Get Started
Core Concepts
Understand the key ideas behind Thesis Notebook.
Notebooks
A notebook is an interactive document that combines code, text, and visualizations. Thesis uses the standard Jupyter notebook format (.ipynb) which stores content as JSON with the following structure:
- Cells - Individual units of content (code, markdown, or raw)
- Metadata - Information about the notebook and kernel
- Format version - nbformat 4.5 for compatibility
Cells
Cells are the building blocks of notebooks. Thesis supports three cell types:
- Code cells - Executable code with syntax highlighting, autocomplete, and output display. Outputs can include text, images, HTML, and interactive visualizations.
- Markdown cells - Rich text with formatting, headers, lists, links, and images. Uses a WYSIWYG editor for easy editing.
- Raw cells - Unformatted text that passes through unchanged during conversion.
Kernels
A kernel is a computational engine that executes code. When you run a code cell, the code is sent to the kernel which processes it and returns results.
Key kernel concepts:
- Execution state - Kernels can be
idle,busy,starting, ordead - Kernel session - Maintains state between cell executions (variables, imports, etc.)
- Interrupt - Stop a running computation without losing state
- Restart - Clear all state and start fresh
Thesis connects to Jupyter kernels via WebSocket for real-time bidirectional communication, enabling features like streaming output and code completion.
Sessions
A session associates a notebook file with a running kernel. When you open a notebook, Thesis creates or reuses a session to manage the connection between your document and the computational backend.
Sessions track:
- The notebook path
- The associated kernel
- Execution history and state
AI Agent
The Thesis AI agent is an intelligent assistant that can understand your notebook context and help with tasks. It operates in three modes with different capabilities:
- Ask mode - Read-only exploration with 4 tools
- Plan mode - Research and planning with 5 tools
- Agent mode - Full autonomous execution with 13 tools
Learn more in the Agent documentation.