> ## Documentation Index
> Fetch the complete documentation index at: https://docs.hired.tools/llms.txt
> Use this file to discover all available pages before exploring further.

# Health and errors

> One tool that tells you whether the instance is working, and one that tells you what failed. Neither can see anyone's content.

**Settings → Admin → Health**, or ask:

> Is anything broken on this instance?

## What health checks

`admin_health` is the first tool to call when something is reported broken, and the one to
call on a schedule if you check on the instance at all. Each check comes back `ok`, `warn`
or `down` with a plain-language summary you can read out as-is.

| Check      | Tells you                                                                                               |
| ---------- | ------------------------------------------------------------------------------------------------------- |
| Database   | Reachable, and how long it took to answer.                                                              |
| Migrations | Whether every one finished. The failure that is invisible until somebody hits the screen that needs it. |
| Email      | Whether it is configured, and whether the **last send actually succeeded**.                             |
| Billing    | Whether Stripe is still calling the webhook.                                                            |
| Assistants | When one last made a tool call.                                                                         |
| Errors     | How many were recorded in the last 24 hours.                                                            |

Two readings worth knowing:

* Billing **down** usually means the signing secret in **Admin → Configuration → Billing** is wrong.
* Billing "never called" means the webhook endpoint was never added on Stripe's side.

## What failed

`admin_recent_errors` is the instance's own event stream, newest first: failed emails,
Stripe webhooks that did not verify or did not sync, tool calls that threw, and pages that
errored.

Each entry has a level — `INFO`, `WARN` or `ERROR` — a source, a one-line message, and the
address of whoever's request hit it.

| Source           | Covers                                 |
| ---------------- | -------------------------------------- |
| `stripe.webhook` | Deliveries, including successful ones. |
| `billing.sync`   | Reconciliation runs.                   |
| `email.send`     | Invitations and test sends.            |
| `mcp.tool`       | Tool calls that threw.                 |
| `app`            | Pages that errored.                    |

<Note>
  Pass `level: "ERROR"` for failures only. The default **includes** `INFO` entries such as
  successful webhook deliveries — those are what prove Stripe is still reaching this
  instance at all, so the noisy default is the useful one when you are debugging billing.
</Note>

Entries older than 30 days are removed automatically. Default 50 entries, maximum 200.

<Check>
  **This never contains anyone's content.** When a tool call throws, the tool's name and
  the failure are recorded — the arguments that caused it deliberately are not, because
  those are somebody's career history and every admin can read this stream.
</Check>

## Two questions this answers well

> Did that invite to [priya@example.com](mailto:priya@example.com) actually send?

`admin_user_detail` says whether it left; `admin_recent_errors` with
`source: "email.send"` says why it did not.

> Is anyone actually using their assistant?

The health check reports when a tool call last arrived. `list_connections` on an individual
account shows the same thing per client, including what called in.