_{▶ Click the preview to watch the full intro with sound.}

SimpleWarp

Warp's terminal, on your machine, with your own AI key.

No account. No cloud. No telemetry. Just a fast, beautiful terminal and an agent that answers to you.

Download · Build from source · FAQ · Thank you, Warp

🚀 New in v0.6.0 — a rebuilt local agent. SimpleWarp's agent loop has been reimplemented from the ground up to run entirely on your machine, modeled on pi: it now remembers every tool result, caches its prompt, tracks its context window, compacts long conversations on its own, takes your steering mid-task and retries flaky providers. See what changed ↓

---

Standing on Warp's shoulders

SimpleWarp is a fork of Warp, and every good thing you will find in it was built by the team at warpdotdev/warp.

They reimagined what a terminal could be. They wrote it in Rust and drew it on the GPU. They built their own UI framework, WarpUI, just to get it right. They turned scrollback into blocks you can select, copy and share; gave the prompt an input editor that behaves like a real text editor; and shipped completions, workflows, notebooks, themes and an agent mode that works inside your shell. Then they opened the client's source to the world, which is the only reason a project like this one can exist.

SimpleWarp adds nothing to that work. It only takes things away: every feature that needs a Warp-hosted service is removed, so the app runs entirely on one machine. That makes it a narrower product, not a better one. If you want accounts, Warp Drive, team sharing, cloud agents or a polished, supported, auto-updating app, go use Warp — it is wonderful.

To the Warp team: thank you for building it, and thank you for sharing it. 💜

Why SimpleWarp

🔒 Local-first, for real No sign-in screen, ever. Notebooks, workflows, agent conversations and settings live in a local SQLite store. The only place your prompts go is the AI provider you chose.	⚡ Fast by inheritance Warp's Rust core and GPU (WGSL) renderer, minus roughly 940,000 lines of cloud plumbing. Less code to build, less code to run.
🤖 Bring your own model Agent mode talks straight to Anthropic, OpenAI or Google with your key — or to any OpenAI-compatible endpoint, such as an aggregator or a model server on your own box. A native agent loop, rebuilt for v0.6, runs on your machine; replies stream in and tools run locally.	🎨 Still unmistakably Warp Blocks, tabs, panes and splits, autosuggestions, command corrections, the command palette, themes and the typography — everything Warp does locally works the way Warp made it.

What's inside

• A modern terminal — GPU-rendered tabs, panes and windows; command blocks with metadata; shell detection and launch configs; searchable history.
• Input that helps — completions, autosuggestions, command corrections and a command palette for everything, settings included.
• An agent in your shell — conversations with streamed replies; the model calls tools (shell commands, file reads and edits, grep, glob) that run on your machine. Add your own MCP servers and they show up as tools and resources.
• Workflows and notebooks — reusable commands and runnable docs, with .ipynb rendering and editable Mermaid diagrams, all stored locally.
• Workspaces and sessions — pick up where you left off.

A rebuilt local agent

When the Warp server went away, so did the brain behind agent mode: what was left could call a model, but it forgot tool output between turns, had no idea how full its context was, and gave up at the first rate limit. In v0.6.0 the whole agent loop is reimplemented natively in SimpleWarp, porting the design of pi to Rust while keeping what only a terminal can do — commands run as blocks in your own shell, file edits go through diff review.

	Before	Now
Memory of the task	Earlier tool results replaced by a placeholder	Every result kept in history (head and tail of long output), and restored with the conversation
Cost and speed	The whole conversation resent at full price every turn	Prompt caching (Anthropic cache breakpoints, a stable prefix for OpenAI); cache reads and writes shown in usage
Context awareness	No token counting	Per-model context windows and a live usage meter in the input footer
Long conversations	Overflow was a hard error; only a manual /compact that dropped all history	Automatic compaction before the window fills, keeping recent turns verbatim; an overflow compacts and retries once
Project memory	AGENTS.md / WARP.md only	Also CLAUDE.md, ~/.agents/APPEND_SYSTEM.md, and a remember tool that saves facts to Knowledge after you approve them
Talking to a busy agent	Wait for the whole task to finish	Steer it: a prompt sent mid-task reaches the model at the next step, without interrupting running tools
Flaky providers	A 429 ended the task	Rate limits, overloads and dropped connections retried with backoff (2/4/8 s or retry-after)
Your shell	Agent commands ran in your session, so a stray set -e could close the tab	Each agent command runs in its own subshell, like a fresh bash -c

