How to Use Warp: The Beginner’s Guide (First Commands, Blocks, AI Search, and Your First Agent)
CodexFuse Team · 2026-10-29T10:03:39.209Z · 11 min read min read
Warp is the AI-powered terminal for developers: Blocks, AI command search, and an agent mode in one window. This beginner’s guide covers install, first commands, and your first agent prompt.
Learning how to use Warp takes about ten minutes, and most of that is installing it. Warp is the AI-powered terminal that groups every command and its output into movable Blocks, suggests commands as you type, and puts a coding agent one keystroke away — while still running your plain shell commands exactly like the terminal you already know. This guide walks through install, Blocks, AI command search, and your first agent prompt, so you can decide in one session whether Warp earns a place in your daily workflow. (Warp's own 10-minute quickstart, refreshed 9 days before this writing, is the reference I verified every step against.)
What is Warp, exactly?
Warp is a terminal with an AI agent built into it — the company now calls it an Agentic Development Environment (ADE), "a space where your primary mode of interaction is through prompts to an agent" (Warp docs, "Welcome to Warp," October 2026). Two facts matter more than the label. First, Warp still runs plain commands exactly like any terminal: type ls or git status and it executes them normally. Second, when you type natural language — like "describe my open git changes" — Warp interprets it as a prompt and launches its agent, which gathers repo context, runs whatever commands are needed, and returns a structured answer. You never context-switch between apps: terminal commands and agent work live in the same window. If you already follow the Claude Code beginner's guide, Warp will feel familiar — its quickstart explicitly notes that it also works with third-party CLI agents like Claude Code and Codex.
How do I install Warp?
Download Warp from warp.dev and run the installer for your platform, or on macOS install it with Homebrew: brew install --cask warp. Warp supports macOS, Windows, and Linux (Warp quickstart, October 2026) — note that older guides claiming Windows is still on a waitlist are stale; the current docs list all three platforms. When you launch Warp you get a terminal session ready for input, and signing up for an account is optional: the quickstart says you can "skip and start using Warp immediately." The honest catch, confirmed across independent reviews, is that the AI features need an account — plain terminal use does not. So install it, skip the signup, and try the terminal first; add the account only when you want the agent.
How do Blocks work in Warp?
Blocks are Warp's signature idea, and they change how a terminal feels. Every command you run and its output are grouped into one contained unit — a Block — that you can copy, search, filter, and navigate independently (Warp quickstart, October 2026). Instead of an endless scroll of text, your session becomes a stack of discrete cards.
Run ls -la and you will see the difference immediately. Then try the three Block moves that matter: click a Block to select it, then press ⌘C on macOS or Ctrl+Shift+C on Windows/Linux to copy its output; use ⌘↑/⌘↓ (or Ctrl+↑/Ctrl+↓) to jump between Blocks; and click the filter icon on any Block to search within its output. Long outputs can be expanded or collapsed, a command can be re-run from its Block, and a useful Block can even be shared as a link. If you have ever scrolled up through 500 lines of logs hunting for one error, you already understand why Blocks exist.
How do I use AI command search in Warp?
Warp's AI command search turns a plain-English description into the command you meant to type. Invoke it from the input line, describe what you want — "find all Python files modified this week" — and Warp proposes the command, which you accept or edit before it runs. (The widely used Warp Terminal Cheat Sheet lists it at Cmd+Shift+K; verify the binding in your build, since shortcuts are customizable.)
Three quieter helpers do as much daily work as the flashy AI search. Warp shows faded autosuggestions from your shell history as you type — press → to accept one instantly. Tab completions cover commands, flags, and file paths, going beyond what your shell offers natively. And if you mistype a command, Warp detects the error and suggests a correction you can accept or dismiss. None of these require the agent, an account, or any AI credits — they work in plain terminal mode from the first minute.
What is Warp Agent Mode?
