Skip to content

Latest commit

 

History

History
185 lines (163 loc) · 4.61 KB

README.md

File metadata and controls

185 lines (163 loc) · 4.61 KB
Raw

Super-Rag

Super-performant RAG pipeline for AI apps.


Features • Installation • How to use • Cloud API

✅ Key features

  • Supports multiple document formats and vector databases.
  • Provides a production ready REST API.
  • Customizable splitting/chunking.
  • Includes options for encoding data using different encoding models both propriatory and open source.
  • Built in code interpreter mode for computational question & answer scenarios.
  • Allows session management through unique IDs for caching purposes.

☁️ Cloud API

Easiest way to get started is to use our Cloud API. This API is free to use (within reasonable limits).

📦 Installation

  1. Clone the repository

    git clone https://github.com/superagent-ai/super-rag 
cd super-rag

  2. Setup virtual environment

    # Using virtualenv 
virtualenv env 
source env/bin/activate 

# Or using venv 
python3 -m venv env 
source env/bin/activate

  3. Install requried packages

    poetry install

  4. Rename .env.example to .env and set your environment variables

  5. Run server

    uvicorn main:app --reload

🤖 Interpreter mode

Super-Rag has built in support for running computational Q&A using code interpreters powered by E2B.dev custom runtimes. You can signup to receive an API key to leverage they sandboxes in a cloud environment or setup your own by following these instructions.

🚀 How to use

Super-Rag comes with a built in REST API powered by FastApi.

Ingest documents

// POST: /api/v1/ingest

// Payload
{
    "files": [
        {
            "name": "My file", // Optional
            "url": "https://path-to-my-file.pdf",
            "metadata": { // Optional
                "key": "value"
            }
        }
    ],
    "document_processor": { // Optional
        "encoder": {
            "dimensions": 384,
            "model_name": "embed-multilingual-light-v3.0",
            "provider": "cohere"
        },
        "unstructured": {
            "hi_res_model_name": "detectron2_onnx",
            "partition_strategy": "auto",
            "process_tables": false
        },
        "splitter": {
            "max_tokens": 400,
            "min_tokens": 30,
            "name": "semantic",
            "prefix_summary": true,
            "prefix_title": true,
            "rolling_window_size": 1
        }
    },
    "vector_database": {
        "type": "qdrant",
        "config": {
            "api_key": "YOUR API KEY",
            "host": "THE QDRANT HOST"
        }
    },
    "index_name": "my_index",
    "webhook_url": "https://my-webhook-url"
}

Query documents

// POST: /api/v1/query

// Payload
{
    "input": "What is ReAct",
    "vector_database": {
            "type": "qdrant",
            "config": {
            "api_key": "YOUR API KEY",
            "host": "THE QDRANT HOST"
        }
        },
    "index_name": "YOUR INDEX",
    "interpreter_mode": true,
    "encoder": {
        "provider": "cohere",
        "name": "embed-multilingual-light-v3.0",
        "dimensions": 384
    },
    "filter": { // Optional
        // Each provider has its own filter schema,
        // for Qdrant it looks like the following:
        "must": [
            {
                "key": "key",
                "match": {
                    "value": "value"
                }
            }
        ]
    },
    "exclude_fields": ["metadata"], // Exclude specific fields
    "interpreter_mode": False, // Set to True if you wish to run computation Q&A with a code interpreter
    "session_id": "my_session_id" // keeps micro-vm sessions and enables caching 
}

Delete document

// POST: /api/v1/delete

// Payload
{
    "file_url": "A file url to delete",
    "vector_database": {
        "type": "qdrant",
        "config": {
            "api_key": "YOUR API KEY",
            "host": "THE QDRANT HOST"
        }
    },
    "index_name": "my_index",
}

🧠 Supportd encoders

  • OpenAi
  • Cohere
  • HuggingFace
  • FastEmbed
  • Mistral (coming soon)
  • Anthropic (coming soon)

🗃 Supported vector databases

  • Pinecone
  • Qdrant
  • Weaviate
  • Astra
  • PGVector
  • Chroma (coming soon)