TheGitAI — Agentic AI Coding Tool for Your Terminal

TheGitAI

3 min read Original article ↗
public preview · agentic ai coding tool

Install and work—no provider account, API keys, or model wiring. Give TheGitAI real autonomy while assistant edits remain reviewable, undoable, and recoverable by checkpoint.

$ npm i -g @thegitai/cli

No API keys Curated frontier models macOS, Linux, Windows

TheGitAI terminal interfacerepresentative session

● System

Interactive mode ready. Type /help for commands.

● You

Trace the signup bug, fix it, run the focused tests, and update the docs.

● Tool grep_code signup validation -> OK

● Tool read_file signup.test.ts -> OK

● Tool web_fetch framework validation docs -> OK

◷ Working · 1m 08s

To-dos · 4/4 done

Map the signup and validation flow

Fix the underlying validation bug

Run the focused test suite

Update the signup guide

∴ ThinkingValidation fixed, 42 tests clean, documentation current. Reviewing the final diff now.

queue>Type to queue the next message

TheGitAI · ~/projects/acme-app

GPT-5.6 Sol (OpenAI)

Enter queues · Esc / Ctrl+C cancel turnAuto-Accept

Real work crosses files, tools, processes, tests, and documentation. TheGitAI keeps that chain coherent from the first investigation to the final handoff—and every change it makes along the way stays reversible.

Several lines of inquiry, one coherent result

Independent file reads, code searches, symbol queries, document reads, and web lookups run together. Mutations stay ordered.

The project keeps running while the agent works

Keep dev servers, watchers, builds, and local APIs alive. Follow output, inspect logs, and stop jobs without blocking the conversation.

Explore the complete capability set

From request to reviewed result

Give TheGitAI the outcome, not a sequence of tiny prompts. It can carry the work through investigation, implementation, verification, documentation, and handoff.

Control without hesitation

Trust does not come from a promise that nothing can go wrong. It comes from seeing what the agent is doing, holding the sensitive actions behind an approval, and knowing where it can and cannot reach.

Approvals

Ask first, or get out of the way

Default mode asks before edits and commands. Auto-Accept keeps trusted work moving. Plan mode stays read-only.

Containment

It stays inside the repository you opened

Work is scoped to the project directory, sensitive paths are guarded, and secrets are stripped from the command output that reaches the model and your session history.

Inspectability

A source-visible local client

Local file, shell, session, approval, containment, and redaction behavior stays inspectable. Provider credentials remain server-side.

We don't store your source files, your prompts, or your conversations on our servers. Session history stays on your machine; the server keeps metadata only.

Privacy policy

There is no dashboard to learn and no configuration ceremony. The shortest path to value is the repository already open in your terminal.

Install the CLI

One global install gives you the ai command. No project plugin or provider setup.

$ npm i -g @thegitai/cli

Sign in once

Use the browser flow locally or a one-time code on a remote machine. No API keys.

$ ai

Open a real repository

The agent works against the repository in your current directory and maps its structure.

$ cd your-project && ai

Ask for the outcome

Describe the finished result. Review progress, approve sensitive actions, and inspect the diff.

❯ trace the bug, fix it, and verify it

Real autonomy, still under your control

Hand it a real repository and a real task. Assistant edits stay reviewable, undoable, and recoverable by checkpoint while TheGitAI carries the work to a verified change.

No credit cardNo provider keysmacOS, Linux, and Windows