How provider charges reach the gateway
Each provider bills differently. The gateway records the cost against the key that made the request, using the provider’s price list.
For a per-token provider, the gateway uses the published price per million tokens:
These figures were read from
GET /v1/model/info on the running gateway. That route
is the authority; this table is a copy and can go stale.
A fixed-price provider shows zero per request. That is correct, not a bug: if the plan is a flat yearly fee, a per-token price has no meaning.
Budgets, keys and teams
The gateway controls spend with two mechanisms:
Why teams matter for cost: if a route charges per request (like OpenRouter), you may not want every team to reach it. The admin can restrict a route to certain teams. A key from a team without access gets a refusal:
Where to see the costs
The admin dashboard athttps://llm.hjc925.cloud/ui shows the numbers:
The request log keeps refused requests too. A refused request shows zero tokens and zero cost, so you can see when someone tried a model their team cannot use.
What to watch for
A sudden jump in spend
A sudden jump in spend
Look at which model jumped. A heavy job pointed at an expensive model, or a looping agent, both show up as a spike against one key.
A key hit its budget
A key hit its budget
The gateway stops the key and returns a budget error. Decide whether to raise the limit or leave it stopped.
A metered route a team should not reach
A metered route a team should not reach
Restrict the route to the teams that are allowed. Access control is the lever; the log shows who tried and was refused.