KOLlateral

Crypto influencers operate with almost no accountability. They post hundreds of "calls" a week, delete the ones that lose, and there is no shared record of whether following them ever made money. KOLlateral fixes both sides of that with one thing: a public, verifiable track record.

For a caller who is actually good, that record is an asset they own. Every explicit call they make in public becomes a structured signal (asset, direction, target, confidence) and gets priced against real DEX history, so their edge shows up as numbers they can point to: what following them returned versus just holding ETH. Losing calls are archived and flagged in red instead of disappearing. Each call is checked against the caller's own on-chain wallet, so "said accumulate, sold four hours later" is a contradiction with a transaction hash attached. And every score is produced by AI inference running inside a verifiable enclave, so the record holds up even against us.

For everyone else, that record is a filter. A leaderboard surfaces the callers who have genuinely earned trust rather than the loudest ones, and each influencer gets a dossier with the equity curve of every call they have made. Once you find someone real, you ride along: copy their calls, or fade the ones who keep getting it wrong, in one click, executed on-chain from a self-custody wallet.

The AI that reads and judges every call runs on 0G Compute, and that inference is the part we cannot fake. Each post goes to 0G's OpenAI-compatible router with verify_tee turned on and a private trust-mode header, so the model executes inside a TDX secure enclave and the router hands back an attestation that this exact inference ran on the stated model, untouched. That is the core claim of the product: a caller's verdict is genuinely the model's output, not a number we typed in, and no one, including us, could edit the reasoning between their tweet and the score. The same verifiable inference powers 0-yap mode, which distills a rambling post down to its bias, a one-line thesis, and the price levels.

The Graph is the live on-chain data all of that scoring reasons over. We query the Uniswap v2 and v3 subgraphs through the gateway (v3 first, v2 as a fallback) for two things the product cannot work without: pricing every call at its exact posted timestamp, so a claim turns into a real return, and pulling a caller's own swap history to run the said-versus-did check. The forensic layer takes each call the AI extracted and matches it against what that wallet actually did on-chain, so a contradiction is grounded in live Subgraph data rather than a static snapshot.

Execution is Uniswap. On Base mainnet we use the hosted Trading API. Base Sepolia is where it got hacky: the Trading API does not index that chain, so we located the deployed WETH/USDC v3 pools on-chain and call SwapRouter02 directly, quoting in USDC and then decoding the ERC-20 Transfer out of the swap receipt so the portfolio records the real fill instead of a nominal number. Wallets and signing are Privy: each user gets an embedded self-custody wallet and delegates a session signer once, after which every Follow or Fade executes server-side with no popup.