Warp — Agentic Development Environment in the Terminal

Context

Warp is an Agentic Development Environment (ADE) built on top of a modern terminal emulator. Unlike traditional terminals like iTerm2 or Windows Terminal — which only run commands — Warp integrates AI capabilities directly into the work surface: it can run coding agents like Claude Code and Codex, offers a code editor with LSP support, and orchestrates local and cloud agents through its Oz platform.

The terminal is built in Rust, giving it native performance. Its UI organizes output into blocks (each command and its result form a navigable unit) instead of the continuous scroll of a conventional terminal. This fundamentally changes how you interact with output: you can select, search, share, and filter individual blocks without infinite scrolling.

Warp went open source under AGPL v3 in 2025, and its code lives at github.com/warpdotdev/warp.

Setup

Installation

macOS (10.14+):

brew install --cask warp

Also available as .dmg from warp.dev/download.

Linux (Debian/Ubuntu):

sudo dpkg -i warp-terminal_<version>_amd64.deb

Also available as .rpm, .tar.zst (Arch), and AppImage for x64 and ARM64.

Windows (11/10):

winget install Warp.Warp

Also available as .exe for x64 and ARM64.

First steps

1. When you open Warp for the first time, select your shell (zsh, bash, fish, PowerShell). Warp respects your existing shell configuration, aliases, history, and environment variables.
2. Choose your visual theme and default AI model in Settings.
3. To start with agents, open Warp Agent from the sidebar or invoke a CLI agent (Claude Code, Codex, etc.) directly from the terminal.

Third-party CLI agent configuration

Warp can run any CLI agent with its agent toolbelt (notifications, code review, rich input). No special configuration needed: just install the agent's CLI on your PATH and Warp detects it automatically.

For Claude Code, for example:

npm install -g @anthropic-ai/claude-code

Then run it from Warp and you get the UI integrations (vertical tabs, notifications, interactive code review).

Examples

Example 1: Warp Agent — refactor a component

In Agent mode (toggle in the top bar or Cmd+E), describe the task in natural language:

Refactor the ProductCard component to accept a variant prop
with values "default" and "compact". Styling should use the tokens
already defined in tailwind.config.ts.

Warp Agent:

1. Reads tailwind.config.ts and existing components.
2. Modifies ProductCard.tsx adding variant logic.
3. Updates imports if needed.
4. Shows an interactive diff in the code review panel.

You can accept, reject, or request adjustments before applying changes.

Example 2: Running Claude Code with Warp's toolbelt

From a new vertical tab, run Claude Code normally:

claude

Warp detects it's a CLI agent and automatically enables:

• Notifications: alerts you when the agent needs approval or finishes a task.
• Code review: interactive diff where you can leave comments and send them back to the agent with one click.
• Shared context: Warp Drive, rules, and MCP servers are available to any agent running inside Warp.

Example 3: Warp Drive workflows

A workflow is a parameterized command you can save and reuse. Defined in YAML:

name: Deploy to staging
command: git push origin {{branch}} && ssh deploy@staging "cd /app && git pull && npm run build && pm2 restart app"
tags:
  - deploy
  - staging

Save it in Warp Drive and execute it from the Command Palette (Cmd+P) by searching "Deploy to staging". Warp prompts you for the branch value before executing.

Particularities

• Blocks, not infinite scroll: each command and its output form a navigable block. You can filter by command, search within a block, and share individual blocks as links. This is the most noticeable UX change compared to traditional terminals.
• Terminal and Agent modes: Warp offers two modes. In Terminal mode it works as a conventional terminal (with blocks and autocompletion). In Agent mode, the input bar becomes a multi-turn conversation with the agent. You can switch between modes without losing context.
• Interactive code review: when an agent generates changes, Warp shows a diff where you can comment on specific lines and send those comments back to the agent. This takes agent work from 80% to 100% without leaving the terminal.
• Vertical tabs with metadata: tabs show git branch, worktree, and associated PR — not just the command title. This is useful when working with multiple agents in parallel.
• Oz for cloud orchestration: the Oz platform lets you run agents on Warp's infrastructure (or your own) with triggers (Slack, GitHub, webhooks), schedules, and parallelism. It's not just a terminal: it's an agent orchestration system.
• Open source (AGPL v3): Warp's client is open source. You can read the code, file issues, and contribute. This is unusual for a commercial terminal and sets it apart from iTerm2 (permissive license but less actively hacked on) and Ghostty (closed or more limited).
• SOC 2 and zero data retention: Warp has SOC 2 Type 2 certification and zero data retention agreements with LLM providers. AI data is not stored or used for training. AI features can be disabled globally.
• Not a full editor: the built-in code editor is useful for quick reviews and light edits, but it doesn't replace VS Code or Neovim for extensive development. Its value lies in integration with the agent workflow, not in being a competitive IDE.
• Cloud account dependency: features like Warp Drive, config sync, and cloud agents require a warp.dev account. The terminal works offline, but AI and collaboration features need connectivity.