> ## Documentation Index
> Fetch the complete documentation index at: https://docs.r3al.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Troubleshooting

> Understand model status and recover a demo without repeatedly starting compute.

| Status | Meaning | What to do |
| - | - | - |
| Not connected | No serving runtime is configured. | Register your model and ask R3AL to connect the runtime. |
| Offline / Demo is off | No active managed session, or the connected server is stopped. | Start the configured demo; for a private runtime, check its server. |
| Loading model | The model is still preparing. | Wait for the displayed stage to complete. Cold starts can take 5–10 minutes. |
| Ready | Model identity and health checks passed. | Open chat or send an API request. |
| Unavailable | The health check failed, the identity differs, or the checkpoint failed to load. | Refresh status. If loading failed, end the demo before retrying. |
| Stopping compute | Shutdown has been requested. | Wait for confirmation; starting another session is disabled. |

## RunPod says Running, but the platform is loading

Running describes the worker container. Ready describes the model inside it. Downloading and loading weights happen after the container starts. The platform reports these stages and waits for the expected model identity before allowing inference.

## Failed requests in RunPod

The provider's request count may include health checks and transport failures, not just model generations. Use the platform status and the actual chat/API response to understand whether inference failed. Avoid repeatedly pressing Start demo.

## A request fails after Ready

A long conversation may exceed the model's context window. Start a new chat or shorten the input. If the demo deadline expired, wait for shutdown and start a new session. For persistent failures, provide R3AL with the project ID, approximate time, and visible error—never an API key or confidential prompt.

## API authentication errors

Use an account-linked API key from a user who still belongs to the project team. Set `model` to the project ID. Unrelated projects return 404. Streaming requests are not supported.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.