LightRAG provides optional API support through FastAPI servers that add RAG capabilities to existing LLM services. You can install LightRAG with API support in two ways:
pip install "lightrag-hku[api]"# Clone the repository
git clone https://github.com/HKUDS/lightrag.git
# Change to the repository directory
cd lightrag
# create a Python virtual enviroment if neccesary
# Install in editable mode with API support
pip install -e ".[api]"Before running any of the servers, ensure you have the corresponding backend service running for both llm and embedding. The new api allows you to mix different bindings for llm/embeddings. For example, you have the possibility to use ollama for the embedding and openai for the llm.
- LoLLMs must be running and accessible
- Default connection: http://localhost:9600
- Configure using --llm-binding-host and/or --embedding-binding-host if running on a different host/port
- Ollama must be running and accessible
- Requires environment variables setup or command line argument provided
- Environment variables: LLM_BINDING=ollama, LLM_BINDING_HOST, LLM_MODEL
- Command line arguments: --llm-binding=ollama, --llm-binding-host, --llm-model
- Default connection is http://localhost:11434 if not priveded
The default MAX_TOKENS(num_ctx) for Ollama is 32768. If your Ollama server is lacking or GPU memory, set it to a lower value.
- Requires environment variables setup or command line argument provided
- Environment variables: LLM_BINDING=ollama, LLM_BINDING_HOST, LLM_MODEL, LLM_BINDING_API_KEY
- Command line arguments: --llm-binding=ollama, --llm-binding-host, --llm-model, --llm-binding-api-key
- Default connection is https://api.openai.com/v1 if not priveded
Azure OpenAI API can be created using the following commands in Azure CLI (you need to install Azure CLI first from https://docs.microsoft.com/en-us/cli/azure/install-azure-cli):
# Change the resource group name, location and OpenAI resource name as needed
RESOURCE_GROUP_NAME=LightRAG
LOCATION=swedencentral
RESOURCE_NAME=LightRAG-OpenAI
az login
az group create --name $RESOURCE_GROUP_NAME --location $LOCATION
az cognitiveservices account create --name $RESOURCE_NAME --resource-group $RESOURCE_GROUP_NAME --kind OpenAI --sku S0 --location swedencentral
az cognitiveservices account deployment create --resource-group $RESOURCE_GROUP_NAME --model-format OpenAI --name $RESOURCE_NAME --deployment-name gpt-4o --model-name gpt-4o --model-version "2024-08-06" --sku-capacity 100 --sku-name "Standard"
az cognitiveservices account deployment create --resource-group $RESOURCE_GROUP_NAME --model-format OpenAI --name $RESOURCE_NAME --deployment-name text-embedding-3-large --model-name text-embedding-3-large --model-version "1" --sku-capacity 80 --sku-name "Standard"
az cognitiveservices account show --name $RESOURCE_NAME --resource-group $RESOURCE_GROUP_NAME --query "properties.endpoint"
az cognitiveservices account keys list --name $RESOURCE_NAME -g $RESOURCE_GROUP_NAME
The output of the last command will give you the endpoint and the key for the OpenAI API. You can use these values to set the environment variables in the .env file.
LLM_BINDING=azure_openai
LLM_BINDING_HOST=endpoint_of_azure_ai
LLM_MODEL=model_name_of_azure_ai
LLM_BINDING_API_KEY=api_key_of_azure_ai
We provide an Ollama-compatible interfaces for LightRAG, aiming to emulate LightRAG as an Ollama chat model. This allows AI chat frontends supporting Ollama, such as Open WebUI, to access LightRAG easily.
