Troubleshooting

Find your problem below. If you see an error code, Error codes explains every one.

The first request takes a long time

What happened: the endpoint was asleep. The first request wakes it, which takes a few seconds on a warm GPU and a few minutes when a new GPU starts. Start-up is never billed.

How to fix it:

• Every request waits on the open connection for up to 8 minutes while the model starts, then answers.
• If you sent X-BitsRun-Wait: 0, you get 503 model_starting after about 45 seconds. Retry after 30 seconds.
• A very large model can take longer than 8 minutes the first time. Then the answer carries model_starting in its body; send it again.
• Select Start on the endpoint's card a few minutes before you need it, or raise Sleep after so it stays awake between bursts.
• For no cold starts at all, turn on always-on. It is billed every second it is on.

Read Cold starts and Sleep, always-on and priority.

401 unauthorized

What happened: the API key is missing, mistyped or revoked.

How to fix it:

1. Check the header is exactly Authorization: Bearer brk_..., with one space after Bearer.
2. Check the environment variable holds the key: echo ${BITSRUN_API_KEY:0:8} should print brk_ and four more characters.
3. If the key was revoked or lost, create a new one on the API keys page.

See API keys.

402 payment required

What happened: your balance reached $0, so endpoints stopped.

How to fix it: add credit on the Billing page. Endpoints work again within seconds; nothing was deleted. Set up a balance email or monthly top-up so it does not happen again.

404 on my endpoint URL

What happened: BitsRun answers 404 for anything it cannot show you, on purpose, so nobody can probe for endpoints. Common causes:

• A typo in the URL, or a missing /v1 (use https://<id>.endpoint.bitsrun.com/v1/chat/completions).
• A path the endpoint does not serve, such as chat on an embedding endpoint.
• A key limited to a different endpoint.
• The endpoint was deleted.

How to fix it: copy the URL from the endpoint's card (or bitsrun status <endpoint>), check the path, and try a key that works with every endpoint.

My model will not deploy

What happened: the dashboard or command line shows a message with a code such as model_not_found, model_gated, model_custom_code or gpu_too_small.

How to fix it:

1. Check the model before you deploy: select Check model in the dashboard, or run bitsrun size <model>.
2. Read the message. It names the fix, for example accepting a repo's terms or picking a GGUF version.
3. Look up the code on Error codes.

The most common reasons and what to pick instead are on Models that cannot be deployed.

The endpoint says no GPU was free

What happened: every GPU of that size was busy for 30 minutes (no_capacity). You were not billed while it waited.

How to fix it: try again in a few minutes, choose another GPU size, or turn on priority so cold starts launch on a dedicated GPU right away.

bitsrun login does not open a browser

What happened: you are on a server over SSH, or the computer has no browser.

How to fix it:

bitsrun login --no-browser

It prints a link. Open it on any device where you can sign in, approve the computer, and paste the code back into the terminal. Codes last 5 minutes and work once; if yours expired (invalid_code), run the command again. See Command line.

Answers stop partway

What happened: the reply reached max_tokens, or the conversation filled the model's memory.

How to fix it: raise max_tokens (up to 32,768), drop older messages, or raise Memory length under Performance. If your app closed the connection, the model stops too.

Still stuck?

Write to support@bitsrun.com. Include the request id from the error, the endpoint name, what you tried and when. A real person answers.