GPTStonks Chatbot API

Welcome to the GPTStonks Chatbot API documentation! This API allows you to interact with a powerful financial chatbot built on top of the openbb framework. Whether you're a developer looking to integrate financial chat capabilities into your application or a trader seeking automated financial insights, this API is designed to provide you with a seamless and customizable experience.

Table of Contents

• GPTStonks Chatbot API
  • Table of Contents
  • Introduction 🌟
  • Features 🚀
  • Supported LLM Providers
  • Supported Embeddings Providers
  • Getting Started 🛠️
    • Prerequisites
    • Full deployment [api + frontend + db] (recommended)
    • Installation 🛸
  • For Production Environments 🏭
  • Usage💡
  • API Endpoints🌐
  • Configuration with environment variables ⚙️
  • Contributing 🤝
  • License 📃
  • Disclaimer

Introduction 🌟

GPTStonks is a financial chatbot powered by LLMs and enhanced with the openbb framework. It provides natural language conversation capabilities for financial topics, making it an ideal choice for a wide range of financial applications, including:

• Learning about the financial markets
• Improving trading strategies
• Financial news analysis: sentiment, trends, etc.
• Customer support for financial institutions

This API allows you to integrate the GPTStonks financial chatbot into your projects, enabling real-time financial chat interactions with users.

Features 🚀

• Real-time Financial Chat: Engage in natural language conversations about financial topics.
• Customizable Responses: Tailor the chatbot's responses to suit your specific use case.
• Easy Integration: Built on FastAPI, this API is designed for straightforward integration into your application or platform.
• Extensive Documentation: Detailed documentation and examples to help you get started quickly.

Supported LLM Providers

• Llama.cpp: optimized implementations of the most popular open source LLMs for inference over CPU and GPU. See their docs for more details on supported models, which include Mixtral, Llama 2 and Zephyr among others. Many quantized models (GGUF) can be found in Hugging Face under the user TheBloke.
• Amazon Bedrock: foundation models from a variety of providers, including Anthropic and Amazon.
• OpenAI: GPT family of foundation models. For now only instruct versions are supported, such as gpt-3.5-turbo-instruct. Chat versions will be added soon.
• Vertex AI: similar to Amazon Bedrock but provided by Google. This integration is in alpha version, not recommended for now.

Supported Embeddings Providers

• OpenAI Embeddings: includes models such as Ada 2.
• Hugging Face: including providers such as BAAI, see their general embedding (BGE) model list.

Getting Started 🛠️

Prerequisites

Before you begin, make sure you have Docker installed on your system.

Full deployment [api + frontend + db] (recommended)

1. Clone this repository to your local machine:

git clone https://github.com/gptstonks/api.git

2. Check the .env.template file for the required environment variables. Modify the values as desired and save the file.

3. Run the docker-compose file:

docker-compose up

4. If you didn't change any port configuration, navigate to http://localhost:3000 to access the frontend.

5. (optional) If you want to use your OpenBB PAT to access resources, you can set it from the API Keys section (sidebar) in the frontend.

Installation 🛸

1. Set up environment variables by creating a .env file in the project directory with the contents specified in .env.template.

2. [Highly Recommended] Option 1: use the latest Docker image in ghcr.io.

docker run -it -p 8000:8000 --env-file .env ghcr.io/gptstonks/api:main

• Option 2: build from source.

  1. Install PDM.

  2. Clone this repository to your local machine:

  git clone https://github.com/GPTStonks/api.git
  

  2. Navigate to the project directory:

  cd projects/gptstonks_api
  

  3. Install the required dependencies:

  pdm install --no-editable --no-self
  

  4. Create openssl.cnf to allow legacy TLS renegotiation, needed for OECD data in OpenBB.

  echo 'openssl_conf = openssl_init\n\
  \n\
  [openssl_init]\n\
  ssl_conf = ssl_sect\n\
  \n\
  [ssl_sect]\n\
  system_default = system_default_sect\n\
  \n\
  [system_default_sect]\n\
  Options = UnsafeLegacyRenegotiation'\
  > openssl.cnf
  

  5. Start the API:

  uvicorn gptstonks.api.main:app --host 0.0.0.0 --port 8000
  

Now your GPTStonks Financial Chatbot API is up and running!

For Production Environments 🏭

For production environments, additional steps are necessary to ensure security and stability:

Ensure that uvicorn is configured with SSL certificates for secure HTTPS communication.

Build the Docker image from source:

docker build -t gptstonks-api:v0.1_pro -f Dockerfile.pro .

Now you can run the Docker image with the following command:

docker run -it -p 443:8000 \
-v /etc/letsencrypt/live/api.gptstonks.net/fullchain.pem:/api/cert.pem \
-v /etc/letsencrypt/live/api.gptstonks.net/privkey.pem:/api/key.pem \
--env-file .env gptstonks-api:v0.1_pro

Usage💡

To use the GPTStonks Financial Chatbot API, send HTTP requests to the provided endpoints. You can interact with the chatbot by sending messages and receiving responses in real-time.

API Endpoints🌐

Check http://localhost:8000/docs once the API is started to access the endpoints' documentation.

