How it works
From a private request to a working API key.
Three steps
Request, negotiate, run.
Step 1
Request
Describe what you need: model, input and output tokens, delivery window, rate limits, region and budget. Send it privately to the verified suppliers you choose, or ask for terms on a listed lot.
Your request is visible to you, the suppliers you invite and Cairva operations. Nobody else.
Step 2
Negotiate
Suppliers reply with firm quotes in a private deal room. Every counteroffer is a new numbered version with the full terms, so you always know exactly what you're accepting.
You accept one specific version. The supplier countersigns, payment is confirmed, and the capacity is locked to you.
Step 3
Run
Activation checks pass and your agreement becomes an allocation. Create a scoped API key, point an OpenAI SDK at Cairva, and send requests.
Each request is metered against the agreement. You and the supplier see the same usage and the same remaining balance.
What a lot is
What you actually own.
You get
- Access to one model on one route, for a fixed UTC window
- A contracted input and output token allowance
- Rate limits written into the agreement and enforced by the gateway
- Scoped API keys and a meter both sides can see
You don't get
- Ownership or control of GPUs
- A speed guarantee, unless the agreement is reserved or dedicated capacity
- Credits from another provider's account
You can't
- Transfer or resell the allocation
- Spend past the allowance
- Call a model the agreement doesn't cover
Capacity classes
Three kinds of capacity.
Metered allowance
Offered at Genesis
- What you're buying
- A prepaid token allowance on a shared route
- Dedicated GPU
- No
- Minimum tokens/sec promised
- No
- Rate limits enforced by the gateway
- Yes
- Can run on a sandbox route
- Yes
- Available at Genesis
- Yes
- On the market today
- None listed right now
Reserved throughput
- What you're buying
- Provisioned capacity with a measured floor
- Dedicated GPU
- Depends on the deployment
- Minimum tokens/sec promised
- Yes
- Rate limits enforced by the gateway
- Yes
- Can run on a sandbox route
- No
- Available at Genesis
- Only once a supplier has a measured reserved deployment
- On the market today
- None listed right now
Dedicated window
- What you're buying
- A whole deployment for a fixed window
- Dedicated GPU
- Yes
- Minimum tokens/sec promised
- As stated in the agreement
- Rate limits enforced by the gateway
- Yes
- Can run on a sandbox route
- No
- Available at Genesis
- Only once hardware is actually set aside
- On the market today
- None listed right now
Agreements
The life of an agreement.
- 1 of 5Awaiting countersign
You accepted a quote version. The capacity is reserved on the lot while the supplier countersigns.
- 2 of 5Awaiting payment
An invoice is issued. You pay it outside Cairva and record the reference. Sandbox agreements skip this step.
- 3 of 5Provisioning
Cairva operations confirmed payment. Activation checks run: supplier verified, lot listed, route enabled and healthy.
- 4 of 5Active
The allocation is live. Create keys and send requests until the allowance is used or the window ends.
- 5 of 5Completed
The full allowance was used. The meter, usage events and reports stay available to both sides.
Agreements can also end as Expired when the window closes with allowance left, Cancelled before activation, or Disputed while operations review a delivery problem.
Privacy
Private by default.
- Who sees a request
- You, the suppliers you invite and Cairva operations. It isn't listed or indexed anywhere.
- Supplier identity
- A supplier can choose to stay undisclosed until you accept their quote. Only verified suppliers can quote, so you know they've passed Cairva's checks either way.
- Prompts and outputs
- Not stored by default. Gateway logs keep the request id, token counts, latency and status.
- Prices on the market
- Hidden unless the supplier chooses to show an indicative price. Negotiated prices are never public.
Read the privacy policy and the terms of service.
Questions
Questions about terms.
The short version. Your agreement's own terms always apply, and you see them in full before you accept.
Can I resell or transfer an allocation?
No. An allocation is a service entitlement for your organization, bound to one model, one route and one delivery window. It isn't a token, a security or a transferable credit, and it can't be moved to another account.
What happens to unused tokens when the window ends?
The allocation expires and stops serving requests. Whether anything is refunded or rolled over depends on the cancellation and under-delivery terms in your agreement, which you see before you accept.
What if I go over my allowance?
You can't. Cairva holds tokens before each request and caps max_tokens at what's left. When the allowance runs out, requests return a 402 and nothing more is charged.
What if the supplier under-delivers?
Open a dispute from the agreement. Operations review the metering evidence both sides already see and apply the remedy written into the agreement.
Can I cancel?
Either side can cancel before the agreement goes live, and the reserved capacity returns to the supplier's lot. Once it's active, the cancellation terms in the agreement apply.
How do I pay?
Against an external invoice. Cairva operations confirm the payment before activation. Cairva doesn't hold card details, wallets or balances.