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

# AI service isn't available

> What the message “Your company's AI service isn't available right now.” means, what you can do, and how Pretty can help.

> **Your company's AI service isn't available right now. Please try again in a moment. If it keeps happening, Pretty support is happy to help.**

## What happened

Your company connected its own AI service to Pretty. Pretty sent your request there and did not get
an answer back. Pretty itself is working; the AI service is briefly unavailable.

Your message is saved and your slides are untouched. Nothing is lost.

## What you can do

1. Press **Retry**. Most of the time the service is back within a minute.
2. Still failing? Wait a few minutes and try once more.
3. Still failing after that? [Contact Pretty support](/help/support) with the time it happened. We can
   see which service did not answer, tell you what is going on, and reach out to the right people at
   your company if needed.

If you need results right away, a workspace admin can temporarily switch the workspace to Pretty's
own AI access under [Model providers](/admin/model-providers) and switch back later.

## For admins

The workspace uses bring-your-own-key, so requests go to the service you connected in
[Model providers](/admin/model-providers) instead of Pretty's own access. Pretty shows this message
when that service answers `502`, `503` or `504`, or the connection drops before a reply.

<AccordionGroup>
  <Accordion title="Your LiteLLM proxy" icon="server">
    `curl -i https://<proxy host>/health` returning 502/503/504 or hanging means the proxy itself
    is down (restarting, out of memory, broken load balancer). Everyone in the workspace is affected
    at once. Restart or repair the deployment, then confirm `/health` answers `200`.
  </Accordion>

  <Accordion title="The model provider behind it" icon="cloud-xmark">
    `/health` answers `200` but the proxy logs show errors from OpenAI, Anthropic, Bedrock or Google.
    Check the provider's status page; once their incident clears, requests recover on their own.
  </Accordion>
</AccordionGroup>

A different message, *"The selected model isn't available on your LiteLLM proxy"*, is a
configuration issue rather than an outage. See [Troubleshooting](/help/troubleshooting#while-the-agent-works).

Pretty support is glad to help with either case: send us the time and the message ID and we can
tell you which host answered what.