Everything lives in the client: the policy sits in a small, unit-tested crate (crates/agent_harness), and the only traffic is still the request to the provider you chose.

Where your bytes go

SimpleWarp has a short, explicit list of network traffic, and nothing else:

Traffic	When
Your AI provider	When you talk to the agent
LSP server installs (GitHub / npm)	Only when you ask for one
gh polling for the branch's PR	Only while a PR chip is visible
MCP servers you add	When you use them (OAuth callback stays on loopback)

Gone entirely: login and accounts, Warp Drive sync, teams and sharing, cloud and ambient agents, session sharing, the server-side AI harness, telemetry, crash reporting, auto-update and computer use. Remote images in markdown and notebooks are never fetched. FAQ.md has the details.

Get it

Download

Grab the latest build from Releases:

Platform	Asset
macOS (Apple silicon)	SimpleWarp-<version>-macos-aarch64.zip
Linux (x86_64)	SimpleWarp-<version>-linux-x86_64.tar.gz
Windows (x86_64)	SimpleWarp-<version>-windows-x86_64.zip

There is no auto-update; check Releases for new versions.

First launch on macOS

The macOS app is ad-hoc signed but not notarized by Apple, so Gatekeeper blocks the first launch of a downloaded copy ("Apple could not verify…"). Since macOS 15 the old right-click → Open trick no longer works. Pick one:

• Terminal (quickest): unzip, move the app to /Applications, then clear the quarantine flag macOS put on the download:

  xattr -dr com.apple.quarantine /Applications/SimpleWarp.app

• System Settings: open the app once and dismiss the warning, then go to System Settings → Privacy & Security, scroll to Security, click Open Anyway next to the SimpleWarp message, and confirm.

Either way you only do this once per downloaded version. A build you make yourself (below) isn't quarantined and opens directly.

Build from source

./script/bootstrap   # one-time, platform-specific setup
cargo run            # build and launch

script/bundle_simplewarp builds a standalone macOS .app; it needs cargo about (script/install_cargo_release_deps), or set NO_LICENSES=1 to skip the licence file.

Then open Settings → SimpleWarp Agent, add your provider API key, and start a conversation.

Warp or SimpleWarp?

	Warp	SimpleWarp
GPU terminal, blocks, completions, workflows, notebooks	✅	✅
Agent mode	✅ Warp-hosted	✅ your own key
Account, Warp Drive, team sharing	✅	—
Cloud agents, session sharing	✅	—
Auto-update, notarized builds, support	✅	—
Works with no account and no Warp servers	—	✅

If the left column is what you need, use Warp. If you want the right one, welcome.

Hack on it

./script/presubmit   # fmt, clippy and tests

AGENTS.md is the engineering guide (build and test commands, code style, WarpUI patterns) and CONTRIBUTING.md the short version. The architecture lives in specs/, and the story of how Warp became SimpleWarp — every removal round and decision — is in simplify-specs/.

Licensing

Same licences as upstream: the UI framework (warpui_core and warpui) is MIT and the rest of the code is AGPL v3. This is a derivative of Warp's open-source client and the upstream copyright notices stand. "Warp" is a trademark of its owners; SimpleWarp is an independent fork, not affiliated with or endorsed by Warp.

Security

See SECURITY.md. The Code of Conduct is upstream's.

_{Made with gratitude, on top of Warp.}