OnboardBot

OnboardBot allows you to direct Chatbot conversations towards collecting data defined by flexible YAML files.

demo.mov

For example, the following YAML will direct the Chatbot to collect Name, email and the user's desired property type:

models:
  - name: Buyer
    class_type: Question
    fields:
      - name: name
      - name: email_address

  - name: PropertyStyle
    class_type: Choice
    description: What is your desired property type?
    fields:
      - name: townhouse
      - name: condo
      - name: single_family
      - name: multi_family
      - name: land
      - name: other

Import Notes

1. Currently , your models config MUST start with a Question model type. Hoping to fix this soon.

Install

1. git clone [email protected]:kvnn/OnboardBot.git
2. cd OnboardBot/src/server
3. (optional) python3 -m venv .venv && source .venv/bin/activate)
4. pip install -r requirements.txt
5. Create an .env file in the server directory. It will require an OPENROUTER_API_KEY OR an OPENAI_API_KEY .

CHAINLIT_AUTH_SECRET="YOUR SECRET KEY"
OPENROUTER_API_KEY="YOUR OPENROUTER_API_KEY"
# Optional
DB_ENGINE_URL="postgresql://postgres:{password}@{host}:5432/postgres"

Data collection

DB_ENGINE_URL is optional, but without it the data retrieved from the chats will not be readily available to you. Its also trivial to send the data to a Slack channel, for example. Built-in support for this is coming soon. Here is an example of an output:

buyer:
  email_address: [email protected]
  name: Kevin
desiredproperty:
  number_of_bathrooms: 2.5
  number_of_bedrooms: 3
multifamilydetails:
  business headquarters: false
  grandparents quarters: true
  off-grid compound: false
  vacation rental: false
musthave:
  must_haves: green gardens , good aesthetics
propertystyle:
  condo: false
  land: false
  multi_family: true
  other: false
  single_family: false
  townhouse: false

Config.yml

1. you can modify config.yml to point to a new model definition to fit your use case
2. for now, it will be best to copy bots/realty.yml , and experiment with modifications
3. there are three model types:
   1. Question : an open-ended question . The user's text answer is saved to the model instance value.
   2. Choice : a single choice . The user's choice is saved to the model instance value.
   3. MultiChoice : coming soon. its working in my fork of chainlit , and the PR should be merged soon . See Chainlit/chainlit#965
4. there is conditions logic. See MultiFamilyDetails in bots/realty.yml. for_choice is a pointer to another model. A model with this config will show only if that model has a value matched in the for_value key of the config. So, You can show / hide questions based on the values of previous questions.

Server

The OnboardBot server is a simple, opinionated and flexible Chainlit project. So you run it like chainlit run app.py from OnboardBot/src/server. This will open a browser tab running the chatbot interface.

OnboardBot (via prompts.py) will use the data model given in config.yml to collect data from the user in a conversational, helpful manner.

Notice that conditional logic, in the case of notification_preference_sms_email_whatsapp_or_combination, is handled in the field name itself. The LLMs (including Mixtral-8x7B-Instruct which is 100x cheaper than gpt-4) handle this exactly how we'd wish. The aim is to push the simplicity as far as possible before implementing logic chains in the models.

Single choice, Multiple choice and conditional logic is supported.

UI

OnboardBot does not use a custom client. It uses the default chainlit UI. Everything OnboardBot wishes to achieve is done via chat.

Support

create an issue or email [email protected]

TODO

• Explain and give good examples of Choice , MultiChoice and conditionals
• Support OpenAI (you'll need to modify llm.py to do this for now. Just remove the base_url property from llm_chat_client_async and change the default models to an OpenAI model.
• Support duplicate key names in data classes (the LLM needs to understand they are unique, even with same key)
• Decide: should User and Goal may need to be first-class models, and should prompt be different for definitions of data models versus class that inherit from them? there is a tricky truth here that we are defining pydantic models for LLM's understanding, when we are conditioned to write data models to make our code more meaningful. So, we may need utility classes (or first-class classes?) that support enabled_models that are intended for the LLMs understanding.