This repo is an implementation of a context-aware AI Agent for Sales using LLMs and can work across voice, email and texting (SMS, WhatsApp, WeChat, Weibo, Telegram, etc.).
SalesGPT is context-aware, which means it can understand what stage of a sales conversation it is in and act accordingly. Morever, SalesGPT has access to tools, such as your own pre-defined product knowledge base, significantly reducing hallucinations.
We are building SalesGPT to power your best AI Sales Agents. Hence, we would love to learn more about use cases you are building towards which will fuel SalesGPT development roadmap, so please don't hesitate to reach out.
If you need assistance, our team is here to help! Please reach out to us at Odyssey AI Partners for expert support and guidance.
- The AI Agent can access data from any data system (new, old or legacy) via an integration to Mindware's AI Solutions Engineer.
- AI Sales Agents can now ACTUALLY sell! They autonomously generate Stripe payment links to sell products and services to customers.
- You can now test your AI Sales Agents via our frontend.
- Sales Agent can now take advantage of tools, such as look up products in a product catalog!
Unload AI Sales Agent Demos - Powered by SalesGPT: Our new virtual workforce? 🤔
Demo #1: Sarah - Patient Coordinator at South Orange Pediatrics
- 100X Your Healthcare Admin with our Virtual Workforce
Demo #2: Ted - Autonomously create payment links and collect customer payments
Try it yourself on our SalesGPT hosted frontend demo.
We are building SalesGPT to power your best AI Sales Agents. Hence, we would love to learn more about use cases you are building towards which will fuel SalesGPT development roadmap.
If you want us to build better towards your needs, or need help with your AI Sales Agents, please reach out to chat with us: SalesGPT Use Case Intake Survey
The AI Sales Agent understands the conversation stage (you can define your own stages fitting your needs):
- Introduction: Start the conversation by introducing yourself and your company.
- Qualification: Qualify the prospect by confirming if they are the right person to talk to regarding your product/service.
- Value Proposition: Briefly explain how your product/service can benefit the prospect.
- Needs Analysis: Ask open-ended questions to uncover the prospect's needs and pain points.
- Solution Presentation: Based on the prospect's needs, present your product/service as the solution that can address their pain points.
- Objection Handling: Address any objections that the prospect may have regarding your product/service.
- Close: Ask for the sale by proposing a next step.
- End Conversation: The user does not want to continue the conversation, so end the call.
- Reference only your business information & products and significantly reduce hallucinations!
- The AI Agent can actually close sales by generating Stripe payment link and closing orders from customers.
- The AI Agent can leverage data from any data system (new, old or legacy) via an integration to Mindware.
- Enhance your sales process with automated email communication. SalesGPT can now send personalized emails to prospects, including follow-ups or product information.
- The AI Agent can now facilitate scheduling meetings by generating Calendly links.
- Thanks to our integration with LiteLLM, you can choose any closed/open-sourced LLM to work with SalesGPT! Thanks to LiteLLM maintainers for this contribution!
- Synchronous & Asynchronous Completion with LLMs
- Synchronous & Asynchronous Streaming from LLMs
- Voice AI Sales Agent boasts <1s round trip response rate to human speakers which includes the entire pipeline - speech to text, LLM inference, and text to speech - while ensuring stability and scalability.
- For use cases where AI sales agent needs human supervision.
- debug, test, evaluate, and monitor chains and intelligent agents built on any LLM framework
- Upcoming integration with PromptArmor to protect your AI Sales Agents against security vulnerabilities (see our roadmap).
import os
from salesgpt.agents import SalesGPT
from langchain_community.chat_models import ChatLiteLLM
from dotenv import load_dotenv
load_dotenv() # make sure you have .env file with your API keys, eg., OPENAI_API_KEY=sk-xxx, MINDWARE_API_KEY etc.
# select your model - we support 50 LLMs via LiteLLM https://docs.litellm.ai/docs/providers
llm = ChatLiteLLM(temperature=0.4, model_name="gpt-4-0125-preview")
sales_agent = SalesGPT.from_llm(llm, use_tools=True, verbose=False,
product_catalog = "examples/sample_product_catalog.txt",
salesperson_name="Ted Lasso",
salesperson_role="Sales Representative",
company_name="Sleep Haven",
company_business='''Sleep Haven
is a premium mattress company that provides
customers with the most comfortable and
supportive sleeping experience possible.
We offer a range of high-quality mattresses,
pillows, and bedding accessories
that are designed to meet the unique
needs of our customers.'''
