Run a pack of agents (on Claude or Codex), each in its own isolated git worktree, so they don’t step on each other. Then review it all in the editor: diffs, git, databases, Docker. Sessions are saved, so you can quit and pick any of them up later.
signed .dmg · macOS & Linux · free
Shipping in beta · rough edges expected · bug reports welcome
A small Symfony service for public-holiday calendars, leave requests, and reliable working-day calculations.
Sync a country calendar from DateNager and upsert it without replacing local leave data.
bin/console app:holiday:sync --country=DEStart up to eight focused agents in this workspace.
claude on your PATH
codex on your PATH
Requires Claude Code and/or Codex CLI installed locally.
“It got annoying, jumping between a stack of terminal sessions and a separate IDE. I wanted both worlds in one place, in a way that’s easy and reliable.”
Shikigami is built and released by me, on my own. It’s free, under active development, and currently in beta. Your feedback matters to me, and it shapes where the app goes next.
Spin up a few agents in a few tabs and they fight over the same working tree, clobber each other’s edits, and lose every bit of state the moment you quit. That’s the problem to solve, not another chat box.
Each agent gets a dedicated git worktree. No collisions, no clobbered edits.
Quit and reopen. Sessions reattach exactly where you left them.
Projects as tabs, agents as panes, all of them at a glance.
A shikigami is a spirit from Japanese folklore, summoned to carry out tasks on someone’s behalf. That’s roughly the model here: you hand work to a set of agents and they go off and run it. The name fit, so it stuck.
Agentic Coding mode runs the agents. A full editor lets you review their work. It’s a native desktop IDE, not a browser wrapper.
The agents
Run as many agents as you need. Each one works in its own git worktree, so they can all touch the same repo at once without overwriting each other.
Run each agent on Claude Code (Opus, Sonnet, Haiku) or OpenAI Codex (GPT-5.6). Pick the model and reasoning effort per agent.
A real PTY behind every agent: a full shell, not a sandbox.
Quit and reopen. Every session reattaches where you left off.
When an agent finishes its turn or needs your input, it sends a desktop notification. Click it and you go straight to that agent’s session.
A full IDE to review their work
Switch to Editor mode for a Monaco code editor with your whole tree, split view, and a diff viewer. Review a diff, walk the git history and per-line blame across every repo in a monorepo, then commit (GPG-signed) without leaving the app.
Go-to-definition, find-references, completion, and live diagnostics for PHP (8.1–8.5) and TypeScript/JavaScript — no language server to install.
Browse MySQL tables, run and save SQL queries, and inspect Redis keys — read-only, right beside the diff.
See your Compose stack, tail per-service logs, open a container shell, and start or stop services — all from inside the app.
Four agents run in the grid. The fifth is the stream where this whole app gets built, live and unedited, commit by commit, by the one developer writing it. Same repo, same branch, nothing cut.
Name a task and pick an agent — Claude or Codex. It spins up in a fresh, isolated git worktree of its own.
It works alongside your other agents while you watch, or while you do something else.
Read the diff in the editor, resume any session, merge what’s good, and dismiss the rest.
One agent per branch, all building at once.
Write the parts you care about in the editor while an agent handles the rest alongside you.
Kick one off, quit, resume it tomorrow.
An agent per failing test, all running at once.
Free. Solo mode. A signed desktop app. No account, no cloud.
signed .dmg · no sign-up