Agent Mode is the dedicated conversation view where you talk to the Warp Agent in natural language instead of typing commands. Press ⌘↩ on macOS or Ctrl+Shift+Enter on Windows/Linux to open it, then prompt it like a colleague: the quickstart's own example is "Explain the architecture of this project." The agent reads your codebase, understands its structure, and answers with context — then you can ask follow-ups, have it write or refactor code, debug errors, or run commands on your behalf, all in the same conversation.
| Terminal Mode | Agent Mode | |
|---|---|---|
| You type | Shell commands | Natural-language prompts |
| Warp does | Runs commands exactly as typed | Gathers context, runs commands, returns structured answers |
| Best for | Commands you already know | Exploring code, debugging, multi-step tasks |
| Account needed | No | Yes |
Two deeper agent features are worth knowing on day one. Full Terminal Use lets you tag the agent into a command that is already running — start psql or your dev server yourself, then bring the agent in with the "Use Agent" button in the terminal footer or Cmd+I; it sees the same terminal buffer you do and proposes actions inside the session, and you can take control back at any time with the Takeover control (Warp docs, "Full Terminal Use," October 2026). And /plan mode turns a goal into a step-by-step plan before anything runs, which is how experienced users keep agents from improvising destructively.
On the integrations side, Warp's enterprise docs list MCP servers — including Linear, Sentry, Figma, and GitHub — so the agent can pull in issues, errors, and designs as context. If the term is new to you, our Model Context Protocol (MCP) explainer covers what those connectors are. Combined with cloud agents that run PR reviews and dependency updates in the background, this is the "development environment" half of the ADE label: the terminal is becoming the place where the work gets delegated, not just typed.
How much does Warp cost?
Warp has a free plan and paid plans, and the pricing has been rewritten twice in the last year — so treat any number below as a snapshot and verify the current page at warp.dev/pricing before you decide. The last full simplification, in December 2025, replaced the old Pro/Turbo tiers with a credit system: every AI action — command generation, agent turns, codebase indexing queries — consumes credits (dev.to "Warp Terminal Review 2026").
| Plan | Price | AI allowance | Notes |
|---|---|---|---|
| Free | $0 | Small monthly credit allowance | All terminal features included; enough to evaluate the AI, not to live on it |
| Build | $20/month | 1,500 credits/month | Bring-your-own-key (BYOK) supported — route requests through your own provider key so Warp credits are not consumed |
| Business | $50/user/month | 1,500 credits/seat | SSO, zero data retention, team controls |
The BYOK detail is the one that matters for heavy users: on a paid plan you can point Warp at your own Anthropic or OpenAI key, and those requests bill through your provider instead of burning Warp credits. Newer coverage also notes a "bring your own AI inference" direction on the free tier, including ChatGPT sign-in since September 2026 — another sign the model is still moving. The stable truth underneath the churn: the terminal itself is free and unlimited, and you only pay when the AI does work for you.
Warp vs iTerm2: should you switch?
Switch to Warp if you live in the terminal and want AI help inline; stay with iTerm2 if you want a minimal, fast, endlessly scriptable terminal and never think about agents. iTerm2 is free, mature, and sips memory; Warp is heavier — one 2026 review measured roughly 380 MB idle, scaling to 500–700 MB under load — and its Blocks and session model fight with tmux, so tmux power users lose most of what makes Warp special (AI Leap Tools review, May 2026). Warp's case is the whole package: Blocks that make output navigable, AI command search for the commands you half-remember, and an agent that can read your repo and run the fix. iTerm2's case is that it does none of that and does everything else brilliantly. Try Warp for a week of real work before uninstalling anything; the terminal you have configured for years is not dethroned by a demo.
If your reason for looking at Warp is the AI-coding wave rather than the terminal itself, compare it with the editor side too — our Cursor AI beginner's guide covers the agentic IDE lane, while Warp owns the terminal lane. And when your agent work starts involving containers, the Docker beginner's guide picks up where the terminal leaves off.
Your first 30 minutes with Warp
Do this in order and you will have touched every important feature. It is the fastest honest evaluation of any terminal.
