Keep agent sessions active
Claude Watcher helps keep Claude Code, Codex, and long terminal agent runs covered when you need to close your Mac or step away.
What it does
Claude Watcher is a session keepalive and receipt dashboard for coding agents.
It is built for developers using Claude Code, Codex, auto-accept, auto prompting, or long terminal agents who want fewer dead sessions, fewer vague prompts, and less guessing when they return.
Review everything in one dashboard
Come back to active, idle, and finished sessions with changed files, token burn, risk labels, and rollback notes.
Coach weak prompts before they burn time
Auto prompting can turn vague Claude Code requests into sharper next actions when your local Ollama setup is enabled.
After the walk
A dashboard receipt for every unattended agent run.
When you get back, Claude Watcher makes the aftermath readable: what changed, which actions looked risky, how many tokens burned, and what to restore if the agent went too far.
Run receipt auth-hardening-42
Done Time38m
Tokens42.1k
Files18
Risk2
Changed files
src/auth/session.tsedited
src/middleware/rate-limit.tscreated
.env.examplechecked
Risk notes
.envsecret-like path
git push --forceblocked
Watcher Phone
Approve a run from across the room.
Glance at live status and answer the one prompt that needs you — from your phone, over your own Tailscale or SSH tunnel. Your Mac stays the host. No hosted server, no cloud, no telemetry.
- G Glance See which run is active, waiting, or blocked without opening the lid.
- A Approve When a step needs a yes or no, tap Allow or Deny. Catastrophic commands are always blocked — never offered for remote approval.
- N Get pinged A local macOS notification fires the moment a run needs you or a risky command is blocked.
Your Mac is the host Your own network No cloud
10:42 Watcher Phone
Needs you
Bash
dangerous keyword 'git push --force'
git push --force origin main
Allow Deny
Live sessions
active auth-hardening 2m
waiting docs-site 9m
How it works
The point is not analytics. The point is closing your Mac.
01
Start the agent.
Run Claude Code, Codex, or a long shell job the way you already do. No new workflow to learn.
agent run02
Close the lid.
Tap Safe Carry before you walk away. Claude Watcher covers the run while you move.
safe carry03
Glance & approve.
Watch live status from your phone over your own network. When a run needs a yes or no, Allow or Deny it without opening the laptop.
watcher phone04
Come back clean.
Check the dashboard receipt for changed files, risk labels, token burn, and rollback notes before you trust the run.
dashboardWatcher apps
Leave the desk first. Understand the run second.
The impulse buy is simple: stop carrying an open laptop around the room just because Claude Code, Codex, or a terminal agent is still working. Glance and approve from your phone instead — over your own network, never our cloud.
Close
Close the laptop before a long Claude Code or Codex run finishes.
Carry
Move rooms, take a call, grab coffee, or commute without babysitting the terminal.
Glance
Open the Watcher Phone on your own network and see which run is active, waiting, or blocked — from another room.
Approve
When a run needs a yes or no, Allow or Deny it from your phone. Catastrophic commands are always blocked, never remotely approved.
Return
Come back to the same run instead of a dead session, wasted tokens, and lost context.
Radar
See which Claude Code run is active, waiting, blocked, or done.
Risk
Catch secret-like paths, destructive commands, force pushes, and custom rules.
Receipt
Save a plain-English run summary with actions, files, result, and risks.
Rollback
Know which files to restore and what steps undo the bad run.
Prompt
Turn vague Claude Code prompts into sharper next actions with local prompt coaching.
In the wild
Three moments it pays for itself.
Not testimonials — the everyday moments Watcher Phone is built for. Step away, stay in control, come back to a receipt.
The commute carry
Kick off a refactor, close the lid at the door, and approve the one risky migration step from the train. The run never died, and you didn't carry an open laptop down the platform.
The coffee-run approve
A test suite wants a force-push to a throwaway branch. You tap Allow from the counter instead of sprinting back to the desk — then read the receipt when you sit down.
