Warp terminal in 2026: features, agents, pricing, privacy

Warp is a free terminal for macOS, Windows and Linux with a coding agent, the Warp Agent, built in. It groups each command and its output into a Block, gives you an editor-style input line, and can host other CLI agents such as Claude Code. The client has been open source under AGPL v3 since April 2026, and an account is optional. The terminal costs nothing; agent usage is the paid part, starting with the Build plan at $20 a month, which includes $20 of usage at API rates.

This is a guide built from Warp's documentation and pricing pages, checked on 9 November 2026. It is not a hands-on review: I have not run Warp on my own projects, so there are no timings, screenshots or a verdict here, only what Warp documents.

Warp at a glance

What it is	A terminal with a built-in coding agent (Warp Agent)
Platforms	macOS 10.14+ (Intel and Apple silicon); Windows 10 version 1903+ and Windows Server 2022+ (x64, ARM64); Linux with glibc 2.31+ (.deb, .rpm, Arch, AppImage; x64, ARM64)
Shells	bash, zsh, fish and PowerShell on macOS and Linux; PowerShell 7 (default), PowerShell 5, WSL2 and Git Bash on Windows
Licence	Client open source under AGPL v3 since 27 April 2026
Account	Optional
Models	OpenAI, Anthropic, Google, xAI and open-weight models, plus Auto routing
Price	Terminal free; Build $20/month, Max $200/month, Business $50/user/month

On Windows, Command Prompt (cmd.exe) is not supported as a session shell.

What changed in 2026

If you last looked at Warp a year or two ago, three things are different.

• The client is open source. The changelog entry for 27 April 2026 reads "Warp is now open source", and the installation docs explain how to build it from the warpdotdev/warp repository. Self-built binaries do not auto-update. The same release made Warp configurable through a TOML settings file.
• Usage is priced in dollars. Paid self-serve plans include a dollar amount of agent usage each month, charged at public API rates, rather than a fixed number of requests.
• The focus has widened. Warp's site now calls the product "the agentic development environment" and leads with Warp Factories, a separate product for running fleets of coding agents, with its own plans on the pricing page. The terminal is still the free entry point, and the Factories FAQ says nobody has to use the terminal to use Factories.

The terminal

These features come with every plan, including Free.

• Blocks. Each command and its output form one Block. You can copy the command or just the output, jump to the start of the output, re-run the command, bookmark it, search and filter within it, and share it with formatting kept.
• An editor-style input line. You can click to place the cursor, select by word and edit across lines as in a code editor. Warp's docs say this editor "replaces the shell's editor capabilities", and Warp ships its own Vim keybindings, so your shell's own vi mode is not what you get.
• Suggestions and search. Autosuggestions from your shell history, fuzzy tab completions for commands, options and paths, command search on Ctrl+R, and a command palette for finding workflows, notebooks and prompts.
• Warp Drive. Saved workflows, notebooks and agent plans you can share with a team. The Free plan has limited Warp Drive and collaboration features.
• Offline use. An internet connection is needed on first launch only. AI and collaboration features need one after that.

The Warp Agent

Warp keeps the two jobs apart. Terminal mode, the default in a new tab, is a plain prompt for commands with agent controls hidden. Agent mode, opened with /agent or /new, is a conversation view with a model selector, voice input, image attachments and a panel for past conversations. /fork branches a conversation and /compact summarises it to free up context.

The documented capabilities:

• Model choice. Models from OpenAI, Anthropic, Google and xAI, plus open-weight models such as GLM, Kimi, Qwen and DeepSeek hosted on Fireworks AI. Four Auto modes route each request for you: Responsive (quality and speed), Cost-efficient, Genius (scales up for hard problems and /plan sessions) and Open-weights.
• Codebase context. Warp indexes Git-tracked code and respects .gitignore and its own .warpindexingignore. The docs say "No code is stored on Warp servers." The Free plan indexes up to 5,000 files per repository; paid plans allow more.
• Planning. /plan writes a plan in a rich text editor. You can edit it, every agent revision becomes a new version you can compare or restore, and you can run the whole plan or one section at a time.
• Rules, skills and MCP. Global rules apply everywhere; project rules live in the repo, and Warp reads AGENTS.md by default (the older WARP.md still works and wins if both exist). Skills package reusable task instructions, and MCP servers connect the agent to outside tools.
• Profiles and permissions. A profile sets the agent's autonomy per action (edit code, read files, run commands and so on), its base and planning models, and regex allowlists and denylists for commands and MCP servers. "Run until completion" auto-approves a task's actions and, by default, also bypasses the denylist, so turn that option off if you rely on it.
• Full Terminal Use. The agent can work inside interactive programs: database shells such as psql, editors, Python REPLs, debuggers, dev servers and cloud CLIs such as aws. You can take over at any point and hand control back.
• Agent CLI. The same agent runs as a warp command in any terminal emulator, including over SSH. The Free plan includes it.

The AGENTS.md default is useful if you already keep one instructions file for several tools. Cursor vs Claude Code shows how to point Cursor and Claude Code at a single AGENTS.md; Warp's agent would read the same file.

Running Claude Code and other CLI agents in Warp

You do not have to choose between the Warp Agent and a CLI agent you already pay for. Warp auto-detects Claude Code, Codex, OpenCode, Gemini CLI, Copilot CLI, Cursor CLI, Amp, Goose, Kiro CLI and several others when you start them, and adds:

• a rich input editor for long prompts, opened with Ctrl-G
• code review comments you send back to the agent from Warp's review panel
• selecting code and attaching it as context
• vertical tabs that show each agent's live status
• Remote Control, to monitor and steer the session from another device
• notifications when the agent needs you, for Claude Code, Codex and OpenCode (and Antigravity, whose CLI emits structured notifications)

For Claude Code, notifications need a small Warp plugin, which in turn needs jq installed. When you run Claude Code without it, Warp shows a chip that installs it in one click. You can also add it yourself from inside Claude Code:

/plugin marketplace add warpdotdev/claude-code-warp
/plugin install warp@claude-code-warp

Then restart Claude Code or run /reload-plugins. Claude Code sessions do not appear in Warp's own conversation panels, and Warp's docs leave sign-in to Claude Code's own setup guide. Claude Code itself needs a paid Claude plan or an Anthropic API key either way; Cursor vs Claude Code covers those plans.

