ignite

mcp
Security Audit
Fail
Health Warn
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Fail
  • exec() — Shell command execution in demo/pantry/src/db.ts
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

The only AI coding app you'll need. Run agents in parallel. Hand off tasks that come back as pull requests. Draw on screenshots to show exactly what's wrong. Native macOS, sandboxed by default.

README.md

Ignite

The only AI coding app you'll need. Hand off work, draw on what's wrong,
review the pull request.

A desktop app for running AI coding agents on your projects, several at
once. Made for macOS; and others.

Ignite, an AI coding agent app: an agent that added dark mode pushes its commit and opens a GitHub pull request, its four changed files listed with their line counts, and one file's diff open beside the conversation The Tasks view: one task needs you, one runs on its own with an annotated screenshot attached, its subtasks checked off as it edits src/timer.ts, three are up next and three are done Drawing on a screenshot of a timer app: a dashed ellipse around 25:00 and the note 'was 11:32!', with the markup toolbar's pen, highlighter, arrow, box, ellipse, text, colours and Add to new task

The agent asking how people should sign in, in the composer: three options with their trade-offs and ⌥1 to ⌥3 shortcuts, a box to write your own answer, and Let the agent decide Two subagents checking a recipes API in parallel while npm run dev runs as a background command, its live output open in a tab beside the conversation The agent showing a screenshot of the timer app in dark mode and a live, clickable preview page of both colour palettes inside the conversation

Open GitHub issues fetched through the GitHub MCP server, shown as a table, and three of them proposed as tasks waiting for approval Ignite settings on the Providers section, with Claude and ChatGPT connected through their subscriptions

Features

  • Run several agents at once. Each conversation works in its own git
    worktree, so agents never touch each other's work or yours, and the sidebar
    shows which are working, done or waiting for you. Its node_modules,
    builds and .env are copy-on-write clones of yours, so a new worktree is
    ready at once and costs almost no disk until something changes.
  • Work comes back as a pull request, every file's diff a click away, and your
    folder updates when it merges.
  • Write a task, attach screenshots or sketches, pick a model and start it.
    The agent plans the subtasks, works through them on its own and brings you
    its pull request to approve, stopping only when it needs you.
  • Click any image, yours or one the agent took, and draw on it: pen,
    highlighter, arrows, boxes and text, zooming in for detail. Then add it to
    the conversation or a task to show the agent what you mean.
  • When a decision is open, the agent asks you, with a few options and what
    each one costs. Pick one, write your own answer, or leave it to the agent.
  • Agents hand parts of the work to subagents running in parallel, and keep
    dev servers and watchers running in the background, their output a tab away.
  • Agents show you screenshots and live, clickable pages right in the
    conversation, and you can draw on either.
  • ⌘J (or ⌘1) opens a terminal beside the conversation. Agents read what it printed,
    and shell code in a reply runs there with one click.
  • Type / for the project's skills and @ for files, terminals and attached
    images.
  • Claude and ChatGPT through your Claude Code and Codex sign-ins, so you use
    the plans you already pay for. API keys work too.
  • Commands run in a sandbox: they write only inside the project, can't read
    your credentials, and may only read from websites. risky asks first, or turn on full access and nothing asks.
  • ⌘K searches every conversation, file and project, and ⌘P jumps straight
    to a file.
  • Skills and MCP servers import from Claude Code, Codex, Cursor and Claude
    Desktop in one click.
  • With cmem, agents remember what they learned about a
    project.
  • Code shows in any theme from VS Code, VSCodium, Cursor or Windsurf.
  • Use it from your phone on the same Wi-Fi:
    Settings → Remote, then scan the QR code.

Install

Needs Node.js 22.18 or newer and
Rust. On macOS, also the Xcode Command Line Tools
(xcode-select --install); on Linux and Windows,
Tauri's prerequisites.

git clone https://github.com/ker0olos/ignite.git
cd ignite
npm install
npm run setup

Ignite is now in ~/Applications on macOS, your app menu on Linux, or the
Start Menu on Windows. It updates itself each time it opens, and goes back
to the last working version if an update fails to start.

Linux sandboxes commands with
bubblewrap (bwrap). Windows, setup
offers to sandbox them too (alpha, one admin prompt the first time); without
it, Auto asks before every shell command. Settings → Agent switches it. Worktrees clone
copy-on-write on Btrfs and XFS; elsewhere agents install and build their own
dependencies.

Make it yours

Every component and function in Ignite is yours to change. Ask its agent to
rework any part of it to fit how you work: the message box, the sidebar, how
tools are approved, how the agent itself runs. Your changes survive every
update, and if one breaks something, Ignite still opens and tells you.

Develop

npm install, then npm run tauri dev, which never updates itself and
ignores ~/.ignite/mods. npm run demo opens two sample projects with
scripted conversations and tasks; no agent runs and nothing is saved.

Reviews (0)

No results found