Venice x402 marketplace

Pick your model.
Pay by the bite.

Resell unused Venice inference credits over x402 — pay-per-use inference for humans and agents, with no API key to mint, store, or trust.

Browse the menu Start an order

Sitting on idle Venice credits? Sell your credits in the Kitchen →

Pay per tokenSettle exact usage
No API keyPayment is the auth
Agent-readyQuote, sign, call
Key stays server-sideSeller keys never ship

How the teahouse works

Three steps from craving to completion

DiemSum turns inference into an a-la-carte menu. Pick, order, and pay by the bite — for a person at a keyboard or an agent on a loop.

Pick from the Menu
Scan a-la-carte offers from independent sellers — every major Venice model, each with a live per-token price you can compare against the Venice list rate before you commit.
Browse markets →
Place your Order
Send your prompt like any chat — no account to provision, no key to paste. DiemSum routes it to the seller's capacity and returns the model's completion.
Start an order →
Pay by the bite
You get a quote first, then settle the exact usage over x402. Payment is verified before a single seller credit is spent — you only ever pay for what you eat.
See the flow →

One marketplace, two seats at the table

Good for who's ordering — and who's cooking

For buyersOrder side

Eat well, pay only for the bite

Inference without the contract. Browse the menu, send your prompt, and settle exact usage — for yourself or the agent acting for you.

Cheaper, honest pay-per-use — no monthly seat, no minimum spend.
Any Venice model, one menu — compare sellers and prices side by side.
No API key to trust, store, or rotate — payment is the only credential.
Agent-ready by design — discover, quote, sign, and call programmatically.
No raw prompt retention — only sha256 hashes are kept for receipts.

Browse the menu

For sellersKitchen side

Plate up your unused capacity

Idle Venice credits are revenue waiting to happen. Open a Kitchen, set your prices, and let buyers pay you per token — keys stay home.

Turn idle Venice credits into revenue instead of letting them expire.
Your Venice API key never leaves the server — buyers never see it.
Set your own price against the Venice list rate, per model.
Payment is verified before any of your credits are spent.
List, pause, and reprice from the Kitchen — you stay in control.

Open your Kitchen

Built for agents

Your agent can order for itself.

There is no API key to provision and no dashboard to babysit. An agent discovers an offer, asks for a quote, signs an x402 + ERC-7710 payment that satisfies it, and gets the completion back — the whole loop is three HTTP calls. Payment is the credential.

DiscoverGET /api/v1/offers
Quote (402)POST /api/v1/inference
Pay & receive+ PAYMENT-SIGNATURE

Read the docs Point your agent at /skill.md

quote-first inference

# 1. Browse the menu — pick an offerId
curl -s https://diemsum.com/api/v1/offers

# 2. Ask for a quote — no payment yet → HTTP 402 + PAYMENT-REQUIRED
curl -s -X POST https://diemsum.com/api/v1/inference \
  -H 'content-type: application/json' \
  -d '{
    "offerId": "offer_...",
    "messages": [{ "role": "user", "content": "Explain x402 in one sentence." }],
    "buyerAddress": "0xYourPayer"
  }'
# → 402 { quote: { quoteId, quotedMaxUsd }, paymentRequirements: { accepts: [...] } }

# 3. Replay the identical body, signing the quote over x402 + ERC-7710
curl -s -X POST https://diemsum.com/api/v1/inference \
  -H 'content-type: application/json' \
  -H 'PAYMENT-SIGNATURE: <base64 x402 payload>' \
  -d '{ ...same body... }'
# → 200 { completion, receipt }  ·  you pay only the bite you ordered

Why it's safe to order here

Trust baked into the recipe

Verified before a credit is spent

Every payment is verified by an independent onchain facilitator before any inference runs. A seller's Venice credits are only ever spent against a payment that already cleared.

No raw prompt retention

Your prompts are never stored — DiemSum keeps only sha256 request hashes for receipts and dispute resolution. Paid outputs are retained server-side so delivery is guaranteed even when settlement needs a retry.

Right delegation for the orderer

ERC-7715 permissions for humans granting from a wallet, ERC-7710 delegations for agents acting on their own — the same settlement either way.

Built on Venice + Base

Real Venice models behind every offer, settled in USDC on Base. Open protocols end to end — x402 for payment, Venice for inference.

Pull up a chair