Cookie.fun 如何結合鏈上與鏈下數據排名 AI Agent
How an AI-Agent Analytics Layer Works: Notes on Cookie.fun's Data Approach" published
Cookie DAO 的 Cookie.fun 將鏈下社交討論與鏈上行為結合，為 1,400+ 個 AI Agent 建立排名。其第三階段推出 API，讓第三方 AI 工具使用數據，並以 Snaps 和 InfoFi 作為相關功能。文章指出，排名應結合活動與關注度訊號，並獎勵經核實的行動，而非單純曝光。
If you wanted to rank AI agents in crypto, where would the data come from, and how would you stop people gaming it? Cookie DAO's Cookie.fun is a useful real-world example: it indexes 1,400+ agents by merging off-chain chatter with on-chain behavior.
Here is the architecture at a conceptual level, based on SoluLab's published case study.
The key design choice is bridging off-chain sentiment with on-chain behavior in one signal dashboard. Social data alone is easy to inflate. On-chain data alone misses attention.
*Mindshare as a ranking primitive
*
Mindshare ranks agents by attention-weighted activity. The weighting details are not public in the case study, so treat this as a concept rather than a formula. For builders, the useful lesson is that a ranking needs both an activity signal and an attention signal.
*Where developers plug in
*
In the third rollout phase, Cookie released APIs so third-party AI tools could consume its data. That turned a dashboard into infrastructure, alongside two primitives:
- Snaps: tokenized actions that campaigns can reward
- InfoFi: attention and signal intelligence
*Practical lessons
*
- Combine at least two independent signal types before ranking anything.
- Expose an API early. Distribution through other tools beat a standalone UI.
- Reward verified actions, not impressions.
Full case study with results and the four-phase rollout: SoluLab: Cookie DAO and MarketingFi
來源：DEV Community · dev.to