Watchman

The agent that watches your agents.

Every automated system fails eventually. A process dies, an API changes underneath it, a deploy breaks something that worked yesterday. Watchman notices in seconds, fixes it, and tells you what it did.

No dashboard to sit in front of. No alert at 3am that you have to get up and act on.

The question

What happens when it stops at 2am?

For most automation the honest answer is: nothing, until a customer notices and complains. The agent is not down in a way anyone can see — it is quietly not doing its job, and the first person to find out is the one you least want finding out.

What it does

Five things, without waking anyone.

Watchman does not page you and wait. It takes the action a person would have taken, and leaves a record.

Restart
The process died. It is back up before the next job is due, with a limit so a crash loop cannot run away.
Retry
One action failed — a call, a lookup, a write. It is tried again on a timer, not abandoned.
Resend
A report or message did not go out. It is sent again, and checked against the provider that it actually landed.
Roll back
A deploy broke something that worked. The last good version is restored, and the bad one is held for you to look at.
Fix the code
The failure is in the code itself. Watchman reads the error, writes the fix and ships it — then watches to confirm the fix held.

The difference

Monitoring tells you it broke. This fixes it.

Every monitoring tool on the market ends at the alert. Something is wrong, here is a red dot, now go and do something about it. That is only useful if somebody is awake, has access, and knows the system.

02:14:07 check failed report send returned 500 02:14:09 diagnosed provider field renamed in API response 02:14:31 patch written mail.py, 3 lines 02:14:44 deployed verified — report delivered 07:00:00 morning note 1 issue found and fixed overnight

You read about it over coffee, after it is already fixed — not at 2am, while it is still broken.

What reaches you

You hear about it only when it matters.

An agent that emails you every time it does its job is just another inbox to ignore. Watchman keeps the small things to itself.

Fixed it
Logged, and summarised in one line the next morning. No alert.
Could not fix it
You get an email with what broke, what it tried, and what it needs from you.
Could not fix it, and it is urgent
Email and a phone call, straight away — because some things cannot wait for morning.

What it watches

Any agent, not only ours.

Watchman is not tied to one product. It watches whatever you point it at — a voice agent, a scheduled job, a scraper, an internal tool — and it watches every agent we run for you.

Want it watching yours?

If you are already running agents that nobody is watching, we would like to hear what they are.