Support OpenRouter as Embedder

README.md

@@ -35,7 +35,7 @@
 - **Ask Feature**: Chat with your repository using RAG-powered AI to get accurate answers
 - **DeepResearch**: Multi-turn research process that thoroughly investigates complex topics
 - **Multiple Model Providers**: Support for Google Gemini, OpenAI, OpenRouter, and local Ollama models
-- **Flexible Embeddings**: Choose between OpenAI, Google AI, or local Ollama embeddings for optimal performance
+- **Flexible Embeddings**: Choose between OpenAI, Google AI, OpenRouter, or local Ollama embeddings for optimal performance
 
 ## 🚀 Quick Start (Super Easy!)
 
@@ -341,6 +341,7 @@ docker-compose up
 |------|-------------|------------------|-------|
 | `openai` | OpenAI embeddings (default) | `OPENAI_API_KEY` | Uses `text-embedding-3-small` model |
 | `google` | Google AI embeddings | `GOOGLE_API_KEY` | Uses `text-embedding-004` model |
+| `openrouter` | OpenRouter embeddings | `OPENROUTER_API_KEY` | Uses `openai/text-embedding-3-small` via OpenRouter |
 | `ollama` | Local Ollama embeddings | None | Requires local Ollama installation |
 
 ### Why Use Google AI Embeddings?
@@ -361,12 +362,52 @@ export DEEPWIKI_EMBEDDER_TYPE=openai
 # Use Google AI embeddings
 export DEEPWIKI_EMBEDDER_TYPE=google
 
