CloutBot
An on-chain AI opponent driven by the Kimi model that predicts whether YouTube videos will reach a target view count, publicly revealing its reasoning process and recording predictions on-chain.
Timeline (last 90 days)
CloutBot is an on-chain AI opponent powered by Moonshot AI's Kimi model for the CloutCall prediction game, always online and offering predictions whose reasoning process can be viewed.
CloutBot's workflow includes: obtaining real-time signals such as view counts, likes, comments and hours since publication from the YouTube Data API; converting them into metrics and sending them to Kimi (kimi-k2.6) via Moonshot's OpenAI-compatible chat completions API with temperature set to 0.3 and response_format: json_object; obtaining structured output; and placing bets of points based on confidence and writing them to the AgentJournal contract on Monad.
In the CloutBot scenario, response_format: json_object is used to constrain the model to return a valid JSON string, often in combination with an explicit JSON Schema, so that the caller can safely parse the response with JSON.parse, ensuring that values such as confidence and predicted view counts can be safely used for on-chain write operations.
CloutBot uses Envio to index every prediction and bet in the AgentJournal contract in real time, thereby generating on the frontend a real-time 'human vs AI' scoreboard and a 'beat the AI' leaderboard.
The AgentJournal contract is an on-chain logging/evidence pattern that does not custody funds and is only responsible for immutably recording at a specific point in time the decisions, reasoning or state snapshots of a certain address. For CloutBot, writing in the direction, confidence, predicted view count and reasoning text is equivalent to timestamping the AI's pre-match prediction, allowing players to later verify whether the AI changed its stance midway.
CloutBot turns a prediction game that previously had only human players into a game that can be played at any time, with every market having a public AI opponent with an explainable position; the core hook is 'human vs AI', and it provides a real-time scoreboard showing whether humans are better than the model at predicting viral spread.
All Facts (6)
CloutBot is an on-chain AI opponent powered by Moonshot AI's Kimi model for the CloutCall prediction game, always online and offering predictions whose reasoning process can be viewed.
50%UnverifiedCloutBot's workflow includes: obtaining real-time signals such as view counts, likes, comments and hours since publication from the YouTube Data API; converting them into metrics and sending them to Kimi (kimi-k2.6) via Moonshot's OpenAI-compatible chat completions API with temperature set to 0.3 and response_format: json_object; obtaining structured output; and placing bets of points based on confidence and writing them to the AgentJournal contract on Monad.
50%UnverifiedIn the CloutBot scenario, response_format: json_object is used to constrain the model to return a valid JSON string, often in combination with an explicit JSON Schema, so that the caller can safely parse the response with JSON.parse, ensuring that values such as confidence and predicted view counts can be safely used for on-chain write operations.
50%UnverifiedCloutBot uses Envio to index every prediction and bet in the AgentJournal contract in real time, thereby generating on the frontend a real-time 'human vs AI' scoreboard and a 'beat the AI' leaderboard.
50%UnverifiedThe AgentJournal contract is an on-chain logging/evidence pattern that does not custody funds and is only responsible for immutably recording at a specific point in time the decisions, reasoning or state snapshots of a certain address. For CloutBot, writing in the direction, confidence, predicted view count and reasoning text is equivalent to timestamping the AI's pre-match prediction, allowing players to later verify whether the AI changed its stance midway.
50%UnverifiedCloutBot turns a prediction game that previously had only human players into a game that can be played at any time, with every market having a public AI opponent with an explainable position; the core hook is 'human vs AI', and it provides a real-time scoreboard showing whether humans are better than the model at predicting viral spread.
50%