Warp AI terminal: Warp Agent, CLI agents, rules, skills and pricing

What is the Warp AI terminal?

Warp is a terminal for macOS, Windows and Linux with a built-in coding agent, the Warp Agent, that can run commands, read their output, edit code and work through multi-step tasks while you approve its actions. It also hosts other agents such as Claude Code and Codex, and since April 2026 its client is open source.

The Warp docs now call it "an open-source Agentic Development Environment." In practice that means three layers:

• The Warp terminal, a modern terminal with an agent conversation view, a code editor and code review.
• The Warp Agent, which you use in the app, in any terminal through the Warp Agent CLI, or in the cloud.
• The Automation Platform and Warp Factories, which run cloud agents from schedules, integrations and APIs for team workflows.

On 28 April 2026 Warp announced that the Warp client is now open-source. The warpdotdev/warp repository says the UI framework crates are MIT licensed and "the rest of the code in this repository is licensed under the AGPL v3." This guide follows the docs as read on 26 October 2026.

How do you install Warp?

The installation page covers macOS (Intel and Apple silicon), Windows (x86_64 and ARM64) and Linux (x86_64 and ARM64). Download it from the Warp site, or use a package manager:

# macOS
brew install --cask warp

# Windows
winget install Warp.Warp

On Linux, Warp ships .deb, .rpm and Arch packages. The .deb and .rpm packages set up Warp's repository for updates, and on Arch the app walks you through it on the first update. All installs auto-update. Creating a Warp account is optional for the terminal, and Warp loads your login shell if it is bash, fish, zsh or PowerShell.

How do Terminal mode and Agent Mode work?

Warp keeps shell work and agent work in two modes. The modes page describes "a clean terminal for running shell commands, and a dedicated conversation view for multi-turn interactions with the Warp Agent."

• Terminal mode is the default in each new tab. It looks like a normal prompt, with a hint bar at the bottom.
• Agent conversation view adds a model selector, voice input, image attachments and conversation history.

On macOS, ⌘↩ starts a new agent conversation and ⌘I switches between command and Agent Mode. When a command fails, Warp offers to attach its output to the agent as context. With auto-detection on, Warp labels input it thinks is natural language as an agent prompt and shows "(autodetected)." In the agent view, prefix input with ! to force it to run as a shell command, for example !git status. Both auto-detection toggles live under Settings > Agents > Warp Agent > Input.

Commands the agent runs stay inside that conversation and feed its context, while your own terminal blocks stay in the terminal list and can be attached to any conversation.

What can the Warp Agent do?

The agents overview says the Warp Agent "works through multi-step tasks on its own, running commands in a real terminal and using the output to decide what to do next." A few capabilities stand out:

• Full Terminal Use. The agent can work inside interactive programs such as psql, vim, python, gdb or a running dev server, reading the live output and typing into it. Start the tool yourself and tag the agent in, or ask it to start one.
• Planning. /plan produces an editable plan in Warp's rich text editor. Each agent revision is a new version, and you can ask it to execute only part of the plan, such as "Implement phase 1 of the plan."
• Codebase Context. Warp indexes Git-tracked code when you open a repo so the agent can ground answers in real files. The docs say "No code is stored on Warp servers."
• Code review. You review the agent's diffs before they land, in Warp's interactive code review.
• Cloud agents. ⌥⌘↩ starts a cloud agent conversation that runs in an isolated environment, useful for parallel or long-running work.

The same agent also runs outside the app. The Warp Agent CLI is a standalone warp command that works in any terminal or over SSH, and picks up your project's rules (such as AGENTS.md), skills and MCP servers:

curl -fsSL https://app.warp.dev/download/agent-cli | bash
warp

How do you control what the Warp Agent may do?

Through Agent Profiles, under Settings > Agents > Profiles. The profiles and permissions page says each profile sets a base model and permissions for actions such as applying code diffs, reading files, creating plans, executing commands and interacting with running commands.

For applying code diffs, Agent Decides currently behaves like Always ask, so you always review diffs unless you choose Always allow. You can add regex patterns such as ls(\s.*)? to a command allowlist so they auto-run, and a command denylist overrides both the allowlist and Always allow (Run until completion is the one exception). Setting everything to Always allow gives the agent full autonomy, which the docs call "YOLO mode."

How do you give the Warp Agent rules and skills?

Rules. The rules page describes Global Rules, which apply everywhere, and Project Rules, which "are stored in an AGENTS.md file (or WARP.md for backwards compatibility)." Warp applies the AGENTS.md in the repo root and in your current directory, and tries to include a subdirectory's file when the agent edits there. The filename must be all caps, and if WARP.md and AGENTS.md share a folder, WARP.md wins. Run /init to index the repo and generate an AGENTS.md, or /add-rule to create a Global Rule. /init can also link an existing rules file to AGENTS.md instead, and supports CLAUDE.md, .cursorrules, AGENT.md, GEMINI.md, .clinerules, .windsurfrules and .github/copilot-instructions.md.

Skills. The skills page says the agent receives every available skill's name and description at the start of a conversation and loads the full instructions when one fits. You can also invoke one with /{skill-name}, and skill bodies can use argument placeholders such as $ARGUMENTS. Warp scans an unusually wide set of folders, so skills written for other tools work without moving:

Discovery follows your working directory up to the repository root, so skills from one project do not leak into another. The format follows the Agent Skills standard.

MCP. Add MCP servers under Settings > Agents > MCP servers. Warp supports command-based servers and Streamable HTTP or SSE servers, and the docs say configurations from most MCP clients can be pasted in directly.

Can you run Claude Code, Codex and other CLI agents in Warp?

Yes, and Warp treats them as first-class. The third-party CLI agents page says Warp "auto-detects supported CLI agents and enhances them" with a rich input editor, notifications, inline code review and Remote Control for steering a session from another device. The supported list includes Claude Code, Codex, OpenCode, Amp, Auggie, Copilot CLI, Cursor CLI, Gemini CLI, Droid, Pi, goose, Antigravity, Hermes, Mistral Vibe, Grok Build and Kiro CLI, though not every feature works with every agent.

These agents run in a normal terminal pane and keep their own login, models and config. They do not appear in Warp's Conversation Panel, which lists only Warp Agent and cloud conversations. If you already use Codex CLI or Claude Code, Warp can simply be the terminal you run them in.

Which models does the Warp Agent use and what does it cost?

The model choice page offers a curated set of models from OpenAI, Anthropic, Google and open source providers, plus Auto options that pick for you: Responsive, Cost-efficient, Genius and Open-weights. You can also bring your own API key for OpenAI, Anthropic or Google, connect an OpenAI-compatible endpoint such as OpenRouter, or connect a ChatGPT or SuperGrok subscription. The BYOK page says: "You can't connect a Claude consumer subscription to the Warp Agent." BYOK is open to individuals and organizations with 10 or fewer employees, and larger teams need Business or Enterprise.

As of October 2026, the Warp pricing page listed these monthly prices:

On privacy, the agents overview says Warp has "Zero Data Retention policies with all contracted LLM providers," and you can turn the AI features off under Settings > Agents > Warp Agent.

Where do you keep the prompts and skills you use in Warp?

Warp reads AGENTS.md and .agents/skills/, so the same files can serve Warp, Claude Code, Codex and the rest. Kipt keeps prompts and Agent Skills as Markdown in a private GitHub repo you own, with {{name}} blanks and Cmd+K search, and installs every skill with one command: npx skills add https://github.com/<you>/<repo>.git --all -g.