One endpoint,
every model.
Tare is an OpenAI-compatible gateway in front of every model you use — ours and the ones you bring yourself. One base URL, one key, and every token accounted for.
Cost attribution
Attribution answers "where did this money go" at a finer grain than an API bill.
Send any of these extra fields in the request body and Tare records them against the call. They are stripped before forwarding, so your provider never sees them.
{
"model": "GLM 5.1",
"messages": [ ... ],
"task_id": 4821,
"task_type": "issue-fix",
"stage_code": "plan",
"action_code": "draft-patch",
"attempt_no": 2,
"usage_label": "acme-corp",
"available_tool_count": 12,
"available_tool_choice_mode": "auto"
}
[!TIP]
Stripping is necessary: OpenAI's own endpoint rejects unknown body parameters outright, so a
BYOK channel pointed at api.openai.com would fail every call, for a reason that looks nothing
like "an extra attribution field was sent".
Charged versus attributed spend
The console breaks spend down by model, by key, by product line and by time. Two numbers to distinguish before reading a report:
- Charged — the amount billed to you by the platform.
- Attributed spend — what the call cost you, including tokens your own provider billed you for directly. Use this value to measure the cost of a given feature.
They differ by design: on a BYOK-only account the first is zero and the second is not.
Retries and tool-call overhead
Spend on retries, and on calls where tools were offered but never used, is tracked separately. Both are usually a small share of the total; when the share rises, they are the cheapest items on the list to address.