Skip to content

Warp

Open-source, AI-first terminal with a built-in coding agent and cloud agents, plus Warp Factories for running agent pipelines.

Reviewed by Andrew Miller · Updated

Visit Warp Free plan

Our rating

4.0/5

Warp puts a capable agent where developers already type commands, but its free tier pays for AI with telemetry and its own agent is shallower than the Claude Code and Codex harnesses it now hosts.

What it is

Warp is a modern terminal for macOS, Linux and Windows that treats commands and their output as blocks you can select, share and ask questions about. Built into it is the Warp Agent (called Oz until August 2026), which can run commands, edit code across a project and hand longer jobs to cloud agents. It works with models including Claude Opus 5, Fable 5, GPT-5.6, Gemini 3.1 Pro, GLM 5.2 and Kimi K3, and you can switch models mid-task.

In April 2026 Warp open-sourced its terminal client on GitHub, with OpenAI as founding sponsor; the repository passed 26,000 stars within hours. In August 2026 it launched Warp Factories, a separate product for teams that defines agent pipelines (from triage to code review) as files in the repo and can also run the Claude Code and Codex harnesses.

Our review

Warp's case is location. The agent sits in the terminal, so you describe a task, it runs the commands, reads the output and fixes what breaks, without switching to an editor or a chat window. The client has been open source under AGPL-3.0 since April 2026, and the GitHub repository has about 65,600 stars, which settles the old complaint about a closed terminal that demanded a login.

The free tier has a catch the pricing page does not spell out. According to Warp's privacy docs, Free users must leave telemetry on to use AI features, while paid users can turn it off and keep them. Business and Enterprise sit under a zero-data-retention agreement, and the pricing page says customer data is not used for training. Warp's general privacy notice, updated on 18 August 2026, still reserves the right to use personal information for training models, so a security review should get the contract wording rather than rely on the marketing page.

Pricing is in dollars rather than credits. Build at $20 a month includes $20 of agent usage at API rates, Max at $200 includes ten times that, and Business at $50 a seat includes $20 per seat. That is honest, and it also means a $20 plan buys exactly $20 of model calls, less headroom for heavy agent work than a $20 Claude or ChatGPT subscription gives.

The product has spread sideways. In one year Warp open-sourced the terminal, renamed its agent from Oz to Warp Agent and launched Factories as a second product with its own plans, where pay-as-you-go usage runs 20% above API rates. Factories can run the Claude Code and Codex harnesses directly, which quietly concedes that many teams will use Warp as the shell and orchestration layer around someone else's agent.

Where it earns its place

  • A DevOps engineer on a five-person platform team

    Debug failing deploys and write one-off scripts without leaving the shell

    The agent reads the error block, proposes and runs the fix and keeps the session as a shareable block. Build at $18 a month billed yearly covers $20 of usage; long agent runs need top-ups at API rates.

  • The engineering manager of a 20-developer team at a fintech

    Roll out an AI terminal that passes a security review

    Business at $45 per seat billed yearly is $10,800 a year for 20 seats, with SAML SSO, admin-enforced data controls and zero data retention. Legal should reconcile the privacy notice's training clause with the contract before sign-off.

  • A staff engineer at a 60-person startup

    Run triage-to-review agent pipelines defined in the repo

    Factories Build at $20 a month runs the Claude Code and Codex harnesses at API rates with factory-as-code in GitHub; usage for inference, compute and agents is billed separately on top.

Before you commit

Setup
Install the app for macOS, Linux or Windows; AI features on Free need telemetry on, and paid plans unlock the privacy toggle.
Integrations
Models from several providers with mid-task switching; Business adds your own API keys and custom inference endpoints, Enterprise routes inference through your own cloud.
Data and privacy
SOC 2; zero data retention for Business and Enterprise and with contracted model providers per the pricing page. Secret redaction is off by default and must be switched on.
Cost at scale
Business is $45 per seat a month billed yearly with $20 of usage per seat; teams over 25 seats need Enterprise with a custom shared pool.
Limits
Cloud agent limits are not quantified on the pricing page; overage rates beyond API pricing are not published.

What's good

  • Open-source AGPL-3.0 client with block-based output
  • Agent runs commands and fixes errors in place
  • Included usage billed in dollars at API rates
  • SAML SSO and bring-your-own keys on Business
  • Factories defines agent pipelines as files in the repo

What's not

  • Free tier requires telemetry to use AI
  • Privacy notice and pricing page disagree on training use
  • Built-in agent less capable than Claude Code or Codex for long tasks
  • Business capped at 25 seats
  • Factories billed separately, 20% above API rates on pay-as-you-go

Best for

Terminal-heavy developers and DevOps engineers who want AI help with commands, scripts and quick code fixes.

Alternatives

Warp: common questions

Is Warp worth it?
Warp puts a capable agent where developers already type commands, but its free tier pays for AI with telemetry and its own agent is shallower than the Claude Code and Codex harnesses it now hosts. We rate it 4.0 out of 5.
How much does Warp cost?
Free terminal with bring-your-own inference and limited cloud agents; Build $20/month ($18 yearly) with $20 of agent usage at API rates; Max $200/month ($180 yearly); Business $50/user/month ($45 yearly); Enterprise custom. Warp Factories is billed separately by usage.
What are the downsides of Warp?
Free tier requires telemetry to use AI; privacy notice and pricing page disagree on training use; built-in agent less capable than Claude Code or Codex for long tasks; business capped at 25 seats; factories billed separately, 20% above API rates on pay-as-you-go.
Who is Warp best for?
Terminal-heavy developers and DevOps engineers who want AI help with commands, scripts and quick code fixes.
How much does Warp cost for a team or in production?
Business is $45 per seat a month billed yearly with $20 of usage per seat; teams over 25 seats need Enterprise with a custom shared pool.
Is Warp safe for company data?
SOC 2; zero data retention for Business and Enterprise and with contracted model providers per the pricing page. Secret redaction is off by default and must be switched on.
What are the best Warp alternatives?
The closest alternatives we have reviewed are Claude Code, OpenAI Codex, Amp and Factory (Droids).

Answers checked Oct 22, 2026.

Make Warp? Show this rating on your siteWarp rated 4.0 out of 5 on ReviewRamble