Troubleshooting

When something feels off in UpServe, this page lists the first things to check, grouped by symptom. Each item follows the same shape: If you see this → Check these in order. Work through the steps top to bottom.

1. The employee isn’t responding

If you see this

You sent a message and got no reply for one or two minutes.
There’s no tool activity label anywhere in the chat (such as “searching the web” or “using the browser”).

Check these in order

Confirm it’s actually running. Look at the chat for tool activity (“searching the web”, “using the browser”, etc.). Some tasks legitimately take tens of seconds to a few minutes.
Check for approval or question cards. A highlighted card saying “approval needed” or “the employee has a question” blocks progress until you respond. See Approvals for details.
Make sure the employee isn’t paused. In employee settings, the status should be Active. An employee is automatically paused when its SU balance reaches zero. Open SU & Billing to check. On the Free plan, if you don’t use UpServe at all for a week (no login, message, or chat view with any of your employees), that account’s employees are automatically deactivated to avoid wasted cost (you get one heads-up notification a day before, and employees on a paid subscription are excluded from this). In that case, sending the employee a message is what reactivates it; just reopening its chat isn’t enough.
Refresh and retry. If the above looks fine, refresh the page and send the same message again. Most transient connection issues clear up on refresh. If the message still gets no response, a second refresh often helps.
Open the logs. From the employee’s side menu, open Logs to see the most recent run result and any error. If the same error keeps coming back, jump to the Advanced section below.

Note: When a model is temporarily unavailable, the employee automatically switches to a standby model within the same AI performance tier for that turn. No notification is shown; this is handled silently.

2. I rejected an approval by mistake

If you see this

You hit “Reject” on a tool approval card, but it was actually the action you wanted.

Check these in order

Ask again. The simplest fix: send a message like “I rejected by mistake, please try that again”. The employee will request the same tool again.
Change auto-approval settings. If you use that tool often, open the employee’s Permissions screen → Run automation → Tool Approval Policy and switch that tool to Run without approval (the mobile apps still have this in the employee’s Settings screen). From then on it runs without asking.
Be careful with risky tools. For things like file deletion or external payments, keeping auto-approval off and reviewing each request is the safer default. The rules for what counts as risky are in Approvals.

3. My scheduled run didn’t run

If you see this

You scheduled the employee to run every morning at 8 AM, but the time passed and nothing happened.

Check these in order

Is the employee active? Paused employees skip their scheduled runs.
SU balance. If the balance is zero, the run is skipped and the employee is automatically deactivated. A near-empty balance may run out mid-task, so upgrade your plan in advance.
Are all required tools enabled? If a tool the scheduled run relies on is turned off, the run fails.
Open the run history. The Automatic runs side page shows each scheduled run’s next run time and a History button that jumps to the Logs tab. The Logs tab is where you see each run’s result status and failure reason, alongside SU usage trends and daily statistics.
Recheck the scheduled run itself. It’s easy to enter the wrong time or repeat interval. See Scheduled runs for how times are interpreted.

Note: During automatic runs, if the employee hits the same tool-call error repeatedly (5+ times in a row), keeps re-issuing a request that was already blocked, or repeats no-effect placeholder commands like echo/true, it automatically wraps up with a summary as a safety measure.

4. Push notifications aren’t arriving

If you see this

An automatic run finished, but no notification showed up on your phone.

Check these in order

In-app notification settings. Under Settings → Notifications, confirm the toggles for the categories you want (like Employee response push) are on.
System permission. In iPhone settings → Notifications → UpServe, Allow Notifications must be on.
Do Not Disturb. If the in-app DND window is on, push won’t be sent during that time range. The notification is still recorded in the app so you can see it when you reopen UpServe.
Device registration. If you reinstalled the app or signed out and back in, open the UpServe app once to re-register the device.
You’re already on that chat. If you’re actively viewing the same employee’s chat, the notification shows inline only, and push is intentionally skipped. This is expected.

The full guide is in Notifications.

