Warp Code

Warp Code is the coding layer of Warp's Agentic Development Environment (ADE). It brings an IDE-like editor, a diff-first review panel, and multi-agent control directly into the terminal so you can go from prompt to production without context-switching.

How it works

1. Open a Git repo in Warp and enable Codebase Context. Warp indexes your repository locally (no code is stored on Warp's servers) so agents can answer questions, write changes in the right files, and reference the correct symbols.
2. Start an Agent conversation from Universal Input (natural language) or with slash commands. You can attach context (files, URLs, images, block output) and set Rules in a WARP.md (also compatible with AGENTS.md/CLAUDE.md/Cursor rules) to steer behavior and guardrails.
3. Review changes as they’re generated. The Code Review panel shows live Git diffs against your branch or against main. You can attach diffs to prompts, ask for modifications, accept or revert individual hunks, or open any file in the native editor.
4. Make surgical edits in the built-in editor. It’s a lightweight editor for quick edits: tabbed files, file tree, find/replace, syntax highlighting, and Vim keybindings. Use it to tweak variable names, copy, small functions, or quick scaffolding.
5. Ship with confidence. Agent task lists make long requests explicit and track progress. Profiles let you define which model and permissions an agent has (e.g., may apply changes, run commands), and you can step in at any time.

What’s different • Diff-first UX for agent-generated code: purpose-built review panel to inspect, accept, or revise diffs without jumping to GitHub. • Terminal-native control loop: one place to run commands, orchestrate agents, open files, and apply changes—no editor/terminal ping-pong. • Codebase Context that stays private: local indexing with ignore rules (.gitignore, .warpindexingignore, .cursorignore, etc.); multi-repo referencing when several repos are indexed. • Rules and MCP: project rules in WARP.md and optional Model Context Protocol (MCP) servers give agents structured tools, data, and constraints. • Model choice + Auto: pick a specific model or let Warp select. Auto commonly pairs a strong coding model with a planning model for long tasks.

Key features • Native editor: file tree, tabs, syntax highlighting, find/replace, Vim bindings. • Code Review: live diffs, hunk-level apply/revert, open-in-editor, attach diffs to prompts. • Codebase Context: private local indexing; file/codebase limits by plan; multi-repo context; respects .gitignore and Warp/Cursor ignore files. • Agents: multi-agent management, task lists, profiles/permissions, Active AI suggestions, voice input, and slash commands. • Universal Input: type commands or natural language; Warp detects which you meant. • Privacy & security: SOC 2 Type 2; Zero Data Retention with contracted LLMs; team-enforced ZDR on Business/Enterprise.

Models supported Warp provides curated access to top models and an Auto mode: OpenAI (GPT-5 family including high-reasoning variants, GPT-4.1, GPT-4o, o3/o4-mini), Anthropic (Claude Opus 4.1/4, Claude Sonnet 4/3.5), Google (Gemini 2.5 Pro). Turbo+ includes a Lite fallback model so you can keep working after hitting quota. Enterprise supports bring-your-own LLM; self-serve BYO availability may vary—check current plan details.

How to use it day-to-day • "Explain my codebase" and "where should this live?" with Codebase Context. • "Add payments with Stripe" → agent drafts changes; you review diffs, tweak a couple lines in the editor, re-prompt, and ship. • Triage CI failures by pasting logs (Blocks as context), ask the agent to fix, then revert/select specific hunks. • Keep WARP.md project rules to encode conventions, secrets policy, architecture notes, or a checklist for PR readiness.

Pricing and limits (high level) • Free: 150 AI requests/month; 3 indexed codebases (5k files/codebase); model choice; Next Command; limited Warp Drive objects; 5 shared sessions; individual ZDR. • Pro: 2,500 requests/month; 40 codebases (10k files/codebase); model choice; overages; unlimited shared Drive objects; email support; individual ZDR. • Turbo: 10,000 requests/month; 40 codebases (20k files/codebase); overages; Lite fallback; email support; individual ZDR. • Business: 10,000 requests/month; 40 codebases (20k); team-enforced ZDR; SAML SSO; dedicated Slack account manager; up to 50 seats. • Lightspeed: 50,000 requests/month; 40 codebases (100k); overages; individual ZDR. • Enterprise: custom limits, team-enforced ZDR, SAML SSO, BYO LLM, onboarding support.

Who it’s for Developers who want an AI coding loop that stays in the terminal but adds just enough editor and review features to steer agents, understand changes, and merge safely—solo or with a team.