PurpleIris

PurpleIris (V1.0 Hermes) is a conversational validator-intelligence chatbot for Monad. Ask the chain anything about a validator in plain language, and every number comes back labeled by data class: Class 1 on-chain, Class 2 derived, Class 3 estimated, so you always know what you're looking at. For delegators: real yield simulations ("stake X, earn Y a year") built on a validator's actual on-chain history and commission; network context showing exactly where a validator sits against the network average in clean percentage-points; and a strictly non-advisory design that hands you the facts and lets you decide. For operators: on-chain execution-layer revenue and staking income ranked against the network, not just uptime; one-question node diagnostics (live status, health, streaks, anomalies); and client version and hash tracking so you know if your node is behind. The problem: staking on Monad means trusting someone with your MON, but validator data is scattered, hard to read, and rarely tells you what you'll actually earn or whether a figure is fact or estimate. PurpleIris puts a live, indexed data layer behind a chat box and labels every number by trust level. How it's built: a dedicated indexer ingests every epoch into a Postgres store; around 24 read tools across delegator and operator modes; reasoning on Claude Sonnet 4.6 with Claude Haiku 4.5 for fast paths and guardrails, and prompt caching for cost-efficient scaling. A live, hosted off-chain app. What I learned: data correctness against live mainnet is the hard part, reconciling baselines and labeling estimates honestly. Honesty in the numbers turned out to be the whole product. Live at purpleiris.live.