5. My SU usage suddenly jumped

If you see this

Your balance is dropping faster than usual, or one employee’s usage looks unusually high.

Check these in order

See what’s running. The employee’s side menu → Logs shows total SU usage, model breakdown, and daily trends; the chat view groups tool calls so you can see how many times each tool ran. If you see the same tool repeating, go to the next step.
Is the employee stuck in a loop? If it keeps calling the same tool as if it doesn’t see its own results, message it: “use the result you already have and wrap up.” That gives it an explicit stop condition.
Drop the AI performance tier. If the employee is on Ultra, move it back to Standard (on the employee’s Permissions screen on the web, or its Settings screen on the mobile apps for now). Simple tasks rarely need Ultra, and the saving is large.
Trim heavy tools. Browser and computer use cost more per call. Keep them on only when needed. There’s a feel for relative tool cost in Adding Tools.
Review your SU and plan. Free SU refresh and subscription options are explained in SU & Billing.

6. An external service connection expired

If you see this

A skill or tool that talks to GitHub, Linear, Notion, Slack, etc. shows “unauthorized” or “please reconnect”.

Check these in order

Check connection status. You can find the connection status in either of these two places:
- Skills menu → My Library : a “Reconnect Required” banner appears directly on the skill card.
- Employee settings → Skill selector: Expand the skill row in employee settings. The external service connection row appears there too; an amber warning icon shows when reconnection is needed.
Reconnect. Hit the Reconnect button from either location and complete the external sign-in again.
Scopes changed. If the external service revoked permissions or the required scopes changed, make sure to grant all the ones you need when you reconnect.
Retry. After reconnection, ask the employee to retry the same task.

7. A skill keeps failing or was deactivated

If you see this

A skill card shows a red skill score dot.
You received a library cleanup notification and the skill is listed under “Often failing”.
Or you received a skill deprecated notification.

Check these in order

Open the clean-up suggestions in the Skills menu (My Library). At the top of My Library, a row reading “N skill(s) to consider removing” appears. Tap Review to see the skill’s category in the candidates list. (Opening it from a notification takes you straight there automatically.)
If the skill is listed as “Often failing”: identify the likely cause and choose an action:
- If an external service login expired, reconnect from employee settings or the Skills menu (My Library) (see section 6 above).
- If the external service changed its API, search Official skills in the Skills menu for a replacement skill that does the same job.
- If you want to fix it yourself, copy and edit the skill to make your own derived copy.
If the skill is listed as “Deprecated”: it is no longer supported. Search Official skills in the Skills menu for a successor or updated alternative, or copy and edit the skill to maintain a derived copy yourself.
If an expired external service login is the cause: follow the steps in section 6 to reconnect.

8. Someone I shared with can’t open the link

If you see this

You sent a share link to someone outside UpServe, and they report seeing “link unavailable” or “temporarily blocked”.

Check these in order

Share toggle. Open the employee’s side menu → Public version and confirm the share is Active. Turning it off cuts access immediately.
Your (owner’s) SU. A public version runs on the owner’s SU. If your balance hits zero, the external user is blocked too.
Session / daily cap reached. If the per-session SU cap is hit, that session is permanently blocked. If the daily link cap is hit, new sessions are blocked for the rest of the UTC day and automatically permitted again after midnight UTC. Raising the cap unblocks access immediately.
Link manually disabled. If you turned off the share toggle for that link, access is blocked. Creating a new public version does not automatically disable previous links. Only manual deactivation or SU exhaustion does.

The full sharing model is explained in Sharing an AI Employee Externally.

9. Computer use / browser tools look stuck

If you see this

The preview hasn’t visibly changed for a while, or there’s progress but no result is coming.

Check these in order

Watch the activity labels. Captions like “taking screenshot”, “click”, “typing” updating every few seconds mean the employee is working. Slow pages and busy sites just take time.
Wait briefly. Heavy pages (sign-in dashboards, ad-heavy sites) can take 10 to 30 seconds to settle.
Ask the employee where it is. Send “what step are you on? if you’re stuck, move on”. The employee will report current state or switch approach.
Retry or simplify. If it keeps getting stuck on the same page, ask it to “find that info another way” or break the task into smaller steps and try again.

