Warp's Agent CLI brings its coding agent to any terminal

Warp released Agent CLI, a standalone command-line product that brings its coding agent to any terminal, including Ghostty, iTerm2, VS Code, macOS Terminal, and Windows Terminal. The CLI carries over Warp's model routing, multi-agent orchestration, and cloud handoff, and is part of Warp's strategy to turn its terminal architecture into reusable agent infrastructure. Warp also launched Oz, open-sourced its client, and added bring-your-own inference, extending its product beyond the terminal application.

Warp's standalone Agent CLI https://www.warp.dev/agent-cli?ref=runtimewire gives developers access to its coding agent inside Ghostty, iTerm2, VS Code, macOS Terminal, Windows Terminal and other terminal environments. The command-line product carries over Warp's model routing, multi-agent orchestration /article/ruflo-claude-code-multi-agent-shared-memory and cloud handoff while using the terminal infrastructure Warp has built over several years. Warp announced the product in a launch post https://www.warp.dev/blog/introducing-the-warp-agent-cli-coding-agent?ref=runtimewire . The release advances Lloyd's original thesis without requiring developers to adopt his terminal. Before founding Warp https://www.warp.dev/?ref=runtimewire , Lloyd worked as a principal engineer at Google, served as CTO of Time and co-founded SelfMade. Warp began with his view that the terminal was one of the two tools almost every programmer used, alongside the code editor, yet had received far less product work. Warp's team profile describes that background https://www.warp.dev/blog/whos-behind-warp-meet-the-team?ref=runtimewire . Opening the agent to other terminals changes how Lloyd is pursuing that idea. Warp can meet developers inside their existing setup and retain its technical bet on controlling the session between an agent and the shell. That makes the CLI a distribution move and an attempt to turn Warp's terminal architecture into reusable agent infrastructure. Warp has also opened parts of its product strategy beyond the terminal application. It launched Oz https://www.warp.dev/newsroom/2026/2/10/warp-launches-oz-the-orchestration-platform-for-cloud-coding-agents?ref=runtimewire as a platform for running and managing cloud coding agents, open-sourced the Warp client https://www.warp.dev/blog/warp-is-now-open-source?ref=runtimewire and added bring-your-own inference https://www.warp.dev/blog/bring-your-own-inference-to-warp?ref=runtimewire . The standalone CLI extends that strategy to the interface itself: developers can use Warp's agent without changing terminals or committing to one model provider. A coding agent built around terminal sessions Warp's main technical claim concerns session control. The CLI manages pseudo-terminal, or pty, connections through a multiplexing layer that Warp compares with tmux. The agent remains connected to the session as the developer changes directories, moves between repositories or connects to another machine over SSH. Warp says the remote workflow does not require installing its binary on the target host. That architecture also lets the agent operate interactive and full-screen programs. Warp demonstrates it working inside SQLite and Python sessions, editing through Vim, setting breakpoints with gdb and monitoring processes with htop. These are workflows where a coding agent needs to understand changing terminal state rather than execute a series of isolated shell commands. Warp also includes tab completion and a classifier that decides whether an input is a shell command or a natural-language prompt. The agent routes requests among frontier and open-weight models according to task complexity, according to Warp. Developers can supply API keys or an OpenAI-compatible endpoint, and teams can define custom routing rules that assign particular models to planning, design or implementation work. Warp's product page says inference purchased through Warp is charged at API cost without a markup. Warp also connects local sessions to its Oz cloud product. A developer can start a task in the CLI, dispatch parts of it to subagents and hand the session to a cloud agent before closing a laptop. Warp says its orchestration layer can delegate work across different models and separate agent harnesses, including Claude Code and Codex, while showing each session in one interface. These capabilities are Warp's claims and have not been established through the launch announcement's comparative testing. Access to the Agent CLI is included in Warp's free tier, which also permits developers to bring their own inference. Warp's pricing page https://www.warp.dev/pricing?ref=runtimewire lists its Build plan at $20 per month, or $18 per month with annual billing, with $20 of included agent usage at API rates. Developers can also purchase additional usage. Lloyd is trading terminal lock-in for wider distribution The command line has become one of the busiest fronts in AI developer tools. Anthropic's Claude Code https://docs.anthropic.com/en/docs/claude-code/getting-started?ref=runtimewire established a direct agent workflow inside the shell, while GitHub Copilot CLI https://github.com/features/copilot/cli?ref=runtimewire now offers model selection, parallel subagents and integration with GitHub issues and pull requests. Model choice and subagent support are already common selling points. Warp is placing its competitive case on control of the terminal session. Persistent shell state, interactive applications, SSH continuity and local-to-cloud handoff are the parts that draw directly from the product Lloyd originally built. Warp calls the resulting experience superior to other agent CLIs. The announcement provides architectural detail for that claim, while performance and reliability will depend on how the agent behaves across real repositories, remote systems and long-running sessions. Separating the agent from Warp Terminal removes one reason developers might switch terminal applications. It also gives Warp a larger pool of potential users and turns the terminal into one client among several for Oz. Warp can sell inference, cloud execution and team controls even when the developer keeps Ghostty, iTerm2 or VS Code as the visible interface. Lloyd has financing for that expansion. Warp raised a $50 million Series B https://www.warp.dev/blog/warp-drive-series-b?ref=runtimewire in June 2023, with participation from existing backers including GV, Neo, BoxGroup and Figma CEO Dylan Field. The round also added angel investors Sam Altman and Tobi Lutke. The Agent CLI turns Lloyd's founding wager inside out. He started Warp because the terminal was neglected. He is now using the infrastructure beneath that terminal to reach developers who never intended to replace the one they already use.