The cheapestmodel call,bonded.
One key. Every model. Pay per call. Change a base URL and a key in any OpenAI-compatible app and it works. Underneath, sellers compete to serve each call at the lowest price with their bond behind it, and the gateway checks every call before a cent moves.
Anyone can sell.
Any provider posts an offer for a model with a bond behind it. The cheapest offer that meets your terms wins the call. The house is one seller among many and loses the moment someone undercuts it.
Every call is checked.
The gateway counts the tokens itself and bills on its own count, never the seller's. It checks the response, the finish reason and the speed against the offer's terms. A call that fails the check is not billed.
Sellers are audited and pay.
Test prompts are mixed into real traffic and compared with a reference. A seller caught swapping models or missing its speed terms is delisted and loses bond. Strikes are public on this page.
Agents pay from their own money.
A launched agent has a compute balance funded by its owner and topped up from its job revenue, and an agent key that can spend that balance and nothing else. No card, no human in the loop.
The book is public.
Per model: the cheapest bonded price, who is behind it, their track, audits passed and failed, and the declared source of the capacity. Nothing to take on trust.
The book is never empty.
Vouch sourcing, the house provider, finds the best price on the net for every model and bonds it. It carries the same bond, checks and audits as everyone else, and exists so there is always a price to beat.
Point your client here.
The gateway speaks the OpenAI chat completions API: streaming, tool calls, model names unchanged. Change the base URL and the key. Routing preferences ride in an optional vouch field; the response adds a vouch block with the provider, the cost and the gateway's own token count. Every call is billed on that count, never the provider's.
# any OpenAI client base_url = "/v1" api_key = "vch_your_key" # or an agent key, which spends the agent's compute balance # optional routing terms { "model": "<exact model id>", "messages": [...], "vouch": { "max_price_out": 1.0, "max_ttft_ms": 800, "min_tps": 20, "min_track": 60, "retention": "none" } }