AI coding agentClaude CodeAnthropic's agentic coding tool for the terminal, IDE, desktop and web; included in paid Claude plans.

What Warp costs

Prices from Warp's pricing page on 9 November 2026:

Plan	Monthly	Billed annually	Agent usage included
Free	$0	$0	None; buy usage (at a premium on Free) or bring your own API key
Build	$20	$18/month	$20 a month at API rates
Max	$200	$180/month	$200 a month at API rates
Business (up to 25 seats)	$50/user	$45/user/month	$20 per seat a month at API rates
Enterprise	Custom	Custom	Custom pools

Model	freemium
Paid plans from	$20/month (Build), or $18/month billed annually
Notes	Free: the terminal, no bundled agent usage (buy usage at a premium or bring your own API key). Build includes $20 of agent usage a month at API rates, Max ($200/month) $200, Business $50/user/month. Included usage resets monthly.
Checked	October 29, 2026

The details that decide your real bill:

• Included usage resets monthly, even on annual plans, and unused usage does not roll over. Your exact allowance and reset date are under Settings > Billing and usage.
• Purchased usage rolls over until it expires 12 months after purchase. On Free, purchases cost more than on paid plans and are one-off only; paid plans can turn on auto-reload.
• A spend limit applies to purchases. One-off purchases and auto-reloads share a team-wide monthly limit that defaults to $200 and can be changed by an admin.
• What uses it. Usage covers inference, plus compute and platform charges for cloud agent runs. Plain shell commands and non-AI terminal features never use it.
• Bring your own key. Personal Anthropic, OpenAI or Google keys work on Free and eligible paid plans, for individuals and organisations of 10 or fewer employees, and only for local agent runs. Your provider bills that inference. Warp's Auto models still use Warp-provided inference and your Warp usage, even with your own keys set.
• Refunds. Monthly plans are refundable within 24 hours of the charge if no usage was consumed that cycle; annual plans within 15 days. Purchased usage is refundable only if none of it was used.

Warp Factories, the cloud-agent product, is priced separately on the same page and is not covered here.

Privacy and telemetry

Warp publishes the full list of telemetry events it sends, shows live traffic in a built-in Network Log, and lets you opt out in Settings > Privacy by turning off "Help improve Warp", "Send crash reports" or both. The points worth reading before you use the agent on private code:

• What "Help improve Warp" covers. High-level product analytics, and "AI interactions and console inputs that power our AI features". The privacy page does not say whether the setting is on by default, so check it after installing.
• On the Free plan, AI needs telemetry. The docs state: "Telemetry must be enabled to use AI features on the Free plan."
• Zero data retention. Business and Enterprise plans are covered by a zero data retention agreement under which no AI interaction or console data is collected. Separately, Warp says it has zero data retention agreements with all its contracted model providers, so they do not retain or train on customer AI data.
• Secret Redaction is off by default. When enabled, it masks text matching your patterns before AI requests are sent.

You can turn Warp's AI features off entirely in Settings > Agents > Warp Agent.

Who Warp suits

Based on what Warp documents, not on my own use:

• You want a modern terminal and no AI. The terminal features are free, need no account, and AI can be switched off globally.
• You already use Claude Code or Codex. Warp adds notifications, a bigger prompt editor and inline review around them without asking you to switch agents or pay Warp for usage.
• You want an agent built into the terminal itself, with model choice across providers, plans you can edit and a fixed monthly dollar allowance. Build at $20 is the entry point, or Free with your own API key if you qualify.
• You care about data handling on a team. Business adds zero data retention for AI interactions and console data.

It is a weaker fit if you depend on cmd.exe on Windows (not supported), if you rely on your shell's own line editor or vi mode (Warp's editor replaces it), or if you want AI on the Free plan with telemetry turned off.

TerminalWarpOpen-source terminal with a built-in coding agent that also hosts CLI agents such as Claude Code.

To try it, download Warp; the terminal is free and needs no account.

Frequently asked questions

Is Warp free?

Yes. The terminal is free on macOS, Windows and Linux, and a Warp account is optional. The Free plan has no bundled agent usage: you can buy usage, priced at a premium on Free, or bring your own OpenAI, Anthropic or Google API key. Warp's docs say telemetry must be enabled to use AI features on the Free plan.

How much does Warp cost?

As of 9 November 2026: Build is $20 a month ($18 billed annually) with $20 of agent usage a month at API rates, Max is $200 a month ($180 annually) with $200 of usage, and Business is $50 per user a month ($45 annually) with $20 of usage per seat. Included usage resets monthly and does not roll over.

Is Warp open source?

Yes, the client is. Warp's changelog entry for 27 April 2026 says "Warp is now open source", and the docs explain how to build the client from the warpdotdev/warp repository on GitHub, under the AGPL v3 licence.

Can you run Claude Code inside Warp?

Yes. Warp auto-detects Claude Code, Codex, OpenCode and over a dozen other CLI agents and adds a rich input editor, inline code review comments and remote control. Notifications for Claude Code need Warp's plugin, which Warp offers to install with one click.

Does Warp work offline?

Mostly. Warp needs an internet connection the first time you open it. After that the terminal runs offline, but AI and real-time collaboration features need a connection.