This page is for the admin. It explains how to watch what the gateway is used for, what it costs, and how to keep each person inside their limits. Everything here is in the admin UI at https://llm.hjc925.cloud/ui, or through the management API on https://llm.hjc925.cloud.

​
Keys

Each person has their own API key. The gateway stores only a hash of the key. Nobody can read the key again after it is created.
1

See all keys

Open the Keys page in the admin UI. It lists every key, the person it belongs to, and its spend.
2

Revoke a key

Delete the key. Access stops at once. The other keys keep working.
3

Make a new key for the same person

Create a key and send it to the person out of band. The old key does not work anymore.
The server keeps only a hash of each key. If you close the window without copying the key, you cannot see it again. Make a new one instead.

​
Usage and spend

Every request is logged against the key that made it. The admin UI shows, per key and per user: The gateway also keeps daily and monthly rollups. These are the tables that power the charts in the admin UI.

​
Budgets and limits

You can put four kinds of limit on a key or a user:
1

Open the user or key

In the admin UI, open the user’s row or the key’s row.
2

Set the budget

Enter a max budget and a budget duration. A monthly budget of 20 means the key resets to 20 on the first of each month.
3

Set rate limits

Enter RPM and TPM values if you want to cap request speed as well as total cost.
4

Save

The limit applies from the next request.
When a key hits its budget, the gateway returns a budget_exceeded error. The person sees a clear message, not a confusing failure.
Start everyone on a small monthly budget. Raise it when the person asks. A budget that is too small is a five-minute fix; a budget that is missing is discovered at the end of the month.

​
User preferences

Each user has a profile in the gateway. The admin can set, per user:
If a user should not reach a model, remove it from the user’s allowed list or from the team’s model list. A key can only call what its team and its user role allow. Do not rely on the key holder “being sensible”.

​
What a key holder can see

A key holder signs in at https://llm.hjc925.cloud/ui. What they see depends on their role: A key holder cannot manage the gateway or see other people’s keys. That is the admin’s job.

​
What the admin sees that users do not

The admin pages — models, pools, fallbacks, budgets, MCP, and the full request logs — are visible only to the admin role. Internal users see their own slice.

​
Monitoring habits

Once a week is enough for a small group. The rollup tables make it quick to spot a runaway key.
Look at which model jumped. A person may have pointed a heavy job at an expensive model, or left an agent loop running. Check the key’s recent requests.
Decide whether to raise the budget or leave it. If the person is mid-task and the spend is legitimate, raise it. If the spend looks wrong, leave the key stopped and investigate.
The gateway can send spend alerts. Set them up per key or per team in the admin UI.

​
Pool health

Keys and spend are one half of monitoring; the other is whether the models behind a public name are actually answering. That lives on its own page: Companion -> Pool status. The page refreshes every 15 seconds and that refresh is free — it reads the gateway’s cached health rather than calling anything. Check now runs a live check, which really calls every member of every pool, so it costs one request per member and is deliberately a button.
The gateway publishes no list of the entries it is currently cooling down, so the page reports what failed its last health check instead — the closest available signal — and labels it that way. A member that has never been checked and has never failed a real call reads as unknown rather than healthy, which is the honest answer rather than the reassuring one.