
-        data = response.json()
-        return data["response"]
-
-    async def close(self):
-        """Close the HTTP client."""
-        await self.client.aclose()
-
-
-class AnthropicProvider:
-    """Anthropic provider for cloud LLM inference."""
-
-    def __init__(self, api_key: str, model: str):
-        """Initialize Anthropic provider.
-
-        Args:
-            api_key: Anthropic API key
-            model: Model name (e.g., claude-3-5-sonnet-20241022)
-        """
-        self.client = AsyncAnthropic(api_key=api_key)
-        self.model = model
-
-    async def generate(self, prompt: str, max_tokens: int = 500) -> str:
-        """Generate text using Anthropic API."""
-        message = await self.client.messages.create(
-            model=self.model,
-            max_tokens=max_tokens,
-            temperature=0.7,
-            messages=[{"role": "user", "content": prompt}],
-        )
-        return message.content[0].text
-
-    async def close(self):
-        """Close the client (no-op for Anthropic)."""
-        pass
 
 
 def create_llm_provider(
@@ -102,18 +23,24 @@ def create_llm_provider(
     ollama_model: str | None = None,
     anthropic_api_key: str | None = None,
     anthropic_model: str | None = None,
-) -> LLMProvider:
+    bedrock_region: str | None = None,
+    bedrock_model: str | None = None,
+) -> Provider:
     """Create an LLM provider from environment variables or arguments.
 
     Args:
-        provider: Provider type ('ollama' or 'anthropic'). Defaults to RAG_EVAL_PROVIDER env var or 'ollama'
+        provider: Provider type ('ollama', 'anthropic', or 'bedrock').
+            Defaults to RAG_EVAL_PROVIDER env var or 'ollama'
         ollama_base_url: Ollama base URL. Defaults to RAG_EVAL_OLLAMA_BASE_URL or 'http://localhost:11434'
-        ollama_model: Ollama model. Defaults to RAG_EVAL_OLLAMA_MODEL or 'llama3.1:8b'
+        ollama_model: Ollama model. Defaults to RAG_EVAL_OLLAMA_MODEL or 'llama3.2:1b'
         anthropic_api_key: Anthropic API key. Defaults to RAG_EVAL_ANTHROPIC_API_KEY env var
         anthropic_model: Anthropic model. Defaults to RAG_EVAL_ANTHROPIC_MODEL or 'claude-3-5-sonnet-20241022'
+        bedrock_region: AWS region. Defaults to RAG_EVAL_BEDROCK_REGION or AWS_REGION env var
+        bedrock_model: Bedrock model ID. Defaults to RAG_EVAL_BEDROCK_MODEL or
+            'anthropic.claude-3-sonnet-20240229-v1:0'
 
     Returns:
-        LLMProvider instance
+        Provider instance
 
     Raises:
         ValueError: If provider is invalid or required credentials are missing
@@ -130,7 +57,9 @@ def create_llm_provider(
             or "http://localhost:11434"
         )
         model = ollama_model or os.environ.get("RAG_EVAL_OLLAMA_MODEL", "llama3.2:1b")
-        return OllamaProvider(base_url=base_url, model=model)
+        return OllamaProvider(
+            base_url=base_url, embedding_model=None, generation_model=model
+        )
 
     elif provider == "anthropic":
         api_key = anthropic_api_key or os.environ.get("RAG_EVAL_ANTHROPIC_API_KEY")
@@ -143,7 +72,18 @@ def create_llm_provider(
         )
         return AnthropicProvider(api_key=api_key, model=model)
 
+    elif provider == "bedrock":
+        region = bedrock_region or os.environ.get(
+            "RAG_EVAL_BEDROCK_REGION", os.environ.get("AWS_REGION", "us-east-1")
+        )
+        model = bedrock_model or os.environ.get(
+            "RAG_EVAL_BEDROCK_MODEL", "anthropic.claude-3-sonnet-20240229-v1:0"
+        )
+        return BedrockProvider(
+            region_name=region, embedding_model=None, generation_model=model
+        )
+
     else:
         raise ValueError(
-            f"Invalid provider: {provider}. Must be 'ollama' or 'anthropic'."
+            f"Invalid provider: {provider}. Must be 'ollama', 'anthropic', or 'bedrock'."
         )

tests/unit/providers/__init__.py

@@ -0,0 +1 @@
+"""Unit tests for provider infrastructure."""

