Publishing capacity
Declare capacity and price executable offers.
Start from a configured endpoint
Provider capacity belongs to an endpoint that identifies the model, adapter, region, retention, and supported capabilities. Complete Provider onboarding before declaring supply.
Only publish quantities and delivery rates you can fulfill. A model’s public context limit or a catalogue price does not establish capacity.
Create a capacity pool
Open Supply & fulfillment, find the capacity section, and enter:
| Field | Meaning |
|---|---|
| Endpoint | The configured supplier route that will serve requests |
| Name | A recognizable name for this delivery pool |
| Total tokens | The pool’s available token quantity |
| Start and end | When the pool is available |
| Tokens per minute | The supported delivery rate |
| Maximum concurrency | How many requests may run at once |
The time window must cover the obligations you offer. A large total quantity does not override throughput or concurrency limits.
Manage a saved pool
Select Manage pool beside the pool you want to review. The dialog shows the pool name, total capacity in BT, availability start and end, tokens per minute, and maximum concurrent requests. Update the intended fields and select Save capacity changes.
One BT in this form is one billion tokens of declared capacity. Capacity pools and provider portions are separate from the minimum new buyer purchase: an accepted new agreement must still meet its 1 BT minimum.
Accepted agreements protect their reserved capacity and delivery windows. The server rejects updates that undermine existing obligations. If a change is rejected, the original pool remains saved; review the affected commitments before proposing another change.
The form displays local time. Unchanged saved timestamps retain their original precision when submitted. Review changed dates carefully, particularly when an existing delivery window is already underway.
Publish an offer or quote
A firm offer makes defined spot or future capacity available for eligible buyers to accept. An RFQ response prices a buyer’s specific requirement.
Both need a capacity pool, model, allowance, delivery window, rate card, expiry, and terms as applicable. Publish separate input and output rates. Add supported cache or other classes explicitly and preserve any tier conditions.
Do not offer an execution mode or capability merely because the public model source lists it. The endpoint and configured gateway adapter must support it.
Respond with a partial allocation
Open an eligible RFQ and find Respond with a provider quote.
- Response. Choose RFQ for full size to quote the requested quantity. You do not type a size. Choose Quote for a different size, then type that size in BT.
- Prices and terms. Select the capacity pool covering the delivery window, enter the INPUT price and the OUTPUT price in $ / 1M tokens, set the quote expiry (defaulting to 14 days), and add any optional cache/batch rate classes or delivery terms.
The partial size must be a positive whole number of tokens and must fit the request and the available pool. It may be below 1 BT because the buyer combines selected allocations to cover the RFQ, and a new agreement is still at least 1 BT.
Select Submit provider quote. The quote uses your selected expiry (14 days by default) and the requested delivery window. A later revision replaces it; a superseded quote is not a current offer.
The buyer can combine your allocation with other eligible provider quotes. All selected capacity is reserved together on acceptance. A submitted partial quote by itself does not reserve its pool or guarantee that the buyer will select it.
Shared capacity is reserved once
A pool may support multiple offers, but the underlying capacity is shared. Acceptance reserves against the pool atomically so two buyers cannot independently consume the same remaining commitment.
If capacity has been reserved by another acceptance, a later acceptance can fail even when the buyer opened the offer earlier. Issue a current quote or provide another verified pool instead of treating the earlier view as a guarantee.
Respond to a future replacement
A replacement quote must identify the new pool, model, prices, expiry, and compatible terms. The buyer reviews it before acceptance. The service moves only eligible unused, unreserved value and records an agreement revision.
See Switching a future model for the buyer’s view of the amendment.