A query prefix in the query string can determines which LightRAG query mode is used to generate the respond for the query. The supported prefixes include:
/local
/global
/hybrid
/naive
/mix
/bypass
For example, chat message "/mix 唐僧有几个徒弟" will trigger a mix mode query for LighRAG. A chat message without query prefix will trigger a hybrid mode query by default。
"/bypass" is not a LightRAG query mode, it will tell API Server to pass the query directly to the underlying LLM with chat history. So user can use LLM to answer question base on the LightRAG query results. (If you are using Open WebUI as front end, you can just switch the model to a normal LLM instead of using /bypass prefix)
After starting the lightrag-server, you can add an Ollama-type connection in the Open WebUI admin pannel. And then a model named lightrag:latest will appear in Open WebUI's model management interface. Users can then send queries to LightRAG through the chat interface.
LightRAG can be configured using either command-line arguments or environment variables. When both are provided, command-line arguments take precedence over environment variables.
For better performance, the API server's default values for TOP_K and COSINE_THRESHOLD are set to 50 and 0.4 respectively. If COSINE_THRESHOLD remains at its default value of 0.2 in LightRAG, many irrelevant entities and relations would be retrieved and sent to the LLM.
You can configure LightRAG using environment variables by creating a .env file in your project root directory. Here's a complete example of available environment variables:
# Server Configuration
HOST=0.0.0.0
PORT=9621
# Directory Configuration
WORKING_DIR=/app/data/rag_storage
INPUT_DIR=/app/data/inputs
# RAG Configuration
MAX_ASYNC=4
MAX_TOKENS=32768
EMBEDDING_DIM=1024
MAX_EMBED_TOKENS=8192
#HISTORY_TURNS=3
#CHUNK_SIZE=1200
#CHUNK_OVERLAP_SIZE=100
#COSINE_THRESHOLD=0.4
#TOP_K=50
# LLM Configuration
LLM_BINDING=ollama
LLM_BINDING_HOST=http://localhost:11434
LLM_MODEL=mistral-nemo:latest
# must be set if using OpenAI LLM (LLM_MODEL must be set or set by command line parms)
OPENAI_API_KEY=you_api_key
# Embedding Configuration
EMBEDDING_BINDING=ollama
EMBEDDING_BINDING_HOST=http://localhost:11434
EMBEDDING_MODEL=bge-m3:latest
# Security
#LIGHTRAG_API_KEY=you-api-key-for-accessing-LightRAG
# Logging
LOG_LEVEL=INFO
# Optional SSL Configuration
#SSL=true
#SSL_CERTFILE=/path/to/cert.pem
#SSL_KEYFILE=/path/to/key.pem
# Optional Timeout
#TIMEOUT=30The configuration values are loaded in the following order (highest priority first):
- Command-line arguments
- Environment variables
- Default values
For example:
# This command-line argument will override both the environment variable and default value
python lightrag.py --port 8080
# The environment variable will override the default value but not the command-line argument
PORT=7000 python lightrag.py| Parameter | Default | Description |
|---|---|---|
| --host | 0.0.0.0 | Server host |
| --port | 9621 | Server port |
| --llm-binding | ollama | LLM binding to be used. Supported: lollms, ollama, openai |
| --llm-binding-host | (dynamic) | LLM server host URL. Defaults based on binding: http://localhost:11434 (ollama), http://localhost:9600 (lollms), https://api.openai.com/v1 (openai) |
| --llm-model | mistral-nemo:latest | LLM model name |
| --llm-binding-api-key | None | API Key for OpenAI Alike LLM |
| --embedding-binding | ollama | Embedding binding to be used. Supported: lollms, ollama, openai |
| --embedding-binding-host | (dynamic) | Embedding server host URL. Defaults based on binding: http://localhost:11434 (ollama), http://localhost:9600 (lollms), https://api.openai.com/v1 (openai) |
| --embedding-model | bge-m3:latest | Embedding model name |
| --working-dir | ./rag_storage | Working directory for RAG storage |
| --input-dir | ./inputs | Directory containing input documents |
| --max-async | 4 | Maximum async operations |
| --max-tokens | 32768 | Maximum token size |
| --embedding-dim | 1024 | Embedding dimensions |
| --max-embed-tokens | 8192 | Maximum embedding token size |
| --timeout | None | Timeout in seconds (useful when using slow AI). Use None for infinite timeout |
| --log-level | INFO | Logging level (DEBUG, INFO, WARNING, ERROR, CRITICAL) |
| --key | None | API key for authentication. Protects lightrag server against unauthorized access |
| --ssl | False | Enable HTTPS |
| --ssl-certfile | None | Path to SSL certificate file (required if --ssl is enabled) |
| --ssl-keyfile | None | Path to SSL private key file (required if --ssl is enabled) |
| --top-k | 50 | Number of top-k items to retrieve; corresponds to entities in "local" mode and relationships in "global" mode. |
| --cosine-threshold | 0.4 | The cossine threshold for nodes and relations retrieval, works with top-k to control the retrieval of nodes and relations. |
Ollama is the default backend for both llm and embedding, so by default you can run lightrag-server with no parameters and the default ones will be used. Make sure ollama is installed and is running and default models are already installed on ollama.
