BambooAI Documentation

BambooAI supports integrating with Pinecone or Qdrant to store and retrieve successful analyses. When a user ranks a solution highly (>6), the intent, plan, code, and metadata are vectorized and stored. For future similar tasks, the system queries the vector index (using a similarity threshold of 0.8) to guide agents using previous successful solutions.

To enable this, set vector_db=True in the BambooAI constructor and configure the appropriate environment variables.

from bambooai import BambooAI
import pandas as pd

# Initialize with vector database enabled
bamboo = BambooAI(
    df=your_dataframe,
    vector_db=True
)

To get started with a basic data analysis session, follow these steps:

1. Install the package: pip install bambooai
2. Configure environment variables: Copy the example environment file and update it with your credentials.

   cp .env.example .env

3. Configure agents and models: Copy the sample configuration file to define your preferred LLM combinations.

   cp LLM_CONFIG_sample.json LLM_CONFIG.json

4. Run a session: Use the BambooAI class to initialize an instance with your dataframe and start a conversation.

import pandas as pd
from bambooai import BambooAI

import plotly.io as pio
pio.renderers.default = 'jupyterlab'

df = pd.read_csv('titanic.csv')
bamboo = BambooAI(df=df, planning=True, vector_db=False, search_tool=True)
bamboo.pd_agent_converse()

To run the Web UI using the pip package:

1. Install the package:

   pip install bambooai

2. Download the web_app folder from the repository.
3. Configure the environment by copying .env.example to .env in the web_app directory.
4. Configure LLM agents in web_app/LLM_CONFIG.json by copying LLM_CONFIG_sample.json and editing the agent_configs array.
5. Run the app:

   cd <path_to_web_app>
   python app.py

The interface is available at http://localhost:5000.

You can install BambooAI using pip or by cloning the repository and installing the requirements manually.

Using pip:

pip install bambooai

Using the repository source:

git clone https://github.com/pgalko/BambooAI.git
pip install -r requirements.txt

Depending on your chosen provider, set the following in your .env file:

For Pinecone:

VECTOR_DB_TYPE=pinecone
PINECONE_API_KEY=<YOUR API KEY HERE>
PINECONE_CLOUD=aws
PINECONE_REGION=us-east-1

For Qdrant:

VECTOR_DB_TYPE=qdrant
QDRANT_URL=http://localhost:6333  # For local Qdrant
QDRANT_API_KEY=<YOUR API KEY HERE>  # Optional for local, required for cloud

The Web Application requires an LLM_CONFIG.json file to define which models and providers each agent uses. If this is missing or incomplete, execution will fail.

Example structure for the agent_configs array:

{
   "agent_configs": [
      {
         "agent": "Code Generator",
         "details": {
            "model": "your-preferred-model",
            "provider": "provider-name",
            "max_tokens": 4000,
            "temperature": 0
         }
      }
   ]
}

BambooAI supports loading longitudinal sports and health data from SweatStack.

1. Authorize: Navigate to /sweatstack/authorize to begin the OAuth flow.
2. Get Users: GET /sweatstack/get_users returns a list of accessible athletes/users.
3. Load Data: POST /sweatstack/load_data fetches and combines data for selected users.
   • Request Body:
     • sports: List of sports to include.
     • metrics: List of metrics to include.
     • users: List of user IDs.
     • start_date / end_date: ISO format strings (YYYY-MM-DD).
4. Remove Data: POST /sweatstack/remove_data clears the primary dataset and resets the BambooAI instance while preserving auxiliary datasets.

You can use BambooAI in two primary ways within a Jupyter Notebook or CLI:

1. Interactive Mode: Uses pd_agent_converse() to start a conversational session.
2. Single Query Mode: Passes a specific string to pd_agent_converse() to execute a one-off task.

import pandas as pd
from bambooai import BambooAI

# Setup
df = pd.read_csv('training_activity_data.csv')
bamboo = BambooAI(df=df, search_tool=True, planning=True)

# Option 1: Interactive Mode
bamboo.pd_agent_converse()

# Option 2: Single Query Mode
bamboo.pd_agent_converse("Calculate 30, 50, 75 and 90 percentiles of the heart rate column")

You can use local or alternative LLM providers by specifying the appropriate provider in the agent_configs section of LLM_CONFIG.json.

Ollama Example:

{
  "agent": "Planner",
  "details": {
    "model": "llama3:70b",
    "provider": "ollama",
    "max_tokens": 2000,
    "temperature": 0
  }
}

VLLM Example:

{
  "agent": "Code Generator",
  "details": {
    "model": "/path/to/model/DeepSeek-R1-Distill-14B",
    "provider": "vllm",
    "max_tokens": 2000,
    "temperature": 0
  }
}

MiniMax Example:

{
  "agent": "Code Generator",
  "details": {
    "model": "MiniMax-M3",
    "provider": "minimax",
    "max_tokens": 8000,
    "temperature": 0.1
  }
}