2026-04-08

• #0003

  #0003Agents & toolsHacker News · AI Agentlast month

  `marimo pair`: reactive notebooks as a live workspace for agents

  60radar

  marimo pairNotebook workspace — agent-ready reactive sessions

  Agents can use a running marimo notebook as working memory, scratchpad, and persistent Python state. Strong fit for research and data workflows; worth trying now, but still early for production defaults.

  • —Instead of one-off scripts, the agent works inside a stateful marimo session and can inspect variables, run code, and read outputs in place.
  • —State can be persisted by adding or deleting notebook cells and installing packages; the notebook stays a plain Python file with an executable trail.
  • —The setup exposes a semi-private code mode, turning the notebook into a reproducible REPL that effectively extends context windows.
  • —marimo guardrails matter here: dependent cells rerun automatically and deleting a cell scrubs its variables, reducing hidden-state drift.
  • —They already use it for data exploration and self-debugging, which signals practical utility beyond a demo even if the project is experimental.

  Source: github.com/marimo-team/marimo-pairRead original →
• #0002

  #0002Agents & toolsHacker News · Show HN AIlast month

  `marimo pair`: Put Agents Inside Reactive Python Notebooks

  60radar

  marimo pairNotebook tool — agent pairing inside live marimo sessions

  Agents can attach to a live marimo notebook, use it as persistent working memory, run scratch code, and rewrite the notebook itself. That turns the notebook into both an execution trace and a human-agent workspace, so it's worth testing for research-heavy workflows now.

  • —A single command, /marimo-pair pair with me on my_notebook.py, binds an agent to a running notebook and exposes the full session state.
  • —Agents can probe state in an ephemeral scratchpad, then persist useful work by adding cells, deleting cells, or installing packages into the notebook file.
  • —marimo's reactive graph removes hidden state: rerunning propagates to dependents, and deleting a cell scrubs its variables from memory.
  • —The team positions code mode as a REPL that extends context windows, but unlike throwaway scripts it incrementally builds a reproducible Python program.
  • —Early uses center on data exploration, hypothesis testing, executable answer traces, and even self-hosted debugging of marimo pair and marimo itself.

  Source: github.com/marimo-team/marimo-pairRead original →
• #0001

  #0001Agents & toolsHacker News · LLMlast month

  `marimo pair`: Agents inside reactive Python notebooks

  50radar

  marimo pairNotebook tool — connects a live marimo session to an agent

  Agents can inspect state, run scratch code, and persist changes inside a live marimo session. Strong pattern for research and data workflows; still early but worth watching.

  • —A running notebook becomes working memory plus a reactive Python runtime, so the agent edits state instead of juggling disposable scripts.
  • —The /marimo-pair pair with me on my_notebook.py flow connects an agent to a live session and lets it add or delete cells and install packages.
  • —code mode works like a REPL that extends context, but every action lands in a reproducible Python notebook rather than hidden session state.
  • —Auto-reruns of dependent cells and variable scrubbing on delete reduce notebook drift, which makes human-agent collaboration more reliable.

  Source: github.com/marimo-team/marimo-pairRead original →