Note: If the same tool call keeps failing, or the employee repeats no-effect commands instead of making progress, it automatically wraps up as a safety measure. Give it more specific instructions and try again.

10. An “infrastructure is unstable” banner appeared

If you see this

A notice strip at the top of the chat screen reads “Our infrastructure is unstable right now. Please try again in a moment.”
You see the same wording on web, iPhone, and Android.

What it means

The execution environment your employees work in (the AI work computer each employee uses) is unstable right now. Nothing is wrong with your account, your billing, or your employee settings, and there is nothing for you to fix. This banner is not an error. It’s there to tell you the problem isn’t on your side.

It turns on in two situations:

Your own request just failed for that reason: the banner shows only on your screen.
Failures have piled up across many employees and the service is judged unstable: the banner then shows for everyone who has an employee chat screen open (it does not appear on the dashboard or in team chat). It turns on only when both the number of failures and the share of failures are high enough, so a single hiccup never triggers it.

Check these in order

Wait a moment and retry. Most incidents clear within a few minutes. Just send the same message again.
The banner clears itself. You don’t need to dismiss it or refresh (it does not display a specific time). Note that the banner going away doesn’t necessarily mean things have recovered. Try again, and if it fails again, wait a little longer before retrying.
Don’t change your settings. There’s no need to delete employees, turn tools off, or reconnect anything; that’s not where the problem is.
If it persists. If it keeps showing for more than 30 minutes, let us know through Report an issue (the bug icon). See the Advanced section below for how to report.

What happens to your SU

If the environment could never be obtained, no SU is charged for that time. An employee’s AI work computer is billed only for the time it was actually running. If it did start and then broke partway through, the time it was up until then is still counted.
However, if a request failed partway through, the AI usage that already happened before the failure is still recorded as used. AI call costs are recorded the moment the call is made, and a failure does not automatically reverse them.
If an incident caused unusually high usage, tell us through Report an issue.

Advanced

The content below isn’t needed by most users. Read it only when an issue keeps repeating or you want to dig into root cause.

Reading the last error category in the activity log

The employee’s Logs side panel records the result status of each run (completed / failed / skipped) along with the last error message. When you see the same failure repeating, here’s how to map it to action.

Category	What the user sees	Recommended action
Model call limit	“please try again shortly”	Wait briefly. If it persists, switch the employee’s AI performance tier (Permissions on the web, Settings on mobile)
Model failover triggered	No visible signal (handled silently)	Normal behavior. A standby model within the same AI performance tier took over for one turn, so no action is needed
Tool error	“tool execution failed”	Check the external service connection / permissions the tool relies on
Out of SU	“not enough SU available”	Subscribe or upgrade your plan
Infrastructure (execution environment) outage	The “Our infrastructure is unstable” banner at the top of the chat screen	Wait and retry (see section 10 above)
Awaiting user reply	A closed question card	Open the card and answer, or re-ask the employee

When the same failure keeps repeating

If the same error fires three or more times, this sequence works well.

Pause the employee. Scheduled runs and other automatic runs stop together.
Message the employee: “summarize what you’ve tried and where you got stuck, and save it in memory.” That tidies its working memory.
Reactivate the employee and retry the task in smaller chunks.
If it still blocks at the same spot, look in Adding Tools for an alternate tool combo that gets you past the obstacle.
If none of this resolves it, use the bug icon Report an issue to report it directly. You’ll find it at the bottom of the sidebar and below the employee list, but also next to each individual run in the employee’s Logs tab, the Dashboard, and a team’s full activity feed. Clicking the icon there points the report straight at that run. Opening the report form lets you pick “which run had the problem” from a list of recent runs (skip it if you’re not sure), and once you describe the symptom, the employee’s metadata, recent automatic run history, whether its memory files are intact, and a screenshot of your current screen are attached automatically so the team can quickly check the cause.