AI Coding Tools and IDE Quickstart

Start with an IDE, Git, the project's runtime, and one coding agent. Install commands, product plans, and limits can change, so verify them against the linked official documentation.

1. Why an IDE Makes the First Steps Easier

An IDE or code editor gives you one place to open the project, search files, read diagnostics, run a terminal, inspect Source Control, and review diffs. A terminal agent can still do the implementation, while the IDE gives you a clear human review surface.

Do not run several agents in the same folder. Use one primary tool until you understand permissions, model selection, diffs, stopping a bad loop, and Git recovery.

2. Install Visual Studio Code

VS Code is a free editor for Windows, macOS, and Linux. Download it from the official VS Code page.

On Windows, the User Installer is the usual choice and does not normally require administrator access. Run the .exe, restart your terminal, and open a project with File > Open Folder or code .. On macOS, open the .dmg, move the application into Applications, then use the Command Palette to install the code command in PATH. On Linux, use the official .deb, .rpm, Snap, or distribution-specific instructions.

Verify the installation:

code --version

Learn Explorer, Search, Source Control, Terminal, Problems, Extensions, and the diff editor before approving repository-wide edits.

3. Install Google Antigravity IDE

Google Antigravity IDE combines an editor, agent, terminal, browser, and artifact review. Download the build for your operating system and CPU from the official Antigravity download page. The page provides Windows x64 and ARM64, macOS Apple Silicon and Intel, and Linux x64 and ARM64 installers with minimum requirements.

After installation, sign in, open one project folder rather than a broad personal directory, and begin with an analysis-only task. Review plans, artifacts, diffs, and commands before allowing changes. Antigravity also offers editor extensions, but a beginner should choose either Antigravity as the main workspace or VS Code with one agent, not two agents writing to the same working tree.

4. Install Git and the Runtime

Install Git and verify it with git --version. For JavaScript or TypeScript projects, install the repository's required Node.js version or a supported LTS release from Node.js downloads, then check node --version and npm --version.

Read package.json, the lockfile, README, and implementation guide before changing a runtime or package manager. A package-lock.json normally indicates npm, pnpm-lock.yaml indicates pnpm, and yarn.lock indicates Yarn.

5. OpenAI Codex

Codex is available through multiple coding surfaces. For the terminal on macOS or Linux, current documentation provides:

curl -fsSL https://chatgpt.com/codex/install.sh | sh
cd /path/to/project
codex

Create a Git baseline, use /init only when repository instructions are appropriate, inspect /status, and begin with a read-only prompt. Ask Codex to identify sources of truth, verification commands, risks, and a plan before editing.

6. Claude Code

Claude Code is a terminal coding agent. Current official installers include:

# macOS, Linux, or WSL
curl -fsSL https://claude.ai/install.sh | bash

# Windows PowerShell
irm https://claude.ai/install.ps1 | iex

Use CLAUDE.md for stable project rules such as stack constraints, commands, security boundaries, and definition of done. Keep temporary ticket details in the task prompt or plan.

7. Gemini CLI

Gemini CLI is Google's active terminal agent. The reviewed installation path requires Node.js 20 or newer:

npm install -g @google/gemini-cli
cd /path/to/project
gemini

You can also use npx @google/gemini-cli. Put stable project context in GEMINI.md, keep the first request narrow, and review Google's current quota documentation instead of assuming a universal five-hour window.

8. OpenCode

OpenCode is a provider-agnostic open-source coding agent, not a model provider. Install using an official documented method:

curl -fsSL https://opencode.ai/install | bash
# or
npm install -g opencode-ai

The model, quota, privacy boundary, and cost follow the provider or credit path you configure. In an existing AI Blueprint ZIP, read its instructions before running /init so a generated AGENTS.md does not overwrite or conflict with the handoff.

9. ZCode, GLM, Zed, and Kimi Code

In the Z.AI ecosystem, ZCode is a coding surface and GLM is a model family. Zed is a separate code editor. Confirm which product you mean before installing anything. Follow Z.AI's current integration documentation because model and client paths change.

Kimi Code is a coding agent in the Kimi ecosystem. Its official terminal installer currently uses:

curl -fsSL https://code.kimi.com/kimi-code/install.sh | bash

Follow the login and launch instructions shown by the installer.

10. A Portable First Prompt

Read the repository instructions, README, manifest, PRD, schema, and relevant docs.
Do not edit files yet.
Report the source of truth, project structure, commands, dependencies,
security boundaries, conflicts, and missing information.
Propose a plan for one small first task and its verification steps.

11. Safety Before Command Access

• Keep API keys out of repository instructions and prompt history.
• Store secrets in ignored local files or a secret manager.
• Read commands before approving destructive actions.
• Do not use a production database for experiments.
• Keep one agent per working tree.
• Commit before a refactor or migration.
• Do not run third-party installer scripts without verifying their source.