+# Use OpenRouter embeddings
+export DEEPWIKI_EMBEDDER_TYPE=openrouter
+
 # Use local Ollama embeddings
 export DEEPWIKI_EMBEDDER_TYPE=ollama
 ```
 
 **Note**: When switching embedders, you may need to regenerate your repository embeddings as different models produce different vector spaces.
 
+## 🔀 Using OpenRouter Embeddings
+
+DeepWiki supports OpenRouter as an embedding provider. OpenRouter exposes an OpenAI-compatible embeddings endpoint, so any embedding model available on OpenRouter can be used.
+
+### How to Enable OpenRouter Embeddings
+
+Set the embedder type and your OpenRouter API key in your `.env` file:
+
+```bash
+OPENROUTER_API_KEY=your_openrouter_api_key
+DEEPWIKI_EMBEDDER_TYPE=openrouter
+```
+
+**Docker:**
+
+```bash
+docker run -p 8001:8001 -p 3000:3000 \
+ -e OPENROUTER_API_KEY=your_openrouter_api_key \
+ -e DEEPWIKI_EMBEDDER_TYPE=openrouter \
+ -v ~/.adalflow:/root/.adalflow \
+ ghcr.io/asyncfuncai/deepwiki-open:latest
+```
+
+### Changing the Embedding Model
+
+The default model is `openai/text-embedding-3-small`. To use a different model, edit `api/config/embedder.json` and change the `model` field under `embedder_openrouter`:
+
+```json
+"embedder_openrouter": {
+ "model_kwargs": {
+ "model": "openai/text-embedding-3-large"
+ }
+}
+```
+
+Any embedding model listed on [OpenRouter](https://openrouter.ai/models) that exposes an OpenAI-compatible embeddings endpoint can be used here.
+
 ### Logging
 
 DeepWiki uses Python's built-in `logging` module for diagnostic output. You can configure the verbosity and log file destination via environment variables:
@@ -411,7 +452,7 @@ docker-compose up
 |----------------------|--------------------------------------------------------------|----------|----------------------------------------------------------------------------------------------------------|
 | `GOOGLE_API_KEY` | Google Gemini API key for AI generation and embeddings | No | Required for Google Gemini models and Google AI embeddings 
 | `OPENAI_API_KEY` | OpenAI API key for embeddings and models | Conditional | Required if using OpenAI embeddings or models |
-| `OPENROUTER_API_KEY` | OpenRouter API key for alternative models    | No | Required only if you want to use OpenRouter models   |
+| `OPENROUTER_API_KEY` | OpenRouter API key for alternative models and embeddings | No | Required if using OpenRouter models or `DEEPWIKI_EMBEDDER_TYPE=openrouter` |
 | `AWS_ACCESS_KEY_ID` | AWS access key ID for Bedrock | No | Required for Bedrock if not using instance/role-based credentials |
 | `AWS_SECRET_ACCESS_KEY` | AWS secret access key for Bedrock | No | Required for Bedrock if not using instance/role-based credentials |
 | `AWS_SESSION_TOKEN` | AWS session token for Bedrock (STS) | No | Required when using temporary credentials |
@@ -421,15 +462,16 @@ docker-compose up
 | `AZURE_OPENAI_ENDPOINT` | Azure OpenAI endpoint | No | Required only if you want to use Azure OpenAI models |
 | `AZURE_OPENAI_VERSION` | Azure OpenAI version | No | Required only if you want to use Azure OpenAI models |
 | `OLLAMA_HOST` | Ollama Host (default: http://localhost:11434) | No | Required only if you want to use external Ollama server |
-| `DEEPWIKI_EMBEDDER_TYPE` | Embedder type: `openai`, `google`, `ollama`, or `bedrock` (default: `openai`) | No | Controls which embedding provider to use |
+| `DEEPWIKI_EMBEDDER_TYPE` | Embedder type: `openai`, `google`, `ollama`, `bedrock`, or `openrouter` (default: `openai`) | No | Controls which embedding provider to use |
 | `PORT` | Port for the API server (default: 8001) | No | If you host API and frontend on the same machine, make sure change port of `SERVER_BASE_URL` accordingly |
 | `SERVER_BASE_URL` | Base URL for the API server (default: http://localhost:8001) | No |
 | `DEEPWIKI_AUTH_MODE` | Set to `true` or `1` to enable authorization mode. | No | Defaults to `false`. If enabled, `DEEPWIKI_AUTH_CODE` is required. |
 | `DEEPWIKI_AUTH_CODE` | The secret code required for wiki generation when `DEEPWIKI_AUTH_MODE` is enabled. | No | Only used if `DEEPWIKI_AUTH_MODE` is `true` or `1`. |
 
 **API Key Requirements:**
 - If using `DEEPWIKI_EMBEDDER_TYPE=openai` (default): `OPENAI_API_KEY` is required
-- If using `DEEPWIKI_EMBEDDER_TYPE=google`: `GOOGLE_API_KEY` is required 
+- If using `DEEPWIKI_EMBEDDER_TYPE=google`: `GOOGLE_API_KEY` is required
+- If using `DEEPWIKI_EMBEDDER_TYPE=openrouter`: `OPENROUTER_API_KEY` is required
 - If using `DEEPWIKI_EMBEDDER_TYPE=ollama`: No API key required (local processing)
 - If using `DEEPWIKI_EMBEDDER_TYPE=bedrock`: AWS credentials (or role-based credentials) are required
 

api/README.md

@@ -29,7 +29,7 @@ GOOGLE_API_KEY=your_google_api_key # Required for Google Gemini models
 OPENAI_API_KEY=your_openai_api_key # Required for embeddings and OpenAI models
 
 # Optional API Keys
-OPENROUTER_API_KEY=your_openrouter_api_key # Required only if using OpenRouter models
+OPENROUTER_API_KEY=your_openrouter_api_key # Required if using OpenRouter models or embeddings (DEEPWIKI_EMBEDDER_TYPE=openrouter)
 
 # AWS Bedrock Configuration
 AWS_ACCESS_KEY_ID=your_aws_access_key_id # Required for AWS Bedrock models
@@ -47,7 +47,7 @@ OLLAMA_HOST=https://your_ollama_host" # Optional: Add Ollama host if not local.
 PORT=8001 # Optional, defaults to 8001
 ```
 
-If you're not using Ollama mode, you need to configure an OpenAI API key for embeddings. Other API keys are only required when configuring and using models from the corresponding providers.
+The embedder type is controlled by `DEEPWIKI_EMBEDDER_TYPE` (`openai`, `google`, `openrouter`, `ollama`, or `bedrock`; default: `openai`). The corresponding API key is required for the chosen embedder. Other API keys are only required when configuring and using models from the corresponding providers.
 
 > 💡 **Where to get these keys:**
 > - Get a Google API key from [Google AI Studio](https://makersuite.google.com/app/apikey)

api/config.py

@@ -152,7 +152,7 @@ def load_embedder_config():
  embedder_config = load_json_config("embedder.json")
 
  # Process client classes
- for key in ["embedder", "embedder_ollama", "embedder_google", "embedder_bedrock"]:
+ for key in ["embedder", "embedder_ollama", "embedder_google", "embedder_bedrock", "embedder_openrouter"]:
  if key in embedder_config and "client_class" in embedder_config[key]:
  class_name = embedder_config[key]["client_class"]
  if class_name in CLIENT_CLASSES:
@@ -174,6 +174,8 @@ def get_embedder_config():
  return configs.get("embedder_google", {})
  elif embedder_type == 'ollama' and 'embedder_ollama' in configs:
  return configs.get("embedder_ollama", {})
+ elif embedder_type == 'openrouter' and 'embedder_openrouter' in configs:
+ return configs.get("embedder_openrouter", {})
  else:
  return configs.get("embedder", {})
 
