Get a daily list of what is wrong

When many schedules and agents run on their own, a message for every run is too much to read. The exceptions digest is one message a day that lists only what needs a look. When nothing is wrong, nothing is sent.

It lists three kinds of problem found on this machine:

Section	What is listed
Schedules	A schedule whose latest run failed or ran out of time, and a schedule that was due but did not run (for example because AgentX was stopped). A schedule that failed and then ran fine is not listed.
Stuck requests	An open request that needs attention, is past its due time, or has had no activity for a day (the requests.staleAfterHours setting). Needs open requests switched on.
Paused agents	An agent, a project, or the whole fleet that has used its monthly budget, so new runs are refused until the month ends. A project shows as project <name>.

Each digest covers what happened since the previous one, so a failed or missed run shows up in the next digest. It is off until you switch it on.

Switch it on

1. Terminal: check that AgentX knows where to send messages. Run agentx notifications and look for a destination. If there is none, follow Get notified first, or give the digest its own destination in step 4.
2. Open agentx.json in a text editor.
3. Add an exceptionsDigest section at the top level (next to crons):

   "exceptionsDigest": {
     "enabled": true,
     "time": "08:30",
     "timezone": "Europe/Paris"
   }

   time is when the digest goes out, on a 24-hour clock. timezone is the time zone that time is read in; leave it out to use this computer's.
4. Optional: to send the digest somewhere other than your usual notification chat, add a destination:

   "destination": { "channel": "telegram", "chatId": "123456789" }

5. Save the file. The daemon picks up the change by itself; no restart needed.

A digest looks like this:

Needs a look on office-mac:
Schedules (2)
- backup-check (reporter): 2026-10-25 05:00 UTC, failed: exit 2
- weekly-summary (writer): due 2026-10-25 06:00 UTC, did not run
Stuck requests (1)
- Renew the shop certificate (helper): past its due time (2026-10-24 17:00 UTC)
Paused agents (1)
- writer: monthly budget used ($50.00 of $50.00)
Details: Operations in the dashboard, or run: agentx schedule list

To have a single schedule stay quiet unless it finds a problem, see Report only when something is wrong.

Check it worked

1. Wait until after the time you set.
2. Terminal: run agentx daemon logs and search for [exceptions].
3. You see either [exceptions] digest sent: …, and the message arrived in your chat, or [exceptions] digest: nothing wrong, nothing sent.

If something is wrong

• No [exceptions] line at all: check that exceptionsDigest.enabled is true and that the daemon is running (agentx daemon status). It goes out at most once a day, after time; if the daemon started after that time today, it goes out straight away.
• It was due but nothing was sent and nothing logged: there is no destination. Set notifications.destination, or exceptionsDigest.destination.
• [exceptions] digest failed: the chat could not be reached. The rest of the line says why. The digest is not retried that day.
• A schedule you expected is not listed: a schedule that has never run is not listed as missed, and a missed run counts only once its time limit has passed (2 hours for an agent schedule). A schedule that is switched off is listed only if its last run failed.
• Stuck requests are never listed: open requests are off. See Keep track of what you asked for.