Every loot drop, gacha pull, and agent decision gets a cryptographic receipt — sealed before the outcome is known, verifiable by any player, any partner, anyone. Prove your game is fair. Prove your agent did what it claimed.
Works for agent decisions and game outcomes. Verify anything at verify.fairseal.io.
One line in your agent config:
No setup. No account. MCP-compatible with Claude, Cursor, Cline, and others.
How It Works
From commitment to receipt — every step is independently re-runnable. Works identically for AI agent decisions and game outcomes.
Lock your rule and inputs into a hash before the outcome is observable. The commitment provably precedes the beacon entropy it uses.
Wait for a public drand beacon. Verifiable randomness from Cloudflare and Protocol Labs — neither you nor FairSeal controls it.
Derive the selection from beacon entropy plus your committed rule. Deterministic — given the same inputs, anyone gets the same result.
Anyone re-runs every step. Per-version cryptographic checks: drand beacon or Wesolowski VDF entropy, EIP-191/Ed25519 signed receipts. No trust required — the math is the verification.
For AI Agents
Agent-to-agent commerce has a fundamental accounting problem. When your system and theirs transact, each side holds its own audit trail — but an audit trail the keeper can rewrite isn't evidence to the other side. Re-verifying every incoming claim at machine speed burns tokens and reintroduces the latency you were trying to avoid. Skipping verification means accepting risk. Neither option scales.
FairSeal sits between parties, not inside either one. Its financial interest is orthogonal to the content of any receipt — it has no stake in what the receipt says, only in issuing it correctly. That structural position is what makes a receipt credible to both sides. (The same reason sports referees cannot own teams, and auditors must be independent of their clients.)
Every agent action gets a cryptographic receipt. It answers three questions: who made the claim, what was committed, and before what — the commitment was sealed before the outcome was observable. Anchored where neither party can quietly modify it.
Paste any receipt at verify.fairseal.io and re-run every check yourself. No trust in FairSeal required — the proof is self-contained.
Pay per call via x402 — USDC on Base, no account needed. Machine-readable payment instructions on every 402 response. Agent-to-agent transactions don't need a billing portal.
Receipts prove provenance, not correctness — a fully receipted agent can still be wrong. Audit your agent's reasoning separately.
What verification does and doesn't establish
Four distinct states — none implies the others. Receipt authentic ≠ commitment fulfilled ≠ claim independently substantiated ≠ unknown.
| State | What it establishes | API signal | What it does NOT establish |
|---|---|---|---|
| ✓ Authentic | Signature verified; entropy source confirmed; commitment structure cryptographically sound. FairSeal issued this receipt. | result.valid === true | That the agent's decision or game outcome was correct, fair, or in your interest. A valid receipt can accompany a wrong answer. |
| ⛓ Anchored | The batch seal is published on Base — timestamped immutably on a public blockchain. No one can backdate or rewrite it, including FairSeal. | result.anchored === true | That future receipts from the same agent will also be valid. Every receipt earns its own seal independently. |
| ◎ External | The agent's claim is independently substantiated by a third-party or verifiable external source that confirms the stated outcome or decision. | — (outside FairSeal scope) | FairSeal does not establish external claim correctness. Audit the agent's data sources and reasoning chain separately. |
| ✗ Unknown | Verification failed or evidence is absent. The receipt cannot be confirmed authentic. Treat as unverified — do not act on unverified claims. | result.valid === false | Unknown does not imply intent to deceive — but it must never be upgraded to "verified" without evidence. Fail closed. |
For Developers
Your agent hashes its input and decision locally — content never leaves the machine. Submit only the hashes to FairSeal: Merkle-batched and anchored on Base mainnet. Anyone can verify offline that this exact decision existed at this exact time.
Pay via x402 with no signup, or request an API key. AWS Bedrock AgentCore pays autonomously — no middleware needed.
Merkle proof + Base tx hash. Use @fairseal/verify or verify manually with any SHA-256 tool.
Add @fairseal/mcp-server to Claude Desktop in 60 seconds. Three tools: hash, notarize, poll.
For Game Studios
Your players can verify every outcome. Your competitors can't say the same.
The decision-maker who buys FairSeal is your compliance lead — the person who must prove fairness to regulators, auditors, and store reviewers. Drop-rate disclosure is mandated in CN, TW, and JP, and required by Apple and Google. A verifiable receipt makes compliance a property of your architecture rather than a manual audit process.
Faster than a console.log. Your game won't notice.
Keep your drop-rate logic private. FairSeal proves the commitment and the entropy source — the outcome derivation stays yours.
Smaller than a thumbnail. Attach to any game event.
Drop-rate disclosure is mandated in CN, TW, JP, and by Apple/Google. A Committed Selection Receipt (CSR) gives regulators, auditors, and store reviewers verifiable evidence of your committed selection process — without exposing your actual drop table. Evidence supports compliance review; it is not itself a certification.
Packages
Start with @fairseal/mcp-server for AI agents. Start with @fairseal/commit for game outcomes. Add verification and anchoring as needed.
MCP server for AI agents. Verifiable entropy, provably-fair selections, and receipt verification — one line in your agent config.
Game API for slots, gacha, and loot boxes. Real drand entropy in v0.2.
Committed selection receipts. Lock rules before the outcome. The core primitive — build your own integrations on top.
Independent verification. Validate receipts, check on-chain anchors.
Cryptographic primitives, signing, and Merkle trees.
Pricing
Trust scores tell an agent whether to pay. FairSeal receipts are the evidence a score can be audited against.
Agent payment platforms now score every endpoint before a transaction settles — and at millions of transactions per day, no human reviews those decisions. But a score is an inference, computed from signals that change and disappear. When a decision is challenged later, the score alone can't testify.
FairSeal notarize turns any agent decision — including a trust decision — into a cryptographic receipt: input hash, decision hash, signed by an independent party, Merkle-anchored on Base. Anyone can verify it offline, years later, without trusting the agent, the scorer, or us.
We don't score endpoints, and a receipt doesn't make a decision correct. It makes the decision provable: what was decided, on what inputs, and before what point in time. More in the FAQ →
USDC on Base — no account, no signup. Machine-readable on every 402 response. Agents discover and pay autonomously.
Run every proof check locally. No key, no account, no x402 — the SDK does the math. Also available at verify.fairseal.io for browser-based verification.
Lock a rule to a future VDF epoch before its output is known. Retrieve the committed epoch output after computation — permissionless, anyone can verify.
Submit a 32-byte payload hash to a Merkle batch on Base mainnet. Returns a tamper-evident receipt with Merkle proof and on-chain transaction hash.
Retrieve a pre-computed verifiable random number with Wesolowski RSA-2048 VDF proof. For demos and low-stakes sampling. Production: use Commit + Reveal.
Anchor a structured AI agent decision or audit record to a Merkle batch on Base. Returns a receipt with schema validation, Merkle proof, and on-chain anchor.
Verify a Verifiable Execution Output receipt — checks EIP-191 or Ed25519 signature and on-chain anchor. Returns pass/fail + evidence chain.
Pattern-based PII detection (regex only, no cryptographic receipt): emails, US/Canada phone numbers (NANP only), unformatted credit-card numbers, US SSNs, IPv4, 64-hex strings (also matches SHA-256 hashes -- review before acting). Partial national-ID coverage -- alphanumeric formats (e.g., Taiwan) and Korean 6+7-digit format not matched. Space/dash-formatted card numbers not detected. Returns findings + sanitized text. hasPII:false ≠ PII-free.
Prices sourced from x402.fairseal.io/.well-known/x402 · currency: USDC · network: Base mainnet