Configuration with environment variables ⚙️

Env variable	Required	Default	Description
MONGO_URI	Yes	-	MongoDB's URI to connect.
MONGO_DBNAME	Yes	-	MongoDB's database to use.
AUTOLLAMAINDEX_VSI_GDRIVE_URI	No	None (Not downloaded)	Google Drive's URL to download Vector Store Index (VSI).
AUTOLLAMAINDEX_EMBEDDING_MODEL_ID	No	"local:BAAI/bge-large-en-v1.5"	Embedding model ID to use with AutoLlamaIndex (must match with VSI).
AUTOLLAMAINDEX_SIMILARITY_POSTPROCESSOR_CUTOFF	No	0.5	Minimum similarity required when retrieving similar documents.
AUTOLLAMAINDEX_REMOVE_METADATA_POSTPROCESSOR	No	None (Postprocessor used)	Whether or not to use a metadata postprocessor.
AUTOLLAMAINDEX_VSI_PATH	Yes	-	Path to the downloaded VSI. If AUTOLLAMAINDEX_VSI_GDRIVE_URI is given, they will match automatically.
AUTOLLAMAINDEX_LLM_CONTEXT_WINDOW	No	4096	Context window to use when a Hugging Face model is loaded in AutoLlamaIndex.
AUTOLLAMAINDEX_QA_TEMPLATE	No	None (LlamaIndex's Default QA Template)	Template to use with LlamaIndex question-answering step.
AUTOLLAMAINDEX_REFINE_TEMPLATE	No	None (LlamaIndex's Default Refine Template)	Template to use with AutoLlamaIndex refine step.
AUTOLLAMAINDEX_VIR_SIMILARITY_TOP_K	No	3	K most similar elements are retrieved with vector search.
AUTOLLAMAINDEX_RETRIEVER_TYPE	No	None (Hybrid retrieved used)	Whether or not to use BM25 with vector search (hybrid) or only vector search.
AUTOMULTISTEPQUERYENGINE_QA_TEMPLATE	No	None (LlamaIndex's Default QA Template)	Template to use with AutoMultiStepQueryEngine question-answering step.
AUTOMULTISTEPQUERYENGINE_REFINE_TEMPLATE	No	None (LlamaIndex's Default Refine Template)	Template to use with AutoMultiStepQueryEngine refine step.
AUTOMULTISTEPQUERYENGINE_STEPDECOMPOSE_QUERY_PROMPT	No	None (LlamaIndex's Default Step Decompose Template)	Template to use with AutoMultiStepQueryEngine step decompose.
AUTOMULTISTEPQUERYENGINE_INDEX_SUMMARY	No	"Useful to search information on the Internet."	The index summary is used by the multi-step agent to understand its own capabilities and formulate new questions.
AGENT_REQUEST_TIMEOUT	No	20	No. seconds to wait before timeout when an API LLM is used (e.g., OpenAI).
AGENT_EARLY_STOPPING_METHOD	No	"force"	How the model should return its final output when early stopping is applied.
LLM_TEMPERATURE	No	0.1	Temperature to use when sampling.
LLM_MAX_TOKENS	No	256	No. max. tokens to sample.
LLM_TOP_P	No	1.0	Top-p parameter to apply when sampling.
LLM_MODEL_ID	Yes	-	ID of the model to use. See tutorials in GPTStonks' blog for more details.
LLM_CHAT_MODEL_SYSTEM_MESSAGE	No	"You write concise and complete answers."	System message when using chat models (e.g., GPT-4).
LLM_VERTEXAI_CLOUD_LOCATION	No	None	Google Cloud location to use with VertexAI.
LLM_LLAMACPP_CONTEXT_WINDOW	No	4000	Context window to apply when a model is loaded with Llama.cpp.
LLM_HF_DEVICE	No	-1	Device to use with Hugging Face's models.
LLM_HF_DISABLE_SAMPLING	No	False	Whether or not to disable sampling when generating with Hugging Face's models.
LLM_HF_DEVICE_MAP	No	None (HuggingFacePipeline default)	Device map to use with Hugging Face's models.
LLM_HF_BITS	No	4	No. bits of quantized Hugging Face's model.
LLM_HF_DISABLE_EXLLAMA	No	False	Whether or not to disable ExLlama with Hugging Face's models.
LLM_HF_TRUST_REMOTE_CODE	No	False	Whether or not to trust remote code with Hugging Face's models.
OPENBBCHAT_TOOL_DESCRIPTION	Yes	-	OpenBB Platform's tool description for the LLM agent.
SEARCH_TOOL_DESCRIPTION	No	None (Default DDG Search description)	DDG's search tool description for the LLM agent.
WIKIPEDIA_TOOL_DESCRIPTION	No	None (Default Wikipedia description)	Wikipedia tool description for the LLM agent.
CUSTOM_GPTSTONKS_PREFIX	No	None (Default LangChain agent prefix)	Prefix to use with LLM agent.

Contributing 🤝

We welcome contributions from the community! If you have any suggestions, bug reports, or want to contribute to the project, feel free to open issues or propose changes.

License 📃

This project is licensed under the MIT License - see the LICENSE file for details.

Disclaimer

GPTStonks Chat serves as an interface for accessing financial data and general knowledge. It is not intended to provide financial or investment advice.