Blitz 0.3.0

A coding agent that does the work and gets out of the way. Blitz reads your project, makes the change, checks it and stops. It has no persona and no chatter. It’s a compiled Go program that runs the model you choose, on your machine, with your keys. It’s open source under the Apache-2.0 license.

Where to get it

Everything is on the GitHub releases page, signed so you can verify each download. On macOS or Linux, one line installs the command-line tools:

curl -fsSL https://github.com/retail-cortex/blitz/releases/latest/download/install.sh | sh

Desktop app: .dmg for macOS 13 and later; .deb for Ubuntu 24.04, Debian 13 and later.
VS Code: the .vsix from the same release puts the Blitz chat beside your code.
Windows: a .zip with the command-line tools.

Windows desktop app: coming soon. Today, Windows gets the command-line tools. They run without the OS sandbox that macOS and Linux use, and shell commands need Git Bash or WSL. The desktop app for Windows is the next platform on the way.

How to use it

# once
blitz config set-key gemini       # or anthropic, openai; stored in your OS keychain
blitz doctor                      # checks the key, the settings and the sandbox

# then, in any project
blitz                             # an interactive session
blitz "why does the build fail?"  # one question, then exit

Ask for what you want. Blitz shows a diff before every edit and asks before running commands. You can allow an action once, for the session, or always. /undo reverts the last turn. Run blitz service install to use the desktop app, the VS Code extension and scheduled jobs. They all share one engine, so a session you start in the terminal carries on in the app.

What you get

Your choice of model: Gemini, Claude, any OpenAI-compatible API, or local models through Ollama. You pay your provider directly, or nothing for local models.
Safe by default: commands run in an OS sandbox on macOS and Linux, edits wait for your approval, and every turn can be undone.
Workspace search: new in 0.3. Search your code, PDFs, notebooks, past chats and notes by their words, or by meaning if you set an embedding model. The index stays on your machine. See Workspace search.
Built for data work: CSV files of any size open as a table you can search, filter by column and sort. With summaries turned on, a model notes what each column holds.
Find in file: ⌘F (Ctrl+F) finds in the open file, in the Markdown preview or the source.
Works with what you have: AGENTS.md, CLAUDE.md and GEMINI.md load as project instructions. MCP servers, hooks, custom commands and Agent Skills work too.
Runs unattended: scheduled workers, background tasks in their own git worktrees, and JSON output with limits on turns, cost and time for scripts and CI.

How it compares

Most AI coding tools now fit one of three shapes: an editor built around AI, an extension for the editor you already use, or an agent that runs in the terminal. Blitz is closest to the third, with a desktop app and a VS Code extension on top of the same engine.

Tool	What it is	Models	How you pay	Source
Blitz	Terminal agent, background service, desktop app and VS Code extension, all on one engine	Gemini, Claude, OpenAI-compatible, local through Ollama	Free. You use your own API key, or a local model.	Open (Apache-2.0)
Google Antigravity	An AI-first editor built on VS Code, managing several agents in parallel, with a built-in browser; a CLI and extensions for other editors since 2026	Gemini, plus some Claude models	Free plan with weekly limits; higher limits with a Google AI subscription	Closed
Cursor	An AI-first editor built on VS Code	Several providers	Subscription, with a free tier	Closed
GitHub Copilot	An extension for many editors, plus a coding agent that works from GitHub issues	Several providers	Subscription, with a free tier	Closed
Claude Code	A terminal agent, with editor and desktop versions	Claude	Claude subscription or API	Closed
Gemini CLI	A terminal agent	Gemini	Free tier, or your API key	Open (Apache-2.0)

Plans and features change often. This reflects public information as of October 2026.

Where Blitz is the better fit

No lock-in to one provider: switch models per task, including a model running on your own machine, with no subscription.
Safety you can see: sandboxed commands, approvals and undo are on from the start, not something you add later.
One agent in every place you work: the terminal, a desktop app and VS Code share sessions instead of being separate products.
Open source: read the code, build it yourself, and check each release’s signature.

Where another tool may suit you better

Suggestions as you type: Copilot, Cursor and Antigravity complete code inline in the editor. Blitz works by conversation and doesn’t do this.
Bundled model access: a subscription can be simpler than managing API keys. Antigravity’s free plan includes model use within its limits.
Maturity: Blitz is at version 0.3. The large tools have bigger teams, longer track records and larger communities.
Windows today: until the Windows app arrives, Windows users get the CLI only, without the sandbox.

Try it in five minutes

Install it, set a key, run blitz doctor, and ask it about your build. Getting started covers the rest, and the repository welcomes issues and contributions.