tests/unit/providers/test_bedrock.py

@@ -0,0 +1,280 @@
+"""Unit tests for Bedrock provider."""
+
+import json
+from unittest.mock import MagicMock
+
+import pytest
+
+from nextcloud_mcp_server.providers.bedrock import BOTO3_AVAILABLE, BedrockProvider
+
+
+@pytest.fixture
+def mock_bedrock_client(mocker):
+    """Mock boto3 bedrock-runtime client."""
+    if not BOTO3_AVAILABLE:
+        pytest.skip("boto3 not installed")
+
+    mock_client = MagicMock()
+    mocker.patch("boto3.client", return_value=mock_client)
+    return mock_client
+
+
+@pytest.mark.unit
+async def test_bedrock_embedding_titan(mock_bedrock_client):
+    """Test Bedrock embedding with Titan model."""
+    # Mock response
+    mock_response = {
+        "body": MagicMock(
+            read=MagicMock(
+                return_value=json.dumps({"embedding": [0.1, 0.2, 0.3]}).encode()
+            )
+        )
+    }
+    mock_bedrock_client.invoke_model.return_value = mock_response
+
+    # Create provider
+    provider = BedrockProvider(
+        region_name="us-east-1",
+        embedding_model="amazon.titan-embed-text-v2:0",
+        generation_model=None,
+    )
+
+    # Test embedding
+    embedding = await provider.embed("test text")
+
+    assert embedding == [0.1, 0.2, 0.3]
+    mock_bedrock_client.invoke_model.assert_called_once()
+    call_args = mock_bedrock_client.invoke_model.call_args
+
+    assert call_args.kwargs["modelId"] == "amazon.titan-embed-text-v2:0"
+    body = json.loads(call_args.kwargs["body"])
+    assert body == {"inputText": "test text"}
+
+
+@pytest.mark.unit
+async def test_bedrock_embedding_batch(mock_bedrock_client):
+    """Test Bedrock batch embedding."""
+    # Mock response
+    mock_response = {
+        "body": MagicMock(
+            read=MagicMock(
+                return_value=json.dumps({"embedding": [0.1, 0.2, 0.3]}).encode()
+            )
+        )
+    }
+    mock_bedrock_client.invoke_model.return_value = mock_response
+
+    # Create provider
+    provider = BedrockProvider(
+        region_name="us-east-1",
+        embedding_model="amazon.titan-embed-text-v2:0",
+        generation_model=None,
+    )
+
+    # Test batch embedding
+    embeddings = await provider.embed_batch(["text1", "text2"])
+
+    assert len(embeddings) == 2
+    assert embeddings[0] == [0.1, 0.2, 0.3]
+    assert embeddings[1] == [0.1, 0.2, 0.3]
+    assert mock_bedrock_client.invoke_model.call_count == 2
+
+
+@pytest.mark.unit
+async def test_bedrock_generation_claude(mock_bedrock_client):
+    """Test Bedrock text generation with Claude model."""
+    # Mock response
+    mock_response = {
+        "body": MagicMock(
+            read=MagicMock(
+                return_value=json.dumps(
+                    {"content": [{"text": "Generated response"}]}
+                ).encode()
+            )
+        )
+    }
+    mock_bedrock_client.invoke_model.return_value = mock_response
+
+    # Create provider
+    provider = BedrockProvider(
+        region_name="us-east-1",
+        embedding_model=None,
+        generation_model="anthropic.claude-3-sonnet-20240229-v1:0",
+    )
+
+    # Test generation
+    text = await provider.generate("test prompt", max_tokens=100)
+
+    assert text == "Generated response"
+    mock_bedrock_client.invoke_model.assert_called_once()
+    call_args = mock_bedrock_client.invoke_model.call_args
+
+    assert call_args.kwargs["modelId"] == "anthropic.claude-3-sonnet-20240229-v1:0"
+    body = json.loads(call_args.kwargs["body"])
+    assert body["messages"][0]["content"] == "test prompt"
+    assert body["max_tokens"] == 100
+
+
+@pytest.mark.unit
+async def test_bedrock_generation_llama(mock_bedrock_client):
+    """Test Bedrock text generation with Llama model."""
+    # Mock response
+    mock_response = {
+        "body": MagicMock(
+            read=MagicMock(
+                return_value=json.dumps({"generation": "Llama response"}).encode()
+            )
+        )
+    }
+    mock_bedrock_client.invoke_model.return_value = mock_response
+
+    # Create provider
+    provider = BedrockProvider(
+        region_name="us-east-1",
+        embedding_model=None,
+        generation_model="meta.llama3-8b-instruct-v1:0",
+    )
+
+    # Test generation
+    text = await provider.generate("test prompt")
+
+    assert text == "Llama response"
+    body = json.loads(mock_bedrock_client.invoke_model.call_args.kwargs["body"])
