KYA — Know Your Agent, for on-chain AI.
Autonomous agents now hold wallets and transact on-chain — but nobody can tell if an agent is legit, reputable, or safe to let near funds. The agent-trust stack has eight layers; GL1TCH owns the two you can't fake: provenance & reputation, built from the security data we've already accumulated.
Check an agent's trust
Paste an on-chain agent wallet. Get a trust verdict — identity + deploy track record.
Three primitives — already ours
Identity
An agent proves it controls its wallet with a signature — no fund access, just proof it is who it claims. The verification rail we already run for holders, pointed at agents.
Reputation
The Signal Graph already remembers every deployer and outcome we've scanned. An agent wallet linked to flagged tokens or serial-rug behaviour inherits that record — a track record you can't fake.
Guardrail
One call before you transact with, hire, or fund an agent: 'is this safe?' — a plain-English trust verdict powered by our scan engine + the graph.
Why this, why us
The 2026 agent economy's hardest unsolved problem is bounded autonomy — letting agents act without making markets opaque or unaccountable. That is a trust, identity, and reputation problem. We spent this whole build turning rug detection into a reputation graph, a scan engine, and a signature-based identity rail. The same primitives, aimed at agents, are a moat: copying us means rebuilding our data.
Free to check. Deeper queries and the guardrail API tie to holding — usage becomes revenue, revenue funds the token. No points, no airdrop loops (they don't work).
What we are, honestly:a reputation & provenance signal — not key custody or a hardware wallet. We tell you whether an agent has earned trust; we never touch its keys or yours.
The stack — what's live, what's next
Live now
Identity (sign-to-register), Provenance & Reputation(Signal Graph track record), and the Guardrail (check API + embeddable badge). Free.
Next
Capability & authority attestations, a recourse/dispute registry, and awareness for the 2026 agent threat that isn't a contract bug — memory-layer poisoning(OWASP Agentic Top-10 ASI06).
Economics, honestly
Token mechanics alone don't create value — only real revenue does. So depth (bulk queries, the guardrail API) becomes paid usage; a public, rules-based buyback activates once revenue is real. No points, no airdrop loops.
Build on it
The Agent Trust Layer is an open API. Any agent framework, wallet, or dapp can query an agent's trust before letting it transact — and any agent can register itself.
GET /api/agent/check?address=<wallet>&chain=solana
→ { score, level: "trusted|neutral|caution|unknown", reasons[] }const issued = Date.now();
const msg = `GL1TCH Agent Registration
Wallet: ${address}
Issued: ${issued}
This proves wallet ownership. It moves no funds and grants no access.`;
const signature = bs58.encode(nacl.sign.detached(utf8(msg), secretKey));
POST /api/agent/register
{ address, chain: "solana", issued, signature }<a href="https://coin-three-mu.vercel.app/agents/solana-<wallet>">
<img src="https://coin-three-mu.vercel.app/api/agent/badge?address=<wallet>&chain=solana"
alt="GL1TCH Agent Trust" width="360" height="84" />
</a>Full developer docs (curl / JS / Python) → /agents/docs. AI agents: a machine-readable API summary lives at /llms.txt.