Repolode
Submit
pgalko

pgalko/BambooAI

AI data analyst that works step by step in a persistent Python kernel, keeps every computed result traceable, and can run on local open-weight models.

View on GitHub
Stars
838
Forks
94
Open issues
18
Last commit
7 days ago

What is it?

BambooAI is an LLM-driven data analysis environment. A primary Analyst model works through a question turn by turn, choosing each time to run a Python cell, inspect earlier output, search or read supporting documents, ask a clarifying question, or write the report. Executed cells stay in a persistent workspace, so later steps and follow-up questions build on earlier computation.

The project focuses on keeping the analysis inspectable. Executed code, outputs, figures, intermediate results and model calls remain available, and numbers in the report are linked to the cells that produced them. After a run, the relevant cells are replayed in a fresh kernel against the original data to check that the cited numbers reproduce. In Adaptive mode, a separate Reviewer model can request further tests or narrow a conclusion.

BambooAI supports OpenRouter, OpenAI, Anthropic, Google, Groq, Mistral, xAI and DeepSeek, as well as locally served models through Ollama and vLLM. Model-generated code runs in a Docker executor container by default. The project began with sports-science data and includes Intervals.icu, Endura and SweatStack integrations, but it works with ordinary CSV, Parquet, JSON and XLSX files.

Key features

  • 01Turn-by-turn analysis in a persistent Python workspace with transactional cell execution.
  • 02Reports whose numerical results link back to the cells that computed them.
  • 03Replay of cited computation in a fresh kernel to check reproducibility.
  • 04Separate Reviewer model in Adaptive mode to challenge the analysis.
  • 05Support for many hosted providers plus local models via Ollama and vLLM.
  • 06Document parsing for PDF, Word, Markdown and text attachments.

Use cases

  • Exploratory analysis of CSV or Excel datasets where the code behind each result must be auditable.
  • Running analyses on sensitive data with locally hosted open-weight models.
  • Follow-up analysis of a dataset over several sessions, reusing earlier computation and saved memory.
  • Turning a completed analysis into a reusable routine that reruns on new data.

In collections

Open-source AI agent frameworks & appsRun LLMs locallyData science courses & tools

Topics

#data-analysis#ai-agents#python#llm#reproducibility#ollama#vllm#data-science#ai#docker

FAQ

What is BambooAI?
AI data analyst that works step by step in a persistent Python kernel, keeps every computed result traceable, and can run on local open-weight models. BambooAI is an LLM-driven data analysis environment. A primary Analyst model works through a question turn by turn, choosing each time to run a Python cell, inspect earlier output, search or read supporting documents, ask a clarifying question, or write the report. Executed cells stay in a persistent workspace, so later steps and follow-up questions build on earlier computation.
Is BambooAI free to use? What license does it have?
BambooAI is open source under the MIT license. Check the license terms before using it in commercial products.
Is BambooAI still maintained?
Actively maintained. The last commit to BambooAI was on Oct 31, 2026.
What can I use BambooAI for?
Exploratory analysis of CSV or Excel datasets where the code behind each result must be auditable. Running analyses on sensitive data with locally hosted open-weight models. Follow-up analysis of a dataset over several sessions, reusing earlier computation and saved memory. Turning a completed analysis into a reusable routine that reruns on new data.
Where can I get BambooAI?
The source code and installation instructions are on GitHub: https://github.com/pgalko/BambooAI

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects