What It Can Do

Within a session, TheGitAI combines local tools on your machine with the model on the server to get real work done in your repo.

Conversation & context

  • Multi-turn chat that remembers prior messages, tool outputs, and repo changes for the duration of the session.
  • Applies lightweight environment hints when it detects a Python virtual environment or a Node package.json at the project root.

Files & edits

  • Inspect files and directories.
  • Create, update, and delete files inside the target repo.
  • Track edit checkpoints during a session and safely restore changes the agent made to files or commands.

Shell

  • Run non-interactive shell commands when a task needs them — always behind the approval rules of the current mode.

Code intelligence

The code-intelligence layer is intentionally LSP-free:

  • Look up definitions, symbols, and signatures across your code.
  • Surface diagnostics from your project's own build and lint commands.
  • Reindex changed files so later turns work against fresh code.

Documents

  • Read documents such as PDF and XLSX as part of a task. Large PDFs are read in bounded chunks so a single read stays within one response.

Web

  • Fetch public URLs when a task needs information from the web. Large pages come back with coverage and continuation details so the agent can keep reading where it left off.
Point the agent at a real task — "migrate this route to the new framework and run the tests" — and it will inspect the code, make the edits, and validate with your build and test commands.