Warp launched a standalone Agent CLI that runs in any terminal, separate from the Warp desktop application. Its differentiator is that the terminal session itself becomes persistent agent state.
Warp's multiplexer and PTY architecture lets the agent remain in context across cd, work through an SSH connection without installing a remote Warp binary, and interact with full-screen or stateful programs such as Python, SQLite, gdb, htop, and Vim. A classifier routes input as a shell command or a natural-language request while preserving normal completions.
The CLI also connects to Warp's multi-agent and cloud orchestration and can delegate work to other coding agents, including Codex and Claude Code. Inference can use a Warp subscription, credits, an API key, an OpenAI-compatible endpoint, or SuperGrok.
Install with curl -fsSL https://app.warp.dev/download/agent-cli | bash. The architectural question to test is whether persistent PTY state improves completion without obscuring the audit trail; terminal continuity is useful only if commands, remote boundaries, and approvals remain inspectable.