GataCode

GataCode

A coding agent that runs in your terminal and on your desktop. It reads your codebase, plans, and edits with care — routing each task to the best model through Gata Inference, so you never juggle provider keys.

Install

GataCode runs on macOS, Windows, and Linux. Grab the desktop app, or install the CLI:

bash
# macOS / Linux — one-line install
curl -fsSL https://gata.net/code/install.sh | sh

# or with Homebrew
brew install gata/tap/gata

# or with npm
npm install -g @gata/code

Then authenticate — this opens a browser to connect your Gata account:

bash
gata login

One account, every model

GataCode uses your Gata account for inference, so any of the 40+ models work out of the box. No separate Anthropic/OpenAI keys to manage.

Usage

Run gata inside a repository and describe what you want in plain language. It reads the relevant files, proposes a plan, shows you a diff, and applies the change only after you approve.

bash
cd my-app
gata "refactor the auth flow to use the new session API"

# GataCode:
#   ▸ read auth/session.ts across 6 files
#   ▸ drafted a 3-step plan
#   ▸ proposed a diff  →  [a]pprove  [e]dit  [s]kip
#   ✓ applied · 142 tests passing

The loop

  • Plan — GataCode maps the task to concrete edits before touching anything.
  • Diff — every change is shown as a reviewable diff; nothing lands without approval.
  • Run — it can run your tests/build and iterate on failures.

Best model per task

Planning, editing, and quick fixes can each route to a different model automatically — you get frontier reasoning where it matters and fast models where it doesn't.

Sessions

Every task is a session you can pause and resume. Review the full diff, re-run a step, or pick up where you left off. The same sessions are available in the cloud IDE inside the console.

bash
gata sessions            # list recent sessions
gata resume <id>         # continue a session
gata diff <id>           # review the changes a session made

Prefer a UI? Open the cloud IDE at /app/code — file tree, agent transcript, and a diff viewer, with every session synced to your account.

Configuration

Pick a model per run, or set defaults in a config file:

bash
gata --model claude-opus-4-8 "write integration tests for the billing module"
toml
# ~/.gata/config.toml
[inference]
base_url = "https://inference.apigata.net/v1"
default_model = "claude-sonnet-4-6"
plan_model = "claude-opus-4-8"   # heavier model for planning

[editor]
auto_run_tests = true
confirm_before_apply = true

Per-project overrides live in a gata.toml at your repo root and take precedence over your global config.

Next steps

GataCode — Docs — Gata