The overnight build
Leave a long Codex job on Safe Carry overnight. A push to a protected path waits on your phone; everything safe keeps moving, and a catastrophic command is blocked outright.
Pricing
$0 to watch. $9 to stop babysitting.
Free lets you feel the radar. Solo is for the moment you want to close the laptop, glance from your phone, and approve a run from across the room — all on your own network.
Free
$0
First runs
- Safe hook install
- Claude live status
- Basic trace
- Basic token view
Solo Lifetime
$9
Pay once
- Keep sessions active
- Safe Carry
- Watcher Phone glance
- Approve from your phone
- Session dashboard
- Safe auto-approve
- Risk labels
- Rollback notes
Team features later. Multi-repo exports, shared rules, and policy templates are not part of the $9 decision today.
Payments by Lemon Squeezy. 14-day refund. License verification is local.
FAQ
Small questions before checkout.
What is Claude Watcher? +
Claude Watcher is a macOS app for Claude Code, Codex, and long terminal agent runs. It helps keep sessions active, safely handles Claude Code approvals, and gives you dashboard receipts when you return.
Can it really keep my agent running when I walk away? +
That is the core job: start a run, use Safe Carry, close the laptop, and come back without losing the session. It cannot help if the Mac powers off, runs out of battery, overheats, or is blocked by company policy.
Why not just leave Terminal running? +
Because the pain is not a command-line flag. The pain is leaving your desk during an agent run and still knowing whether the run is alive, what changed, and what to inspect when you return.
Does Claude Watcher replace auto accept or bypass mode? +
It complements them. The local Claude Code hook can auto-approve safe tool prompts, stop for risky actions, and leave you with run status, risk signals, and a dashboard receipt when the run finishes.
What is auto prompting? +
Auto prompting is local prompt coaching. When enabled with your local Ollama setup, Claude Watcher can turn vague Claude Code requests into sharper one-line next actions and copy the rewrite to your clipboard. It is local and optional.
Why would I pay if auto accept already exists? +
Auto accept answers the prompt. Claude Watcher answers the practical question: can I close my laptop without killing the session, and what happened while I was away?
Does it support Codex? +
Yes for Safe Carry and long-running terminal coverage. Automatic live-session triggering and safe auto-approve are currently Claude Code-first; deeper Codex tracking is on the connector roadmap.
What is a receipt? +
A receipt is the post-run dashboard summary: time away, timeline, changed files, risk classification, token burn, result, and rollback notes. Solo can export it as Markdown.
Does my code leave my Mac? +
No. Claude Watcher is local-first. It does not upload traces, repo contents, prompts, or terminal history to a hosted replay service.
Is the phone control cloud-based? +
No. Your Mac stays the host. The Watcher Phone page is served by Claude Watcher itself on 127.0.0.1; you reach it from your phone over your own Tailscale network or an SSH tunnel. There is no hosted server, and nothing is uploaded.
How do I connect my phone? +
In Settings - Watcher Phone, scan the pairing QR with your phone while it is on the same Tailscale network or SSH tunnel. The QR carries a bearer token in the URL fragment, so treat it like a password.
Can someone on my network approve my prompts? +
The interface stays bound to 127.0.0.1, so it is not exposed to your LAN - only a device on your private tunnel with the token can reach it. Catastrophic commands are always blocked and never offered for remote approval, and if no answer arrives in time the run falls back to asking on the Mac.
Can it roll back automatically? +
It focuses on rollback notes and pre-run context instead of blindly rewriting your repo. You stay in control, with the exact files and steps called out.
Why $9 lifetime? +
Because this should be an easy yes for developers already spending money and attention on coding agents. The price is intentionally impulse-level.
Is this affiliated with Anthropic or OpenAI? +
No. Claude Watcher is an independent third-party tool. Claude is a trademark of Anthropic.
The impulse
"I just want to close my laptop."