)
sales_agent.seed_agent()
sales_agent.determine_conversation_stage() # optional for demonstration, built into the prompt
# agent
sales_agent.step()
# user
user_input = input('Your response: ') # Yea, sure
sales_agent.human_step(user_input)
# agent
sales_agent.determine_conversation_stage() # optional for demonstration, built into the prompt
sales_agent.step()
# user
user_input = input('Your response: ') # What pricing do you have for your mattresses?
sales_agent.human_step(user_input)
# agent
sales_agent.determine_conversation_stage() # optional for demonstration, built into the prompt
sales_agent.step()
Conversation Stage: Introduction: Start the conversation by introducing yourself and your company.
Thought: Do I need to use a tool? No
Ted Lasso: Hello, my name is Ted Lasso and I'm calling on behalf of Sleep Haven. We are a premium mattress company that provides customers with the most comfortable and supportive sleeping experience possible. I was wondering if you would be interested in learning more about our products and how they can improve your sleep?
User: Yea, sure
Conversation Stage: Value proposition: Briefly explain how your product/service can benefit the prospect.
Thought: Do I need to use a tool? No
Ted Lasso: Great to hear that! Our mattresses are specially designed to contour to your body shape, providing the perfect level of support and comfort for a better night's sleep. Plus, they're made with high-quality materials that are built to last. Would you like to hear more about our different mattress options?
User: What pricing do you have for your mattresses?
Thought: Do I need to use a tool? Yes --> AI Agent uses tools here.
Action: ProductSearch
ActionInput pricing for mattresses
Observation: Sleep Haven offers three mattresses: the Classic Harmony Spring Mattress for $1,299, the Plush Serenity Bamboo Mattress for $2,599, and the Luxury Cloud-Comfort Memory Foam Mattress for $999. The sizes available vary by mattress.
Thought: Do I need to use a tool? No
Ted Lasso: Sleep Haven offers three mattresses at different price points. The Classic Harmony Spring Mattress is priced at $1,299, the Plush Serenity Bamboo Mattress is priced at $2,599, and the Luxury Cloud-Comfort Memory Foam Mattress is priced at $999. The prices may vary depending on the size you choose. Would you like more information about the specific sizes and features of each mattress?
User: No, I will take two Classic Harmony Spring Mattress please.
Thought: Do I need to use a tool? Yes --> AI Agent uses tools here.
Action: GeneratePaymentLink
Action Input: Classic Harmony Spring Mattress, 2, John Doe Observation: {"response":"https://buy.stripe.com/test_9AQ6rygc397Bbv228p"}
Thought: Do I need to use a tool? No
Ted Lasso: Fantastic choice! I've generated a payment link for you to order two Classic Harmony Spring Mattresses. You can complete your purchase by clicking on this link: https://buy.stripe.com/test_9AQ6rygc397Bbv228p. Is there anything else I can assist you with today?
Make sure you have a python >=3.8,<3.12:
Create a virtual environment at a location on your computer. We use the generic "env" name for our virtual environment in the setup. You can rename this, but make sure to then use this name later when working with the environment (also rename the VENV variable in the Makefile accordingly to be able to use make commands successfully after cloning our repository):
- Open Command Prompt or PowerShell.
- Navigate to your project directory:
cd path\to\your\project
- Create a virtual environment:
python -m venv env
- Activate the virtual environment:
.\env\Scripts\activate
- Open Terminal.
- Navigate to your project directory:
cd path/to/your/project
- Create a virtual environment:
python3 -m venv env
- Activate the virtual environment:
source env/bin/activate
To deactivate a virtual environment after you have stopped using it simply run: deactivate
Clone the SalesGPT Github repository:
git clone https://github.com/filip-michalsky/SalesGPT.git
Navigate to the repository and in case you used a different venv name rename the VENV variable in the Makefile:
cd SalesGPT
If you simply want to work with SalesGPT as an end user without local changes you can install from PyPI using:
pip install salesgpt
If you want to work on your own version of SalesGPT or contribute to our open-source version install by activating your virtual environment as aforementioned and then run:
make setup
For more detailed installation steps along with the reasons for doing each please visit CONTRIBUTING.md
Finally, for use of SalesGPT create an .env
file just as our .env.example
and put your API keys there by specifying a new line just as we have done.
Navigate into the SalesGPT directory:
cd SalesGPT
SalesGPT can be run in various ways, tailored to your preferences and setup. Below are the methods available:
For those who prefer containerization, Docker offers an isolated and consistent environment. Ensure Docker is installed on your system by following the official Docker installation guide.
