AnyT Notebook

The workflow development environment for AI agents.

AI agents are powerful, but they work like chatbots -- you give them a prompt, hope they get everything right in one shot, and when things break halfway through, you start over. AnyT Notebook changes this. It gives AI agents a workflow layer: visible steps, human checkpoints, and recoverable state.

Why AnyT Notebook

When you ask an AI agent to do something complex, the one-shot chatbot model falls apart:

• No visibility -- You can't see the plan before it runs
• No control -- You can't pause, review, or redirect mid-execution
• No recovery -- If step 4 of 8 fails, you redo everything
• No separation -- AI handles things a simple shell command could do faster

AnyT Notebook solves this with a Jupyter-style notebook for AI agent workflows. Break tasks into cells, add human checkpoints, mix AI and shell scripts, and run with full control.

Supported Runtimes

AnyT Notebook works with multiple AI runtimes:

Select your runtime from the dropdown in the notebook toolbar. Switch between runtimes at any time.

Quick Start

1. Install -- Search "AnyT Notebook" in VS Code Extensions

2. Try a Sample -- Cmd+Shift+P -> "AnyT: Open Sample Notebook" -> pick one -> click Run

3. Or Create Your Own:

• Cmd+Shift+P -> "AnyT: New Notebook"
• Click + Task, + Shell, + Input, + Break, or + Note to add cells
• Type your instructions directly into each cell
• Click Run to execute the notebook step by step

Cell Types

AnyT Notebook provides five cell types. Mix and match them to build workflows that combine AI execution, automation, and human oversight.

Task

Task cells are executed by your AI agent (Claude Code, Codex, or Gemini). Write a natural language instruction and the agent carries it out. Use tasks for anything that requires reasoning, code generation, or multi-file changes.

Example: "Create a REST API with Express.js and TypeScript, including user authentication routes."

Shell

Shell cells run scripts directly -- no AI involved. They're fast, deterministic, and ideal for commands you already know. Use them for installs, builds, tests, linting, or any CLI operation.

Example: npm install && npm run build && npm test

Input

Input cells pause execution and present a form to the user. Use them to collect configuration values, choose between options, or supply information that downstream cells need. Form responses are available to subsequent cells.

Example: "Choose deployment target" with a dropdown for staging vs. production.

Note

Note cells are markdown documentation checkpoints. They auto-complete instantly when reached and serve as section headers, context annotations, or progress markers in your workflow.

Example: A section header like "Phase 2: Testing and Validation" to organize your notebook.

Break

Break cells pause execution so you can review what's happened so far. Inspect outputs, verify results, then click Continue when ready. Add breaks liberally while developing a workflow; remove them once you're confident it runs end-to-end.

Example: "Review the generated project structure before adding API routes."

Workflow Lifecycle

AnyT Notebook treats AI workflows like code -- build, test, and refine them iteratively:

1. Create -- Add cells visually: tasks for AI work, shell for scripts, inputs for user decisions, notes for documentation
2. Debug -- Insert break cells as checkpoints and run step by step to inspect each output
3. Iterate -- Review results, fix failing steps, reorder or add cells as needed
4. Harden -- Remove breakpoints once confident and run the full workflow end-to-end
5. Share -- Version the .anyt.md file and share with your team

Sample Notebooks

Open Command Palette -> "AnyT: Open Sample Notebook" -> pick one -> click Run.

Commands

All other actions (run, stop, reset, add cells) are available directly from the notebook toolbar.

Requirements

• Claude Code CLI installed and authenticated (install guide), Codex CLI installed, or Gemini CLI installed (install guide)
• VS Code 1.85+

Configuration

Links

• Product Overview
• User Guide
• GitHub
• Report Issues