macOS menu bar · Claude Code + Codex

Your agent waits.
You tap. It keeps going.

Nudge turns the moments your coding agent needs a human — an approval, a question, a "what next?" — into a notification you can act on. Approve, answer, or redirect the run without ever opening its window.

Buy Nudge — $7
Claude Code Codex Runs 100% on your Mac
Claude Code needs approval ↗✕
Bash rm -rf dist/ && git push --force
Deny
Approve
Codex has a question ↗✕
Which migration approach for the v2 schema?
Dual-writeWrite to both schemas, backfill, then cut over
Expand-contractAdd columns, migrate reads, drop old
The gap

You start a long run and step away. Ten minutes later it's still sitting there — blocked on one approval, waiting for one answer, doing nothing until you come back and notice.

What you can do from a notification

Not a ping. A control surface.

Other tools tell you your agent finished. Nudge lets you act — the run pauses on the notification and resumes the instant you respond.

APPROVE / DENY

Answer the permission prompt from the notification

When your agent hits a gate — a shell command, a force-push, a file write — the run blocks and your tap is the decision. Approve and it continues; deny and it's told why. You never touch the agent's dialog.

Claude Code needs approval↗✕
apply_patchedit src/auth/tokenStore.ts
Deny
Approve
REPLY TO CONTINUE

Redirect a finished run with one message

An agent wraps up — but you want one more thing. Type it straight into the notification and the run keeps going with your instruction. No switching windows, no re-reading context. Multi-line, paste, the works.

Claude Code finished↗✕
Refactored the auth layer and all 42 tests pass.
Now add rate-limiting to the refresh endpoint…
Send
ANSWER QUESTIONS

Real options, laid out in full

When your agent asks which path to take, Nudge shows the full question and every option as a proper list — with a text box for anything custom. Pick, and the agent proceeds. No more truncated system notifications.

Codex has a question↗✕
Which deploy strategy for this release?
Blue-greenParallel environment, instant switch and rollback
CanaryRoute 5% first, watch metrics, then ramp
Type a custom answer…
One panel, both agents

Two agents. One panel.

Most people run more than one agent. Nudge speaks to both natively and stacks their notifications in a single place — so you're not wiring up a separate script for each, or checking two apps.

How it works

Quietly local. Nothing to babysit.

Private

Stays on your Mac

The app talks to your agents over localhost. No account, no cloud, no code or prompts leaving your machine — ever.

Native

Uses the agents' own hooks

One-click setup wires Nudge into Claude Code and Codex. It only steps in when a human is actually needed.

Safe by default

Fails open

Quit the app and nothing breaks — your agents simply fall back to their normal built-in prompts. You're never locked out of your own tools.

Pricing

Buy once. It's yours.

One purchase, one Mac utility, yours for good. No subscription, no seats, no telemetry.

Requires macOS 13+ and Claude Code and/or Codex.

$7 lifetime
  • Approve, deny, answer & reply from the notification
  • Claude Code + Codex, one panel
  • One-click setup, launch at login
  • Lifetime updates included
  • 14-day money-back guarantee
Buy Nudge

Instant download · Apple-notarized

Questions

Good to know

Is any of my code or data sent anywhere? +

No. Nudge runs entirely on your Mac and talks to your agents over localhost. There's no account and no server — your prompts, code, and decisions never leave the machine.

Which agents does it work with? +

Claude Code (the desktop app and CLI) and Codex (the CLI, and Codex inside the ChatGPT desktop app). Both are set up in one click.

Will it slow my agent down? +

Only when you want it to. When you're at the keyboard it stays out of the way. It holds a run open just for the moments it's waiting on your approval or reply — which is the whole point — and you can turn that off per situation.

What if the app isn't running? +

Nothing breaks. If Nudge is closed, your agents fall back to their normal built-in prompts. It fails open, never blocking your tools.

Can I reply and have it show in the chat? +

Your reply continues the run and the agent acts on it. Note that it arrives as agent input rather than a chat bubble — a current limitation of how the agents expose this, not the app.