A coding agent that works inside your projects

What it does

Use any model

Connect a hosted provider with an API key, or run a model locally with Ollama or LM Studio. tBelt Code doesn't pick a provider for you.

Works through the task

It searches and reads your code, edits files with targeted search and replace, then runs your tests or build to check the change. Each step shows up in the session as it happens.

Undo the last turn

Before it edits a file, tBelt Code saves a git checkpoint. Type /undo to put the files back the way they were. Checkpoints are kept under their own ref, so your branches, index and stash are left alone.

Remembers what you tell it

Type /remember followed by a fact, and later sessions will recall it, in this project or in all of them. Anything that looks like a key or password is redacted before it is saved.

Starts with a map of your repo

At the start of a session it lists your files and the classes, functions and imports in each, and gives that map to the model.

Runs commands in a sandbox

Commands run inside the operating system's sandbox. Pick Read Only, Workspace Write or Full access for each session. Commands such as rm -rf, git reset --hard or a force push wait for your confirmation before they run.

Keeps your sessions on your machine

Sessions are saved per workspace on your computer. tBelt Code doesn't upload them anywhere; only the requests to your model provider leave the machine. Open an old one to continue, search through them, or archive the ones you are done with.

Splits up large jobs

Hand parts of a task to the Inspector, Coder and Tester subagents, and save agent presets for the kinds of work you repeat.

Connects to your tools

Add MCP servers, skills and plugins from the settings.

Stays up to date

On Windows and Linux the app checks for new versions and tells you when one is out. One click downloads it and restarts into the new version. On macOS, download the new version from this page.