+    assert body["prompt"] == "test prompt"
+    assert "max_gen_len" in body
+
+
+@pytest.mark.unit
+async def test_bedrock_both_capabilities(mock_bedrock_client):
+    """Test Bedrock with both embedding and generation models."""
+    # Mock responses
+    embed_response = {
+        "body": MagicMock(
+            read=MagicMock(return_value=json.dumps({"embedding": [0.1, 0.2]}).encode())
+        )
+    }
+    gen_response = {
+        "body": MagicMock(
+            read=MagicMock(
+                return_value=json.dumps({"content": [{"text": "Response"}]}).encode()
+            )
+        )
+    }
+
+    # Mock to return different responses based on modelId
+    def mock_invoke(modelId, body, **kwargs):
+        if "embed" in modelId:
+            return embed_response
+        else:
+            return gen_response
+
+    mock_bedrock_client.invoke_model.side_effect = mock_invoke
+
+    # Create provider with both models
+    provider = BedrockProvider(
+        region_name="us-east-1",
+        embedding_model="amazon.titan-embed-text-v2:0",
+        generation_model="anthropic.claude-3-sonnet-20240229-v1:0",
+    )
+
+    assert provider.supports_embeddings is True
+    assert provider.supports_generation is True
+
+    # Test both capabilities
+    embedding = await provider.embed("test")
+    assert embedding == [0.1, 0.2]
+
+    text = await provider.generate("test")
+    assert text == "Response"
+
+
+@pytest.mark.unit
+async def test_bedrock_no_embeddings():
+    """Test Bedrock provider with no embedding model raises error."""
+    provider = BedrockProvider(
+        region_name="us-east-1",
+        embedding_model=None,
+        generation_model="anthropic.claude-3-sonnet-20240229-v1:0",
+    )
+
+    assert provider.supports_embeddings is False
+
+    with pytest.raises(NotImplementedError, match="no embedding_model configured"):
+        await provider.embed("test")
+
+    with pytest.raises(NotImplementedError, match="no embedding_model configured"):
+        await provider.embed_batch(["test"])
+
+    with pytest.raises(NotImplementedError, match="no embedding_model configured"):
+        provider.get_dimension()
+
+
+@pytest.mark.unit
+async def test_bedrock_no_generation():
+    """Test Bedrock provider with no generation model raises error."""
+    provider = BedrockProvider(
+        region_name="us-east-1",
+        embedding_model="amazon.titan-embed-text-v2:0",
+        generation_model=None,
+    )
+
+    assert provider.supports_generation is False
+
+    with pytest.raises(NotImplementedError, match="no generation_model configured"):
+        await provider.generate("test")
+
+
+@pytest.mark.unit
+async def test_bedrock_dimension_detection(mock_bedrock_client):
+    """Test dimension detection for Bedrock embeddings."""
+    # Mock response with specific dimension
+    mock_response = {
+        "body": MagicMock(
+            read=MagicMock(
+                return_value=json.dumps(
+                    {"embedding": [0.1] * 1536}  # 1536-dim embedding
+                ).encode()
+            )
+        )
+    }
+    mock_bedrock_client.invoke_model.return_value = mock_response
+
+    provider = BedrockProvider(
+        region_name="us-east-1",
+        embedding_model="amazon.titan-embed-text-v2:0",
+    )
+
+    # Dimension not detected yet
+    with pytest.raises(RuntimeError, match="not detected yet"):
+        provider.get_dimension()
+
+    # Detect dimension
+    await provider._detect_dimension()
+
+    # Now dimension should be available
+    assert provider.get_dimension() == 1536
+
+
+@pytest.mark.unit
+async def test_bedrock_cohere_embedding(mock_bedrock_client):
+    """Test Bedrock with Cohere embedding model."""
+    # Mock response
+    mock_response = {
+        "body": MagicMock(
+            read=MagicMock(
+                return_value=json.dumps({"embeddings": [[0.1, 0.2, 0.3]]}).encode()
+            )
+        )
+    }
+    mock_bedrock_client.invoke_model.return_value = mock_response
+
+    provider = BedrockProvider(
+        region_name="us-east-1",
+        embedding_model="cohere.embed-english-v3",
+    )
+
+    embedding = await provider.embed("test text")
+
+    assert embedding == [0.1, 0.2, 0.3]
+    body = json.loads(mock_bedrock_client.invoke_model.call_args.kwargs["body"])
+    assert body == {"texts": ["test text"], "input_type": "search_document"}

