What is Warp?
Warp is an AI terminal for developers that runs AI coding agents directly inside the command line environment. It produces shippable code features from natural language prompts and handles everything from writing and editing code to reviewing agent output. Warp combines a redesigned command-line interface with built-in support for multiple AI coding agents, including Claude Code, Codex, Gemini CLI, and Opencode. It also includes a knowledge store for centralizing context and a cloud agent orchestration platform called Oz. Warp runs on Mac, Windows, and Linux, and is open-source.
Features & Benefits
- Block-Based Navigation: Navigate terminal input and output in discrete, selectable blocks for faster review and editing of commands and agent responses.
- IDE-Like Input Editor: Write and edit multi-line commands and long prompts directly in the terminal with text-editor-style controls.
- Built-In AI Agent Support: Run AI coding agents natively inside the AI terminal, with support for Claude Code, Codex, Gemini CLI, Opencode, and Warp’s own agent.
- Agent Autonomy Controls: Set exactly how much autonomy each agent has, from requiring approval on every step to allowing fully autonomous operation, configurable per user or across the whole team.
- Prompt Context Attachments: Reference files and images directly in prompts to give agents accurate context for each task.
- Voice Transcription Input: Dictate prompts using voice transcription via Wispr Flow integration to compose agent instructions hands-free.
- Interactive Code Review: Review, comment on, and send agent-generated code changes back to the agent for revision without leaving the terminal.
- File Tree and File Editor: Browse project file structure, attach paths as context, and open and edit files inside the terminal without switching to a separate editor.
- Vertical Tabs: Organize multiple terminal sessions as vertical tabs with configurable metadata including git branch, worktrees, and pull request details.
- MCP Server Support: Connect the AI terminal to external data sources via MCP servers, such as Figma for design context or Sentry for exception data, so agents can interact with live external information.
- Knowledge Store (Warp Drive): Store plans, command history, and team decisions in a shared knowledge base to improve agent accuracy across sessions.
- Notebooks, Workflows, and Prompts: Save and share reusable, parameterized processes, including onboarding instructions, command sequences, and prompt templates, so teams can standardize how common tasks get run.
- AI Context Rules: Define rules inside Warp Drive to shape how the AI terminal responds, keeping agent output consistent with team conventions and project-specific preferences.
- Oz Cloud Agent Orchestration: Deploy and track multiple agents simultaneously across multiple repositories from a central orchestration platform.
- Secret Redaction: Automatically detect and redact secrets from agent context using custom regex patterns.
- Password Manager Integration: Read secrets directly from connected password managers during terminal sessions.
- Access Controls: Manage team permissions with SAML-based SSO, role-based access control, and centralized admin controls and usage reporting.
- Flexible Hosting: Support for bring-your-own-LLM (BYOLLM) and flexible hosting options to keep code inside your own infrastructure.
Real-World Applications
A developer working on a production codebase can go from prompt to pull request without leaving the command line. Describing a feature in natural language, attaching relevant files as context, letting an agent write the code, then reviewing and commenting on the diff before sending it back for revision — the entire loop happens inside a single AI terminal environment.
Engineering teams at larger organizations may find the knowledge store, shared workflows, and multi-agent support especially useful. When multiple developers work on related systems, reusable Notebooks and Prompts keep processes consistent and reduce repetition. MCP server connections let the AI terminal pull in live context from tools like Figma or Sentry, so agents work with real data rather than assumptions. Oz can spin up several agents across different repositories at the same time, making it practical to coordinate parallel development work.
Security-conscious teams and enterprise engineering departments can deploy the AI terminal in environments with strict data policies. BYOLLM support means the LLM can run inside private infrastructure. Secret redaction, SAML-based SSO, and role-based access controls give IT and security teams the oversight they need. Agent autonomy controls let teams decide how much independent action agents can take, which matters when working in sensitive or production environments.
Developers who switch between multiple AI coding agents may find this workflow particularly practical. Rather than managing separate tools for each agent, running Claude Code on one tab and Codex on another is possible from the same AI terminal interface. The vertical tab system and agent-aware metadata make it easier to keep context organized across concurrent sessions. AI context rules and Warp Drive keep agent behavior aligned with project standards no matter which agent is running.
