Quick Verdict
Warp is no longer accurately summarized as a Rust-based terminal with an AI command generator. Its current product is an open agentic development environment: Warp Terminal provides the command line and local agent surface; Warp Agent can plan, inspect code, edit files, and run commands; third-party harnesses such as Claude Code, Codex, and OpenCode run with richer input, notifications, and review; and Oz moves work into cloud agents, parallel orchestration, schedules, and integration-triggered automation. This breadth is valuable to terminal-first engineers and platform teams, but it makes permission and cost configuration more important than in a conventional terminal.
Best For
Warp Terminal officially supports macOS 10.14+, Windows 10/11, and Linux through deb, rpm, Arch, and AppImage packages, with x64 and broad ARM64 availability. It works with common shells and supported CLI agents. Oz cloud agents can be launched from the web, CLI, API/SDK, schedules, or integrations and can run on Warp-hosted or enterprise self-hosted infrastructure. Experimental Computer Use operates only in isolated cloud sandboxes; it does not give a local terminal agent direct desktop control.
Key Features
- Modern terminal UX with blocks, editor-like input, history, workflows, and shell integration.
- Local agents with planning, full terminal use, code edits, and indexed codebase context.
- Third-party CLI agent hosting plus Oz orchestration for local and cloud work.
- File tree, code editor, LSP, diffs, and interactive code review.
- MCP, Skills, Rules, schedules, and integrations including Slack, Linear, and GitHub Actions.
Use Cases
Agent Profiles separately govern reading files, applying diffs, creating plans, executing commands, interacting with running commands, and asking questions. Levels include Agent Decides, Always Ask, Always Allow, and Never. A command denylist takes precedence over the allowlist, with risky examples such as rm, curl, wget, and eval; MCP servers have separate rules. Codebase Context indexes Git-tracked files, and Warp states that source code is not stored on its servers for that index. AI use on Free requires telemetry; paid personal plans may opt out, while Business and Enterprise are covered by Zero Data Retention. Secret Redaction is always applied to AI interactions, but users should still withhold credentials.
Pricing
| Plan | Monthly starting price | Main allowance |
|---|---|---|
| Free | $0 | Core terminal, BYO inference, limited cloud agents, Drive, and storage |
| Build | $20 | 1,500 credits, full Warp Agent, extended cloud agents and indexing |
| Max | $200 | 18,000 credits |
| Business | $50/user | 1,500 credits per seat, metrics, data controls, SAML SSO |
| Enterprise | Custom | Governance, BYO LLM, self-hosted agents, custom indexing and support |
Annual prices shown by Warp are roughly 10% lower for Build, Max, and Business. These are pay-as-you-go starting points; reload credits and model usage affect the final bill.
Pros
Create a cautious profile for sensitive repositories with Always Ask and only read-only commands in the allowlist. Ask for an editable plan, then inspect code changes hunk by hunk, leave review comments, and request revisions. “Run until completion” should be reserved for isolated branches and least-privilege environments. Cloud agents need additional repository, secret, network, and integration scoping. Their output should return through normal pull requests, CI, and human approval.
Cons
Warp fits terminal development, bug investigation, repository refactors, first-pass PR review, incident response, and coordinating several agent harnesses in one workspace. Compare Cursor for editor-first work, Kiro for spec-driven engineering, and Claude Code for a focused terminal agent.
Alternatives
| Tool | Main surface | Permission and review model | Best fit |
|---|---|---|---|
| Warp | Terminal, code panels, and Oz cloud | Profiles, allow/deny lists, diff review | Multi-agent terminal and cloud orchestration |
| Claude Code | Terminal | Tool permissions and project rules | Focused CLI agent work |
| Cursor | Editor | Agent diffs and IDE controls | Visual editor development |
| Kiro | IDE, CLI, and web | Supervised review and protected paths | Spec-first engineering |
FAQ
Is Warp still a terminal?
Yes, but it now includes code editing, agents, review, and Oz cloud orchestration around the terminal core.
Does it support Windows and Linux?
Yes. Warp provides Windows 10/11, major Linux package formats, and macOS builds.
Can it run Claude Code?
Yes. Warp supports Claude Code, Codex, OpenCode, and other CLI agents with enhanced UX.
Can Free users disable telemetry and keep AI?
No. AI on Free requires telemetry; paid users can opt out, and Business/Enterprise receive ZDR.
Does codebase indexing upload source code?
Warp says the semantic index does not store source code on its servers, though agent requests still use necessary context.
Will the agent auto-run dangerous commands?
That depends on the profile. The denylist prompts for common risky commands, and users should not loosen it casually.
Bottom Line
Warp unifies a terminal, code review, third-party CLI agents, and cloud orchestration, with more granular controls than a single auto-execute toggle. Its client is open source under AGPL v3, and users can inspect network telemetry and control collection according to plan. The downside is complexity: product surfaces and usage pricing are broader, Free AI requires telemetry, cloud and local data boundaries differ, and a permissive profile can still authorize dangerous actions.