uv.lock

@@ -233,6 +233,34 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/f8/aa/5082412d1ee302e9e7d80b6949bc4d2a8fa1149aaab610c5fc24709605d6/authlib-1.6.5-py2.py3-none-any.whl", hash = "sha256:3e0e0507807f842b02175507bdee8957a1d5707fd4afb17c32fb43fee90b6e3a", size = 243608, upload-time = "2025-10-02T13:36:07.637Z" },
 ]
 
+[[package]]
+name = "boto3"
+version = "1.40.74"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "botocore" },
+    { name = "jmespath" },
+    { name = "s3transfer" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/a2/37/0db5fc46548b347255310893f1a47971a1d8eb0dbc46dfb5ace8a1e7d45e/boto3-1.40.74.tar.gz", hash = "sha256:484e46bf394b03a7c31b34f90945ebe1390cb1e2ac61980d128a9079beac87d4", size = 111592, upload-time = "2025-11-14T20:29:10.991Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/d2/08/c52751748762901c0ca3c3019e3aa950010217f0fdf9940ebe68e6bb2f5a/boto3-1.40.74-py3-none-any.whl", hash = "sha256:41fc8844b37ae27b24bcabf8369769df246cc12c09453988d0696ad06d6aa9ef", size = 139360, upload-time = "2025-11-14T20:29:09.477Z" },
+]
+
+[[package]]
+name = "botocore"
+version = "1.40.74"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "jmespath" },
+    { name = "python-dateutil" },
+    { name = "urllib3" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/81/dc/0412505f05286f282a75bb0c650e525ddcfaf3f6f1a05cd8e99d32a2db06/botocore-1.40.74.tar.gz", hash = "sha256:57de0b9ffeada06015b3c7e5186c77d0692b210d9e5efa294f3214df97e2f8ee", size = 14452479, upload-time = "2025-11-14T20:29:00.949Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/7d/a2/306dec16e3c84f3ca7aaead0084358c1c7fbe6501f6160844cbc93bc871e/botocore-1.40.74-py3-none-any.whl", hash = "sha256:f39f5763e35e75f0bd91212b7b36120b1536203e8003cd952ef527db79702b15", size = 14117911, upload-time = "2025-11-14T20:28:58.153Z" },
+]
+
 [[package]]
 name = "caldav"
 version = "2.0.2.dev47+g3e44cf827"
@@ -1296,6 +1324,15 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/2f/9c/6753e6522b8d0ef07d3a3d239426669e984fb0eba15a315cdbc1253904e4/jiter-0.12.0-graalpy312-graalpy250_312_native-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:c24e864cb30ab82311c6425655b0cdab0a98c5d973b065c66a3f020740c2324c", size = 346110, upload-time = "2025-11-09T20:49:21.817Z" },
 ]
 
