Warp Terminal
Warp is an agentic development environment (ADE) that combines a modern Rust-built terminal, an integrated code editor, and AI agents into a single application. Named a TIME Best Invention of 2025,1) Warp enables developers to run multiple AI agents, review their work, and ship code from prompt to production.
ai_agent terminal ide coding agent_mode rust multi_agent
Overview
Warp 2.0 reimagines the terminal as an agentic development environment.2) Rather than a passive command-line interface, Warp provides an integrated workspace where AI agents can plan, code, execute terminal commands, review output, and iterate – all under developer supervision. Its Oz agent ranks #1 on Terminal-Bench (52%) and top-5 on SWE-bench Verified (71%), with unique Full Terminal Use capabilities that no other product offers.3)
Key Features
Oz Agent – State-of-the-art coding agent with Full Terminal Use and Computer Use capabilities
Full Terminal Use – Agent interacts with REPLs, debuggers, and full-screen apps like top – unique to Warp
Multi-Agent Support – Run and monitor multiple agents in parallel with human-in-the-loop oversight
Code Editor – Built-in editor with LSP support, file tree, inline diffs for agent output review
/plan Command – Spec-driven development where agent generates implementation plans for review before execution
Interactive Code Review – Review agent code like a teammate's, with inline comments the agent addresses
Drive – Centralized knowledge and context storage for agents and teammates
Modern Terminal – Rust-native, no Electron; modern text editing, block-based output, secret redaction
4)
Multi-Model – Instant access to latest models from OpenAI, Anthropic, and Google
Architecture
graph TD
A[Developer] --> B[Warp ADE]
B --> C[Terminal Layer]
B --> D[Code Editor]
B --> E[Agent System]
C --> F[Rust Shell Engine]
C --> G[Block-Based Output]
C --> H[Full Terminal Use]
D --> I[LSP Integration]
D --> J[File Tree]
D --> K[Inline Diffs]
E --> L[Oz Agent]
L --> M[Planning - /plan]
L --> N[Code Generation]
L --> O[Terminal Execution]
L --> P[Code Review]
E --> Q{Model Selection}
Q --> R[OpenAI]
Q --> S[Anthropic]
Q --> T[Google]
E --> U[MCP Context]
U --> V[Slack]
U --> W[Linear]
U --> X[Figma]
U --> Y[Sentry]
B --> Z[Drive - Shared Context]
E --> AA[Multi-Agent Threads]
Agent Workflows
Warp agents follow a structured workflow from prompt to production:
Prompt – Developer describes the task via Universal Input with @-mentions for files, images, URLs
Plan – Agent generates an implementation plan (/plan command) for review and editing
Execute – Agent writes code, runs terminal commands, interacts with REPLs and debuggers
Verify – Agent uses Computer Use to inspect logs, servers, and file systems
Review – Developer reviews diffs inline, leaves comments; agent addresses feedback
Ship – Changes committed and pushed; plans can be attached to PRs
Integrations
Slack – Trigger agents from Slack messages
Linear – Assign tasks to agents from Linear issues
GitHub Actions – CI/CD pipeline integration
MCP (Model Context Protocol) – External context from Slack, Linear, Figma, Sentry
WARP.md5) – Compatible with agents.md and claude.md for controlling agent behavior
Benchmarks
| Benchmark | Score | Rank |
| Terminal-Bench | 52% | #1 |
| SWE-bench Verified | 71% | Top 5 |
Installation
# Download from warp.dev
# Available for macOS and Linux
# Rust-native, no Electron
# Source: https://github.com/warpdotdev/Warp
# After installation, Warp replaces your default terminal
# Sign in for AI features
# Free tier available with pro features for teams
See Also
-
-
-
Plandex – AI coding agent with plan/apply workflow
References