solcleus.com

Endpoint reference

The endpoints this host serves. Prices and behaviour below are what the services report about themselves; the canonical machine-readable copies are /v1/info, /.well-known/x402 and /llms.txt. Where this page and those disagree, they are right — and please tell us, because a wrong claim here is a defect.

Start with the demo, it costs nothing: GET /v1/burst/demo shows both outcomes — an answer that verifies and gets charged, and a wrong answer that gets caught and does not — with no key, no wallet and no payment. Read the honesty field in the response before anything else; it states exactly what is real and what is stipulated. See what the demo is and isn't below.

Verified Burst

Pay-per-checkable-decision inference for agents. It samples best-of-N and charges a service fee over x402 only when the verifier passes. A miss is free. Token costs are billed to your own provider key and are not marked up.

Endpoint	What it does	Price
GET /v1/burst/demo	Both outcomes, end to end. Needs nothing from you. Rate-limited and cached — see below.	free
GET /v1/info	Full service descriptor: resources, input schema, verifier modes, safety limits, MCP package.	free
GET /v1/quote	Current base price with its breakdown.	free
POST /v1/burst	Buy a verified decision. Returns `402` with a payment challenge until settled.	from $0.004
POST /v1/best-price	"Best Price Now" — one micropayment buys one broad, real-time best-price search.	$0.004
GET /.well-known/x402	x402 discovery document (also at `/.well-known/x402.json`).	free
GET /burst	The human-readable page for the same service.	free

What it costs, and why you may see two numbers

/v1/quote returns the base price — $0.004, broken down as $0.002 burst + $0.001 candidates + $0.001 verify. The price scales with your configuration: more samples or a stronger verifier costs more. The demo's confirm scenario runs n=3 with an independent judge and reports would_charge_usd: 0.0045. Both numbers are correct; they describe different requests. Quote your actual request before you commit.

Paying, and what "confirmed" means

Settlement is x402 on Base mainnet (eip155:8453) in USDC, through a facilitator we run ourselves rather than a third-party service.

The honest version, because the code is more careful than a slogan: settlement is normally confirmed against the chain. But if the RPC endpoint is unreachable, the broker resolves the receipt as unknown and trusts the facilitator rather than blocking a legitimate payment. So "confirmed on-chain" is the normal path, not a guarantee on every request. The single public RPC this depends on is a known resilience gap and is written down as one.

Keys and wallets

You need a funded wallet — that is the spend ceiling, enforced on-chain rather than by our promise. A provider key goes in the X-Provider-Key header and is used for generation only.

Your first 3 bursts can run without a provider key (free_trial_bursts: 3), on ours. They are per wallet and the payment is still validated, so a funded wallet is required even for those; after three, bring your own key.

# see both outcomes, free, no credentials
curl https://solcleus.com/v1/burst/demo

# what the base price is right now
curl https://solcleus.com/v1/quote

# a real burst: 402 + payment challenge until settled
curl -X POST https://solcleus.com/v1/burst \
  -H 'Content-Type: application/json' \
  -H 'X-Provider-Key: <your provider key>' \
  -d '{"request":"...","verifier":"self_consistency","n":3}'

There is also an MCP server: pip install verified-burst, command verified-burst, tool buy_verified_burst.

Verifier modes

Passed as verifier. This is what decides whether you are charged.

Mode	How the answer is checked
self_consistency	N-of-M samples agree. Pair it with `answer_key` — without one, agreement is measured on the full answer text and rarely matches on prose.
judge	Adversarial LLM check.
independent_judge	A different model family checks the answer, decorrelated from your own model's blind spots.
independent_quorum	Multiple models from different families must agree (k-of-M; pass `quorum_k`). Independent of your model, not of our infrastructure — they are reached through one upstream account, so they share that single point of failure.
none	No gate — always charged.

Pass a candidate to have your agent's own answer verified with no generation at all.

What the demo is, and what it isn't

It is a demonstration, not a live benchmark, and it says so itself:

Responses are cached (about 15 minutes), so the latency_s you see is from the cached run, not yours.
In the catch scenario the wrong answer is stipulated — it stands in for a model hallucinating. The judge calls that catch it are real and independent. The response's honesty field says precisely this.
Limits: 6 per IP per day, 500 per day in total.

To see it run on your input, use a burst — the first three need no provider key.

Records anyone can check

Endpoint	What it is
GET /transparency.json	The append-only log's public state.
GET /.well-known/solcleus-attestation	The Ed25519 public key every signed result verifies against.
GET /catalog.json	Published results. Opt-in only — nothing appears here that the submitter did not agree to publish.

Other pages on this host

Path	What it is
GET /	What Solcleus accepts and how it charges. Job submissions start here.
GET /eleven	Eleven Ball.
GET /cerebras	Cerebras consulting page.
GET /robots.txt	Crawl policy.
GET /healthz	Liveness of this host.

Getting a decision from a person

Describe the job on the front page and submit it. Replies are written and sent by hand, by one person — there is no automated mailer here, so allow several days, and more if you catch a bad week. If a job is declined you will be told which gate it missed, not just that it was declined.

If you have heard nothing and want to chase it, resubmit with the same contact address and say so in the description — that is currently the only inbound channel.