Platform

Support

Where to ask, and what to include so that the answer takes one round trip instead of four.

Where to ask

Channel	Best for
hello@lobstack.ai	Everything: questions, bugs, account and billing. It reaches one person, which is also why it gets answered.
github.com/lobstack-ai	Issues against published code.
x.com/lobstack	Release announcements. Not a support queue.

Check the status page first, then tell us anyway/status is computed from the requests the Gateway actually served over the last thirty days, so it can tell you whether failures are ours. It cannot tell you about a failure nobody has hit yet. We would rather field a duplicate report than have three people each assume the other one filed it.

There is no DiscordThis page linked discord.gg/lobstack and that invite is dead — the API answers Unknown Invite. Email is the support channel. If a community server opens, this is where it will be linked from.

The request id

For anything involving the Gateway, the single most useful thing you can send is the value of x-lobstack-request-id. It is on every response, success or failure, and it is allocated before authentication runs, so even a 401 carries one. Error bodies repeat it as error.request_id.

That id is the key to a stored trace row: the model you asked for, the model that served you, the provider, the status code, the error class, the latency, the token counts and the priced cost. With it, a report is a lookup. Without it, it is an interview.

Getting the id from a requestbash

curl -sS -D - -o /dev/null \
  https://www.lobstack.ai/api/gateway/v1/chat/completions \
  -H "Authorization: Bearer $LOBSTACK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"auto","messages":[{"role":"user","content":"hi"}]}' \
  | grep -i x-lobstack-request-id

If you did not capture the headers, you can still find the request in Console → Logs, which lists one row per request newest first and filters by key, model, status and period; expanding a row shows the id. Log the id on your side from the start and you will not have to.

What to put in a report

request_idstringrequired: From x-lobstack-request-id. One is enough; three showing the same failure is better.
key prefixstring: The lsk_live_… prefix only, never the whole key. The prefix is not secret and identifies the key; the rest of it is the secret and we cannot use it for anything.
what you sentstring: The model you asked for, whether you streamed, and whether tools were in the request.
what came backstring: The status code and the error.type, or a description of how the response was wrong if it was a 200.
whentimestamp: With a timezone. "This morning" is four hours wide.

Things worth checking first

Symptom	Most likely cause
401 with “invalid API key”	The key does not match the shape `lsk_<env>_…`, or it is from a different environment. Revoked and expired keys say so explicitly.
403 mentioning a scope	The key was minted without `inference`. Scopes are fixed at creation; mint a new key.
402	The monthly allowance is exhausted. `x-lobstack-quota-meter` says which allowance — dollars of model spend, requests, or legacy messages — the other quota headers say by how much, and retry-after says when it resets.
temperature appears to be ignored	Some models reject sampling parameters and the Gateway drops rather than forwards them. Check for `x-lobstack-dropped-params`.
A cheaper model than expected	Routing scored the prompt into a lower tier. `x-lobstack-complexity` and `x-lobstack-tier` show the decision; the routing preview reproduces it for free.
x-lobstack-savings-usd is empty	There was no baseline. Either nothing was routed away from, or a price was missing for one of the two models. On `"auto"` a priced plan does get a baseline — check `x-lobstack-baseline-reason`, which reads `plan_ceiling` in that case.
a savings figure that looks too good	Read `x-lobstack-baseline-reason`. `plan_ceiling` means the comparison is against the most expensive model your plan allows, not against a model anyone asked for.

The six error classes and what to retry are on Errors & retries.

Security reports

Send anything exploitable to hello@lobstack.ai rather than to a public issue. Include enough detail to reproduce, and the request id if the Gateway is involved. More context is on the security page.