Every agent, one window.
Skoga is the desktop cockpit for running AI coding agents in parallel. Each agent works in its own sandbox, the ones that need you rise to the top, and the good runs ship as PRs.
Twelve agents, one glance.
The roster shows every agent you have working right now, color-coded by state. Blocked agents float to the top of your attention; running agents stay quiet until they have something to say. No tabs. No Slack threads. No guessing which terminal holds the agent that just opened a PR.
Status at a glance
Active count, what needs you, PRs awaiting review — the four numbers that decide what you do next.
Spend & pulse
Token spend against your daily budget, plotted live. Catch a runaway loop before it costs you a coffee.
Blocked, with context
When an agent stalls, you see the question, the branch, the diff progress — and you drop into the thread in one keystroke.
Your guild, by name
Agents have names, not UUIDs. Astrid does frontend. Björn does backend. Magnus is on CI. You learn their strengths the way you would with people.
Drop in, finish the thought.
When an agent stalls or asks a question, jump into the embedded terminal, see what they're seeing, and unblock them. The activity feed remembers what already happened; the run details panel tells you where you are. When you're done, you push the commits or open the PR from the same view.
Run twelve agents in parallel. Break nothing.
Every dispatched agent gets its own sandboxed environment, rooted in its own git worktree. They can't see each other's edits. They can't touch your working copy. The cockpit lets you parallelize aggressively because the runtime makes parallelization safe.
- 01 One worktree per agent Every run lives on its own branch in its own checkout. No collisions on shared files; no agent stomping on another's WIP.
- 02 Sandboxed processes Filesystem and network scopes per agent. Shell access is gated; secrets are mounted by role, not blanket-exposed.
- 03 Your working copy is yours While agents run, you keep editing on
main. Their patches arrive as PRs you review at the end of the loop, not commits that surprise you.
Built for the work, not the demo.
Three ideas we keep coming back to. They show up in every screen and every keyboard shortcut.
Quiet until it matters.
An agent that is working should not be louder than an agent that is stuck. The cockpit reorders itself so the question you can answer is the one in front of you.
Agents earn trust by name.
You dispatch individual agents with their own credentials, scopes, and audit trail. Magnus broke CI yesterday; Magnus is now on a shorter leash.
Ship the good runs.
When a run is done, you don't copy diffs around. You review the patch in-place, sign off, and Skoga opens the PR with the right reviewers and labels.
Ready when you are