To run SalesGPT with Docker, execute the following steps:
-
Start the Application with Docker Compose:
Use the command below to start SalesGPT in detached mode:
docker-compose up -d
If you've made changes and want them to reflect, append
--build
to the command above. -
Stopping the Application:
To stop and remove all running containers related to SalesGPT, execute:
docker-compose down
Troubleshooting:
- Clean Up Docker Resources: If you encounter errors, you can clean up Docker by removing all unused containers, networks, images, and volumes with caution:
docker system prune --volumes
- Rebuild Without Cache: To rebuild and start the services afresh without using cache, run:
docker-compose up -d --build --no-cache
After successful setup, access SalesGPT at localhost:3000/chat in your browser.
If Docker is not part of your workflow, you can directly launch the SalesGPT user interface. Please refer to the README.md
file in the frontend directory for instructions on setting up the UI locally.
For terminal enthusiasts or automation scripts, run SalesGPT with the following command:
python run.py --verbose True --config examples/example_agent_setup.json
For those who wish to integrate SalesGPT's backend with their own user interface or application, running only the backend is a straightforward process. This allows you to leverage the powerful features of SalesGPT while maintaining full control over the user experience.
To run only the backend of SalesGPT, follow these steps:
-
Start the Backend Service:
Use the following command to start the backend service. This will initiate the server on port 8000 by default, making the API accessible:
docker-compose up -d backend
If you need to rebuild the backend image, perhaps after making changes, you can add
--build
to the command above. -
Accessing the Backend:
With the backend running, you can access the API endpoints at
http://localhost:8000
. Refer to the API documentation for details on available endpoints and their usage. -
Stopping the Backend:
To stop the backend service, execute:
docker-compose stop backend
If you wish to remove the backend container entirely, use:
docker-compose down
This setup is ideal for developers looking to integrate SalesGPT's backend into custom applications or those who prefer to use a different frontend technology.
- Activate your environment as described above. (run
source env/bin/activate
on Unix-like systems and.\env\Scripts\activate
on Windows. Replace env with the name of your virtual environment) - cd
SalesGPT
If you haven't already navigated to the SalesGPT home directory make test
All tests should pass. Warnings can be ignored.
Regular updates are added to the SalesGPT repository. If your code is not working please check the current README file for possible changes in launch. With updates often extra environment variables are added, which are however required for certain functionality. Please check current .env.example file.
For any problems you can not solve please don't hesitate to reach out to our discord #support channel.
To delete the virtual environment you used for SalesGPT programming and your SalesGPT repository from your system navigate to the directory where you installed your virtual environment and cloned SalesGPT and run:
make clean
We leverage the langchain
library in this implementation, specifically Custom Agent Configuration and are inspired by BabyAGI architecture.
LangSmith is a platform for building production-grade LLM applications.
It lets you debug, test, evaluate, and monitor chains and intelligent agents built on any LLM framework and seamlessly integrates with LangChain, the go-to open source framework for building with LLMs.
LangSmith is developed by LangChain, the company behind the open source LangChain framework.
To switch on the LangSmith tracing you have to do the following steps:
- Create a LangSmith account
- Create an API key in settings
- Add you API key and Project name from LangSmith account to .env file or run.py module
- Switch on the "LANGCHAIN_TRACING_V2" setting in run.py to "true"
- That's it. You'll get better understanding of your agents and chaing performance in LangChain admin panel.
For futher reading take a look at the docs
- Improve SalesGPT observability
- Enhance prompt versioning
- Add prompt evaluation
- Write more documentation
- Better code documentation
- Refactor
Improve reliability of the parser issue here and here- Improve Deployment Instructions
- Calling Functionality - sample code
- Enterprise-Grade Security - integration with PromptArmor to protect your AI Sales Agents against security vulnerabilities
- LLM evaluations
- Resolve tickets and PRs (ongoing)
- Add example implementation of OpenAI functions agentissue here
- Add support for multiple tools issue here
- Add an agent controller for when stages need to be traversed linearly without skips issue here
- Add
tool_getter
to choose a tool based on vector distance to the tasks needed to be done - What tools should the agent have? (e.g., the ability to search the internet)
- Add the ability of Sales Agent to interact with AI plugins on your website (.well-known/ai-plugin.json)
- More SalesGPT examples
Lead Maintaner: Filip Michalsky
- Contact Email
- Follow us on X at @FilipMichalsky
Our Support Team:
- AI Engineering: Honza Michna (LinkedIn)
- Software Engineer: Illia Makovoz
Contributions are highly encouraged! Please fork and submit a PR.