# Run lightrag with ollama, mistral-nemo:latest for llm, and bge-m3:latest for embedding
lightrag-server
# Using specific models (ensure they are installed in your ollama instance)
lightrag-server --llm-model adrienbrault/nous-hermes2theta-llama3-8b:f16 --embedding-model nomic-embed-text --embedding-dim 1024
# Using an authentication key
lightrag-server --key my-key
# Using lollms for llm and ollama for embedding
lightrag-server --llm-binding lollms# Run lightrag with lollms, mistral-nemo:latest for llm, and bge-m3:latest for embedding, use lollms for both llm and embedding
lightrag-server --llm-binding lollms --embedding-binding lollms
# Using specific models (ensure they are installed in your ollama instance)
lightrag-server --llm-binding lollms --llm-model adrienbrault/nous-hermes2theta-llama3-8b:f16 --embedding-binding lollms --embedding-model nomic-embed-text --embedding-dim 1024
# Using an authentication key
lightrag-server --key my-key
# Using lollms for llm and openai for embedding
lightrag-server --llm-binding lollms --embedding-binding openai --embedding-model text-embedding-3-small# Run lightrag with lollms, GPT-4o-mini for llm, and text-embedding-3-small for embedding, use openai for both llm and embedding
lightrag-server --llm-binding openai --llm-model GPT-4o-mini --embedding-binding openai --embedding-model text-embedding-3-small
# Using an authentication key
lightrag-server --llm-binding openai --llm-model GPT-4o-mini --embedding-binding openai --embedding-model text-embedding-3-small --key my-key
# Using lollms for llm and openai for embedding
lightrag-server --llm-binding lollms --embedding-binding openai --embedding-model text-embedding-3-small# Run lightrag with lollms, GPT-4o-mini for llm, and text-embedding-3-small for embedding, use openai for both llm and embedding
lightrag-server --llm-binding azure_openai --llm-model GPT-4o-mini --embedding-binding openai --embedding-model text-embedding-3-small
# Using an authentication key
lightrag-server --llm-binding azure_openai --llm-model GPT-4o-mini --embedding-binding azure_openai --embedding-model text-embedding-3-small --key my-key
# Using lollms for llm and azure_openai for embedding
lightrag-server --llm-binding lollms --embedding-binding azure_openai --embedding-model text-embedding-3-smallImportant Notes:
- For LoLLMs: Make sure the specified models are installed in your LoLLMs instance
- For Ollama: Make sure the specified models are installed in your Ollama instance
- For OpenAI: Ensure you have set up your OPENAI_API_KEY environment variable
- For Azure OpenAI: Build and configure your server as stated in the Prequisites section
For help on any server, use the --help flag:
lightrag-server --helpNote: If you don't need the API functionality, you can install the base package without API support using:
pip install lightrag-hkuAll servers (LoLLMs, Ollama, OpenAI and Azure OpenAI) provide the same REST API endpoints for RAG functionality.
