A tray app for Windows and Linux

Know the moment your coding agents need you

foolscap watches Claude Code, Codex and any ACP agent from your tray. Answer them without switching windows, and queue the next prompt: it goes out only when your tests pass.

14-day free trial · $19 once · open source under GPL-3.0

A real run, replayed

  1. 0:00Starts the task in its own worktree
  2. 0:12Asks to edit src/math.mjs. You approve.
  3. 0:29Asks to run npm test. You approve.
  4. 0:43foolscap runs npm test itself: 2/2 green
  5. 0:44Your queued next step goes out on its own
  6. 0:47Edits the code and its test, asking each time
  7. 1:26Runs npm test again, with your OK
  8. 1:40foolscap checks again: 3/3 green. Ready to merge.
  9. 1:58Merged as one commit: 2 files, +10 −1

Claude Code on a throwaway repo, recorded on 18 October 2026: 1 min 58 s from start to merge, shown 4× faster.

Works with the agents you already use

The pill

Answer without switching windows

A small capsule floats above your taskbar with one mark per task: a dot when one needs you, a ring while one works. Hover it to approve, deny or merge right there.

  • On Windows the notification itself has Approve and Deny, and takes a reply.
  • Claude Code sessions you start in your own terminal show up too, approvals included.
The pill opened on hover, asking to approve an edit to the CI workflow.

The orbit

Urgency as distance

Click the pill and the orbit opens: you in the middle, every task a planet. Whatever needs you sits closest; finished work drifts to the edge.

  • Tab walks the planets, A approves, D denies, Esc puts it away.
  • Each task keeps its place, so the picture never reshuffles under you.
The orbit with its three rings: your move, working, done.

The queue

The next prompt waits for green

Give a task its “then…” steps. After each turn foolscap runs your repo's own tests, outside the agent's sandbox, and sends the next step only when they pass. Red or unclear holds the queue and tells you why.

  • Green means exit code 0 and at least one test that actually ran, read from the runner's own counts.
  • Every task gets its own git worktree. Merge lands it as one commit; Discard throws it away.
A working task with two queued prompts waiting.
A finished task: 103 of 103 tests, checked by foolscap, ready to merge.

Voice

Say the next step

Press the mic, talk, and foolscap writes it down as a clean prompt for the agent. Every licence comes with 2,000 voice words a week.

  • foolscap Cloud keeps no audio, transcripts or prompts: only a hash of your licence key and a word count.
The composer: pick a repo and an agent, type or speak the task, and tick Own worktree.

While you were away

Come back and the panel says what finished green, what went red and what is waiting for you. Counted from a local journal, never guessed.

Your settings, or none

Start Claude Code with only the repo's own settings, or with your CLAUDE.md. foolscap never reads or stores your agents' logins.

Personal assistants too

Hermes and OpenClaw connect over ACP like any agent. Their queue always waits for you.

Open source

The whole app is GPL-3.0 on GitHub. Build it yourself and it is never gated.

Pricing

Pay once

Same price in every country. Try the app free for 14 days, no card needed.

First 50 buyers: $9

foolscap

$19once

  • The full app, for life
  • 2,000 voice words a week
  • 3 computers, 7 days offline
Opens at launch

foolscap + Voice

$99once

  • The full app, for life
  • 25,000 voice words a month, for life
  • 3 computers, 7 days offline
Opens at launch

Voice

$8a month, or $79 a year

  • 120,000 voice words a month
  • For people who talk to their agents all day
  • Cancel any time
Opens at launch

Prices in US dollars. Building from source is free forever.

Download

Get foolscap

You need at least one agent installed: Claude Code, Codex, or any agent that speaks ACP.

Builds come from GitHub Releases.

Windows

Windows 10 and 11, 64-bit. Installs for you alone, no admin rights.

Linux

64-bit. Pick the package for your distribution.

FAQ

Questions

Something else? Ask on GitHub.

Does foolscap see my code or my agent logins?

It runs your own claude and codex on your computer, the same way you would in a terminal. It never reads, stores or sends your agents' logins or API keys, and your code never leaves your machine through foolscap.

Which agents does it work with?

Claude Code and Codex natively, plus anything that speaks the Agent Client Protocol: Gemini CLI, OpenCode, Goose, Qwen Code, Kimi, Copilot CLI, Kiro, and personal assistants like Hermes and OpenClaw. foolscap only shows the ones it finds installed.

What counts as “green”?

foolscap runs your repo's own test command after each turn, outside the agent's sandbox. It counts as passing only with exit code 0 and at least one test that actually ran, read from the runner's own summary. Zero tests, a missing exit code or a sandbox block is never green, so the queue holds and tells you why.

Is it really open source?

Yes. The whole app is on GitHub under GPL-3.0. The official installers need a licence after the 14-day trial; a build you make from the source is never gated.

What happens to my voice?

Your audio goes through foolscap Cloud to a speech service that turns it into text. foolscap Cloud keeps no audio, transcripts or prompts: only a hash of your licence key and how many words you used, so it can count your allowance.

How many computers can I use?

A licence works on 3 computers at once, and keeps working for 7 days offline. You can move it from one computer to another whenever you like.

Is there a Mac version?

Not yet. foolscap is built for Windows and Linux first, where nothing like it existed.

Your agents, one tray away

Stop checking terminals. foolscap tells you when it is your move.