@@ -235,19 +237,47 @@ def is_bedrock_embedder():
  client_class = embedder_config.get("client_class", "")
  return client_class == "BedrockClient"
 
+def is_openrouter_embedder():
+ """
+ Check if the current embedder configuration uses OpenRouter for embeddings.
+
+ Detects OpenRouter by inspecting the loaded configuration's initialize_kwargs
+ for a base_url pointing to openrouter.ai, consistent with how other is_*_embedder
+ helpers inspect the config rather than reading environment variables directly.
+
+ Returns:
+ bool: True if using OpenRouter for embeddings, False otherwise
+ """
+ embedder_config = get_embedder_config()
+ if not embedder_config:
+ return False
+
+ initialize_kwargs = embedder_config.get("initialize_kwargs", {})
+ base_url = initialize_kwargs.get("base_url", "")
+ if base_url and "openrouter.ai" in base_url:
+ return True
+
+ # Fallback: explicit OpenRouterClient class
+ model_client = embedder_config.get("model_client")
+ if model_client:
+ return model_client.__name__ == "OpenRouterClient"
+ return embedder_config.get("client_class", "") == "OpenRouterClient"
+
 def get_embedder_type():
  """
  Get the current embedder type based on configuration.
- 
+
  Returns:
- str: 'bedrock', 'ollama', 'google', or 'openai' (default)
+ str: 'bedrock', 'ollama', 'google', 'openrouter', or 'openai' (default)
  """
  if is_bedrock_embedder():
  return 'bedrock'
  elif is_ollama_embedder():
  return 'ollama'
  elif is_google_embedder():
  return 'google'
+ elif is_openrouter_embedder():
+ return 'openrouter'
  else:
  return 'openai'
 
@@ -341,7 +371,7 @@ def load_lang_config():
 
 # Update embedder configuration
 if embedder_config:
- for key in ["embedder", "embedder_ollama", "embedder_google", "embedder_bedrock", "retriever", "text_splitter"]:
+ for key in ["embedder", "embedder_ollama", "embedder_google", "embedder_bedrock", "embedder_openrouter", "retriever", "text_splitter"]:
  if key in embedder_config:
  configs[key] = embedder_config[key]
 

api/config/embedder.json

@@ -30,6 +30,18 @@
  "dimensions": 256
  }
  },
+ "embedder_openrouter": {
+ "client_class": "OpenAIClient",
+ "initialize_kwargs": {
+ "base_url": "https://openrouter.ai/api/v1",
+ "env_api_key_name": "OPENROUTER_API_KEY"
+ },
+ "batch_size": 500,
+ "model_kwargs": {
+ "model": "openai/text-embedding-3-small",
+ "encoding_format": "float"
+ }
+ },
  "retriever": {
  "top_k": 20
  },

api/tools/embedder.py

@@ -9,7 +9,7 @@ def get_embedder(is_local_ollama: bool = False, use_google_embedder: bool = Fals
  Args:
  is_local_ollama: Legacy parameter for Ollama embedder
  use_google_embedder: Legacy parameter for Google embedder 
- embedder_type: Direct specification of embedder type ('ollama', 'google', 'bedrock', 'openai')
+ embedder_type: Direct specification of embedder type ('ollama', 'google', 'bedrock', 'openrouter', 'openai')
 
  Returns:
  adal.Embedder: Configured embedder instance
@@ -22,6 +22,8 @@ def get_embedder(is_local_ollama: bool = False, use_google_embedder: bool = Fals
  embedder_config = configs["embedder_google"]
  elif embedder_type == 'bedrock':
  embedder_config = configs["embedder_bedrock"]
+ elif embedder_type == 'openrouter':
+ embedder_config = configs["embedder_openrouter"]
  else: # default to openai
  embedder_config = configs["embedder"]
  elif is_local_ollama:
@@ -37,6 +39,8 @@ def get_embedder(is_local_ollama: bool = False, use_google_embedder: bool = Fals
  embedder_config = configs["embedder_ollama"]
  elif current_type == 'google':
  embedder_config = configs["embedder_google"]
+ elif current_type == 'openrouter':
+ embedder_config = configs["embedder_openrouter"]
  else:
  embedder_config = configs["embedder"]