When something goes wrong

Before hunting through settings, say what happened in the chat: "the reply has been thinking for ten minutes".

First, ask your assistant

Your assistant can repair most of what goes wrong: a missing chat, a stale model list, a disk filling up, a stalled job. It looks before it changes anything, fixes one fault at a time, and asks before anything that could cost you work. Ask from the page where the trouble is, because the in-app help knows which screen you are on.

It is not replying, or it is slow

1. Read the message where the reply stopped. "Response stopped" means the turn was cancelled, not broken: send it again.
2. Open the provider status panel from the top bar button: red means an account is throttled, out of credit or signed out.
3. If it is throttled, pick a model on another provider from the chat composer, or wait for the reset time quoted.
4. If every message fails at once, ask whether the provider is having an outage.

The next message should run normally.

A provider sign-in has died

An expired sign-in gets a red Signed out tag on its card and an alert row on your Dashboard.

1. Go to Admin > Providers (/admin/providers) and find the tagged card.
2. Press Test. It forces one real check, so a dead sign-in reaches its verdict in one click.
3. Sign in again from the card. Only you can do this: no refresh, cache flush or agent can revive it.

The tag clears itself and the usage bars return. Until it does, your other accounts carry the work.

The machine is full or under pressure

1. Go to Admin > Resources (/admin/resources). The pill beside the title reads ok, elevated, high or critical.
2. In Storage, click a category to see what is filling that disk.
3. Ask for a clean-up, or press Clean up storage there, which hands the job to your assistant. It reclaims only caches, expired backups and old logs, and asks before anything else. Read its plan first.
4. For memory rather than disk, stop heavy jobs: "stop the research job", or Stop on the job's page under Automations > Jobs (/jobs). Stop keeps the work for later.

Pressure should fall back to ok. If a disk gets critically low, Player One stops agent work itself and keeps your main chat open for the cleanup.

A job or a flow is stuck

1. Open Automations > Jobs (/jobs) and the Needs input view (/jobs/view/needs-input). A job waiting on a decision sits there; answer in the chat it started from.
2. Ask your assistant for the job's status. One that looks frozen is often waiting for model capacity, and it will say when it resumes.
3. Say "resume it", or press Resume on the job's page. Stopping parks work rather than destroying it, so the job comes back with its tasks, workers and files intact.

Activity should start again. A flow run that was waiting for a person or a timer carries on after a restart; one interrupted in the middle of a step is not picked up, so start that one afresh.

A channel stopped delivering

1. Ask your assistant to send you a test message on that channel. If nothing arrives, look at Admin > Messaging (/admin/channels), where each connection has a card with its live status and a Test button (email has none, so check it by sending yourself a message).
2. If the card shows an error, ask for it to be repaired, or press Repair on the card. If that fails, disconnect and set it up again: the last step is yours, such as scanning a code.
3. If alerts rather than conversation stopped, check Admin > Alerts (/admin/alerts), whose History tab shows what was sent.

The card should read as connected and your next message arrive.

You were signed out

1. Sign in again on your installation's sign-in page. A tab you are using renews its sign-in by itself, but a browser nobody has used for a day asks again.
2. If it happens while you are working, open Admin > Devices (/admin/devices) from another browser. A browser that was forgotten there has to sign in again, and one that was revoked there is refused from then on.
3. Clearing the browser's data for the site, or a private window, also means signing in again, because the sign-in is kept in that browser.

Opening several tabs, an update and a restart do not sign you out. Your installation has the rest.

The page looks wrong after an update

1. Hard refresh: Ctrl+Shift+R on Windows or Linux, Cmd+Shift+R on a Mac.
2. Try a private window, another browser or another device. If it looks right there, the fault is on the first device.
3. If a red Required update banner is showing, finish the walkthrough rather than skipping it: your assistant repairs what the new version needs.

The missing menu or button should come back.

Restart and caches

Flush Caches on Admin > Runtime (/admin/gateway) clears the model and provider caches without a restart, and costs nothing. Use it when the model list looks stale after you connect an account or change a plan.

Restart Runtime is a light restart: the page drops for a few seconds and comes back. It costs any reply that was mid-sentence and any flow that was running, though jobs resume. It is yours alone to press.

Your version numbers are on that page too, the first thing to quote in a report.

A command keeps asking, or a scheduled run was refused

1. Open the agent's page: Manage > Agents (/agents), the agent, then Advanced > Permissions. The card's Change what link goes straight there.
2. For one program, add it to Allowed for good, or answer Always the next time the card asks.
3. For less asking in general, choose a wider mode, such as Most things. A scheduled task or a job has nobody to answer a card, so the programs it needs must be allowed before it runs.

Stay in control explains the modes and the card.

Report a bug

If none of that helps, ask your assistant to report it.

It asks one short question at a time: what you were trying to do, what happened, which screen, and a name and email for the reply. A link to the chat and permission for support to look are optional.

Nothing is sent until it has shown you the finished report and every private detail in it, and you have said yes. Passwords, keys, payment or health details and whole conversations are never included.

For a fault it first asks you to install any update offered on your account at 1p.dev, because support handles the current version only. Afterwards you get the reference and a note of what was sent. No fix and no date are promised.

Sidekick says no

The agent tells you what refused it and what changes it: the access to choose, or the tick box to tick, on that connection's page in the app. A computer your assistant calls paired but not connected only needs Sidekick opened and signed in on it. Connect your own computer has the rest.

Where next

• Your installation and Help inside Player One.
• Models and providers, Connect your AI accounts, Connect Telegram, WhatsApp, Slack or email.