Data analysis by committee, now in Python

What it does

BambooAI is a Python library that lets you interrogate dataframes and external sources in plain English. It routes your question through a pipeline of specialized agents—planner, code generator, error corrector, reviewer—that write and execute Python code, returning results or visualizations. It works in Jupyter notebooks or as a Flask web app, and can pull in internet search results or auxiliary datasets when the main table isn’t enough.

The interesting bit

Instead of trusting one prompt to do everything, BambooAI treats analysis like an assembly line where each station can run a different LLM model assigned via LLM_CONFIG.json. It also caches successful solution chains in a Pinecone vector database as “episodic memory” and supports custom OWL ontologies for semantic memory—more architecture than most notebook helpers attempt.

Key highlights

• Multi-agent pipeline with role-specific agents including Expert Selector, Planner, Code Generator, Error Corrector, and Reviewer
• Per-agent LLM configuration across OpenAI, Anthropic, Gemini, Groq, Mistral, Ollama, and vLLM
• Self-healing execution: generated code is run, and errors are fed back to an LLM for correction with retries
• Optional Pinecone vector database to store high-quality solutions for semantic search
• Optional OWL/TTL ontology support to improve dataframe understanding
• Web UI and Jupyter interfaces; can fetch external data or search the internet if no local dataframe is provided

Caveats

• The README explicitly labels the project as experimental.
• A single query can trigger multiple LLM calls across different agents, so token usage can stack up quickly.
• The broad surface area of agents, vector databases, ontologies, and external search implies a complex configuration landscape.

Verdict

Data analysts who want an open-source, conversational layer over pandas will find the agent routing genuinely useful; if you just need a quick LLM wrapper for one-off CSV questions, this is overkill.