Query the RAG system with options for different search modes.
curl -X POST "http://localhost:9621/query" \
-H "Content-Type: application/json" \
-d '{"query": "Your question here", "mode": "hybrid", ""}'Stream responses from the RAG system.
curl -X POST "http://localhost:9621/query/stream" \
-H "Content-Type: application/json" \
-d '{"query": "Your question here", "mode": "hybrid"}'Insert text directly into the RAG system.
curl -X POST "http://localhost:9621/documents/text" \
-H "Content-Type: application/json" \
-d '{"text": "Your text content here", "description": "Optional description"}'Upload a single file to the RAG system.
curl -X POST "http://localhost:9621/documents/file" \
-F "file=@/path/to/your/document.txt" \
-F "description=Optional description"Upload multiple files at once.
curl -X POST "http://localhost:9621/documents/batch" \
-F "files=@/path/to/doc1.txt" \
-F "files=@/path/to/doc2.txt"Trigger document scan for new files in the Input directory.
curl -X POST "http://localhost:9621/documents/scan" --max-time 1800Ajust max-time according to the estimated index time for all new files.
Get Ollama version information
curl http://localhost:9621/api/versionGet Ollama available models
curl http://localhost:9621/api/tagsHandle chat completion requests
curl -N -X POST http://localhost:9621/api/chat -H "Content-Type: application/json" -d \
'{"model":"lightrag:latest","messages":[{"role":"user","content":"猪八戒是谁"}],"stream":true}'For more information about Ollama API pls. visit : Ollama API documentation
Clear all documents from the RAG system.
curl -X DELETE "http://localhost:9621/documents"Check server health and configuration.
curl "http://localhost:9621/health"Contribute to the project: Guide
For LoLLMs:
uvicorn lollms_lightrag_server:app --reload --port 9621For Ollama:
uvicorn ollama_lightrag_server:app --reload --port 9621For OpenAI:
uvicorn openai_lightrag_server:app --reload --port 9621For Azure OpenAI:
uvicorn azure_openai_lightrag_server:app --reload --port 9621When any server is running, visit:
- Swagger UI: http://localhost:9621/docs
- ReDoc: http://localhost:9621/redoc
You can test the API endpoints using the provided curl commands or through the Swagger UI interface. Make sure to:
- Start the appropriate backend service (LoLLMs, Ollama, or OpenAI)
- Start the RAG server
- Upload some documents using the document management endpoints
- Query the system using the query endpoints
- Trigger document scan if new files is put into inputs directory
When starting any of the servers with the --input-dir parameter, the system will automatically:
- Check for existing vectorized content in the database
- Only vectorize new documents that aren't already in the database
- Make all content immediately available for RAG queries
This intelligent caching mechanism:
- Prevents unnecessary re-vectorization of existing documents
- Reduces startup time for subsequent runs
- Preserves system resources
- Maintains consistency across restarts
Important Notes:
- The
--input-dirparameter enables automatic document processing at startup - Documents already in the database are not re-vectorized
- Only new documents in the input directory will be processed
- This optimization significantly reduces startup time for subsequent runs
- The working directory (
--working-dir) stores the vectorized documents database
Create your service file: lightrag-server.sevice. Modified the following lines from lightrag-server.sevice.example
Description=LightRAG Ollama Service
WorkingDirectory=<lightrag installed directory>
ExecStart=<lightrag installed directory>/lightrag/api/start_lightrag.sh
Create your service startup script: start_lightrag.sh. Change you python virtual environment activation method as need:
#!/bin/bash
# python virtual environment activation
source /home/netman/lightrag-xyj/venv/bin/activate
# start lightrag api server
lightrag-serverInstall lightrag.service in Linux. Sample commands in Ubuntu server look like:
sudo cp lightrag-server.service /etc/systemd/system/
sudo systemctl daemon-reload
sudo systemctl start lightrag-server.service
sudo systemctl status lightrag-server.service
sudo systemctl enable lightrag-server.service