Show HN: Pantheon-CLI – Open-Source Python Claude Code and Smart Notebook

Unlike most AI coding assistants, Pantheon-CLI runs entirely on your machine (or server). No data upload required. It blends natural language and code in a single workflow, keeping variables in memory and letting you switch seamlessly between typing code and asking in plain English.

What it does: 1. Chat with your data: Directly process CSV, Excel, AnnData, Pickle, Torch tensors, or any format supported by Python/R/Julia. 2. Mixed programming: Variables persist across natural language and code; the CLI auto-generates and runs code for you. 3. MCP-like agent integration: Read/create files, run commands, fetch web pages, generate/revise code. 4. Human-like learning: Feed it a PDF paper or tutorial—Pantheon-CLI reads it, plans steps, and replicates methods before analysis. 5. Task planning: Builds scientific agents by learning from papers/tutorials (not just fixed, human-predefined steps). 6. Multi-model support: Works with OpenAI, Anthropic, Gemini, DeepSeek, Qwen, etc. + offline local LLMs (ollama, deepseek, gpt-oss). 7. Multi-RAG support: Pre-learns from docs/web into a local “brain” for more credible outputs without massive token costs. 8. Built-in biology toolsets: For omics analysis (alignment, annotation, differential expression, full paper reproduction). 9. Notebook mode: Brings the same agentic workflow into Jupyter—automatically runs and revises code, operates on files, and learns from tutorials/papers.

Pantheon-CLI is our attempt to push beyond “AI writes code for you.” Instead, it’s an agentic operating system for data analysis, spanning both terminal and notebook.

Repo: https://github.com/aristoteleo/pantheon-cli

Tutorial: https://pantheonos.stanford.edu/cli/docs/intro/getting-start...

Home page: https://pantheonos.stanford.edu/

Would love to hear feedback from the HN community—what use cases would you try this for, and what features would make it more useful to you?