Warp Ships a Standalone Agent CLI: pty-Multiplexed Terminal Agent Runs Alongside Claude Code

Warp shipped a standalone Agent CLI (installs in any terminal) on August 4 — pty-native control for interactive-terminal tasks that pure file-editing agents can't handle.

Warp Agent CLI separates the agent from the Warp Terminal application. Install it in Ghostty, iTerm, VS Code, or any system shell with one line:

curl -fsSL https://app.warp.dev/download/agent-cli/install.sh | sh
# Windows: winget install Warp.WarpAgentCLI

Pricing: included in Warp subscriptions ($18/mo with $20 inference credits), ad-hoc credits from $10, or bring your own API key.

What's architecturally different from Claude Code

Claude Code reads files, writes code, and runs shell commands — it treats the terminal as an I/O pipe. Warp Agent CLI runs a pty (pseudoterminal) multiplexer between itself and your shell. That means the agent sees and drives full-screen terminal UIs:

• A sqlite3 REPL — type SQL, read results interactively, step through a migration
• gdb with breakpoints set and stepped
• htop or a live log tail it monitors over time
• Any interactive prompt that requires keystrokes, not just piped command output

Claude Code can approximate some of these with enough prompting, but pty-native control is deterministic.

A pattern that works: both in parallel

# Tab 1 — Warp Agent watching a live test suite
warp-agent "run pytest in watch mode; alert me when coverage drops below 80%"

# Tab 2 — Claude Code doing a large refactor in the same repo
claude "rename every AuthService method from camelCase to snake_case and update all callers"

Each sends desktop notifications when it needs input. They don't block each other: Warp holds the interactive terminal loop, Claude Code holds the file-editing loop. Warp's own docs show this as a supported pattern — Claude Code runs as a CLI agent inside Warp tabs, with Warp adding native notification and workspace split.

When to reach for which

Task	Reach for
Read files, write code, manage git	Claude Code
Interactive REPL / debugger / live monitor	Warp Agent CLI
Long-horizon autonomous task (hours)	Claude Code (/goal or --continue)
Multi-agent orchestration across terminals	Warp Agent CLI (built-in cloud handoff)

Why it matters: The Warp Agent CLI fills the "interactive tool driving" gap in terminal-only setups. If your workflow has tasks that need pty control — database migrations, interactive test runners, debugger sessions — you now have a CLI-native option that works in whatever terminal you already use.

Sources: Warp Agent CLI launch post · Warp Agent CLI docs · Warp vs Claude Code — warp.dev