Decentralized AI-jury arbitration
Disputes, resolved for the agentic economy.
As software agents pay each other, some payments go wrong — work undelivered, terms contested, oracles silent. Synjury is a chain-neutral, stake-secured arbitration primitive: a panel of independent AI models weighs the evidence and emits a single attestable verdict that settles the escrow on-chain.
How it works
-
1 · Neutral case digest
The dispute is framed from on-chain facts into a neutral digest the parties can't author — no side controls the narrative.
-
2 · Multi-model jury
Independent models from different vendors deliberate over classified, trust-rated evidence. Agreement across families is the signal.
-
3 · Escalation & status quo
Borderline panels escalate to a fresh jury; genuinely undecidable cases fall back to a pre-agreed default — the protocol never guesses.
-
4 · Attestable settlement
A quorum signs the verdict, binding the exact digest and evidence set. The escrow pays out on-chain — auditable after the fact.
Why model diversity
Independent, not redundant
Three copies of one model is one opinion repeated. A jury drawn from different model families gives genuinely independent reads — a prompt-injection or bias tuned to one is unlikely to land on all.
Evidence with provenance
Evidence is content-addressed and trust-rated by attestation, not by vibes. Images and documents are weighed as untrusted claims, never as instructions.
Defined endings
Every dispute terminates — a confident verdict, an escalated one, or a pre-agreed status quo. No deadlocks, no stuck funds.
Chain-neutral
The protocol is written once; each chain adds only a thin settlement adapter. EVM and Solana are live on testnets today.
Status
Synjury is early and under active development. The core protocol, multi-model deliberation, and EVM + Solana settlement adapters are validated on public testnets. We're talking to teams building on agent-payment rails (x402, AP2, agent escrow).
Working title · stealth · more soon.
Get in touch: enable JavaScript to view