Tine

Tine is an AI notebook runtime that lets developers run, branch, and share data experiments with AI agents in a single local workspace. Built for data scientists and developers who work with AI agents, Tine replaces linear notebooks with a tree-native model: each experiment contains branches, and each branch contains cells, so you can explore multiple approaches in parallel without overwriting prior work.

The runtime is written in Rust and runs entirely on your machine, starting a local server at http://127.0.0.1:9473 that both your browser UI and any connected AI agent share simultaneously. Because the Web UI and agent always see the same state, logs, and outputs, there is no sync lag between what you see and what the agent executes. This makes Tine a practical AI notebook for developers who want a deterministic, reproducible experimentation environment rather than a cloud-hosted black box.

Tine is MCP-native out of the box. You can connect it to VS Code, Cursor, or Claude as an agent host by running a single command — tine mcp register --host vscode — which writes the configuration directly into your host's standard config location. The tool is open source under the Apache 2.0 license and installable via pip install tine or a signed desktop bundle (.dmg for macOS, .msi for Windows) that ships with its own private Python runtime, removing any system Python dependency.

Tine is used for running branching data science and AI experiments locally, with native MCP support that lets AI agents connect to the same runtime as the developer's notebook UI. It is also used to build reproducible, parallel experiment workflows without relying on cloud-based notebook services.