+[[package]]
+name = "jmespath"
+version = "1.0.1"
+source = { registry = "https://pypi.org/simple" }
+sdist = { url = "https://files.pythonhosted.org/packages/00/2a/e867e8531cf3e36b41201936b7fa7ba7b5702dbef42922193f05c8976cd6/jmespath-1.0.1.tar.gz", hash = "sha256:90261b206d6defd58fdd5e85f478bf633a2901798906be2ad389150c5c60edbe", size = 25843, upload-time = "2022-06-17T18:00:12.224Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/31/b4/b9b800c45527aadd64d5b442f9b932b00648617eb5d63d2c7a6587b7cafc/jmespath-1.0.1-py3-none-any.whl", hash = "sha256:02e2e4cc71b5bcab88332eebf907519190dd9e6e82107fa7f83b1003a6252980", size = 20256, upload-time = "2022-06-17T18:00:10.251Z" },
+]
+
 [[package]]
 name = "jsonschema"
 version = "4.25.1"
@@ -1849,6 +1886,7 @@ dependencies = [
 [package.dev-dependencies]
 dev = [
     { name = "anthropic" },
+    { name = "boto3" },
     { name = "commitizen" },
     { name = "datasets" },
     { name = "ipython" },
@@ -1891,6 +1929,7 @@ requires-dist = [
 [package.metadata.requires-dev]
 dev = [
     { name = "anthropic", specifier = ">=0.42.0" },
+    { name = "boto3", specifier = ">=1.35.0" },
     { name = "commitizen", specifier = ">=4.8.2" },
     { name = "datasets", specifier = ">=3.3.0" },
     { name = "ipython", specifier = ">=9.2.0" },
@@ -3270,6 +3309,18 @@ wheels = [
     { url = "https://files.pythonhosted.org/packages/e5/80/69756670caedcf3b9be597a6e12276a6cf6197076eb62aad0c608f8efce0/ruff-0.14.5-py3-none-win_arm64.whl", hash = "sha256:4b700459d4649e2594b31f20a9de33bc7c19976d4746d8d0798ad959621d64a4", size = 13433331, upload-time = "2025-11-13T19:58:48.434Z" },
 ]
 
+[[package]]
+name = "s3transfer"
+version = "0.14.0"
+source = { registry = "https://pypi.org/simple" }
+dependencies = [
+    { name = "botocore" },
+]
+sdist = { url = "https://files.pythonhosted.org/packages/62/74/8d69dcb7a9efe8baa2046891735e5dfe433ad558ae23d9e3c14c633d1d58/s3transfer-0.14.0.tar.gz", hash = "sha256:eff12264e7c8b4985074ccce27a3b38a485bb7f7422cc8046fee9be4983e4125", size = 151547, upload-time = "2025-09-09T19:23:31.089Z" }
+wheels = [
+    { url = "https://files.pythonhosted.org/packages/48/f0/ae7ca09223a81a1d890b2557186ea015f6e0502e9b8cb8e1813f1d8cfa4e/s3transfer-0.14.0-py3-none-any.whl", hash = "sha256:ea3b790c7077558ed1f02a3072fb3cb992bbbd253392f4b6e9e8976941c7d456", size = 85712, upload-time = "2025-09-09T19:23:30.041Z" },
+]
+
 [[package]]
 name = "shellingham"
 version = "1.5.4"