- Install and skip the account.
brew install --cask warpon macOS, or the installer from warp.dev. Launch it, skip signup, and confirm plain commands work. - Run three commands and feel Blocks.
pwd,ls -la,git status— three Blocks. Click one, copy its output, jump between them with⌘↑/⌘↓, filter one for a word. - Try the input editor. Type three
echolines separated byShift+Enter, then pressEnteronce — all three run, each in its own Block. - Use AI command search. Describe a command in plain English and accept the suggestion. Then mistype something on purpose and accept Warp's correction.
- Ask your first agent question. Open a project folder, press
⌘↩, and prompt "Explain the architecture of this project." Ask one follow-up. - Save a Workflow. Turn a command you run constantly into a reusable Workflow so you never retype it — this is the seed of Warp Drive, the shared command library teams use.
One more freshness note for 2026: Warp's client source code was released under the AGPL license in April 2026 (The Rundown's Warp 2.0 review), answering the old "proprietary black box" objection for the terminal itself — hosted agents and model inference remain commercial services. That, plus the agent-first reframe of the docs this month, is why every independent beginner guide you find is at least slightly out of date, and why this one was verified against the vendor docs dated October 2026.
Sources
- Warp, "Quickstart" (docs.warp.dev, updated ~October 20, 2026) — install methods, Blocks, input editor, autosuggestions, Agent Mode keybindings, third-party CLI agents, cloud agents.
- Warp, "Welcome to Warp" (docs.warp.dev, updated ~October 21, 2026) — Agentic Development Environment framing; plain commands run normally; natural language launches the agent.
- Warp, "Full Terminal Use" (docs.warp.dev, updated ~October 20, 2026) — tagging the agent into a running PTY session, Takeover control.
- Warp, enterprise developer docs (docs.warp.dev, updated ~October 21, 2026) — MCP servers (Linear, Sentry, Figma, GitHub), /plan, cloud agents.
- Jovan Chan, "Warp Terminal Review 2026: Open-Source ADE, the $20 Build Plan, and Who Should Actually Pay For It" (dev.to, ~May 2026) — December 2025 pricing simplification, credit allowances, BYOK economics.
- Gabriel Anhaia, "Hermes IDE vs Cursor vs Warp vs Zed" (dev.to, ~October 12, 2026) — free-tier "bring your own AI inference," ChatGPT sign-in since September 2026, AGPL client.
- MeshWorld, "Warp Terminal Cheat Sheet" (meshworld.in, ~May 2026) — shortcut reference including AI command search.
- The Rundown AI, "Warp 2.0 Review: Features, Pricing & Alternatives" (therundown.ai, ~August 2026) — AGPL client release April 2026, zero-data-retention terms, plan pricing.
- AI Leap Tools, "Warp Terminal Review 2026" (tools.aileapers.com, May 2026, verified against warp.dev) — memory footprint, tmux friction, account requirement notes.
Frequently asked questions
Is Warp free?
Yes, for individual use there is a free plan that includes every terminal feature — Blocks, the input editor, autosuggestions — plus a limited monthly allowance of AI credits to try the agent. Heavy AI use needs a paid plan (Build is $20/month at the last check), and you can also bring your own provider key. Verify current numbers at warp.dev/pricing, because Warp rewrote its pricing in December 2025 and keeps adjusting it.
Does Warp work on Windows?
Yes. The current Warp docs list macOS, Windows, and Linux as supported platforms. Older articles that say Windows is on a waitlist are out of date — that was true in early 2025 and has not been true for a long time.
Can I use Warp without an account?
For the terminal itself, yes — the quickstart explicitly lets you skip signup and start using Warp immediately. The AI features (command search, the agent) require an account, which is the one thing the free signup buys you beyond syncing.
Does Warp replace my shell?
No. Warp is a terminal emulator, not a shell — it runs your existing bash, zsh, or fish underneath, so your aliases, configs, and scripts transfer over untouched. Your commands behave identically; only the interface around them is new.
Is Warp open source?
Partly. Warp released its client source code under the AGPL license in April 2026, so the terminal application itself is open. The hosted side — model inference, cloud agents, collaboration services — remains commercial and follows Warp's paid plans.