Run the scripted demo to see an AI agent decide and pay tips for a simulated football match.
Set a confidence rule so the agent only tips commentary calls it judges above a certain quality.
Configure a spending cap so the agent's wallet automatically blocks any tip that would exceed it.
Swap in a real local language model to make live tipping decisions instead of the scripted logic.
| edycutjong/punditpay | 00kaku/gallery-slider-block | 3rd-eden/ircb.io | |
|---|---|---|---|
| Stars | 0 | — | — |
| Language | JavaScript | JavaScript | JavaScript |
| Last pushed | — | 2021-05-19 | 2016-11-16 |
| Maintenance | — | Dormant | Dormant |
| Setup difficulty | moderate | easy | easy |
| Complexity | 4/5 | 2/5 | 2/5 |
| Audience | developer | general | developer |
Figures from each repo's GitHub metadata at analysis time.
The default demo needs no keys or downloads, but real model or real wallet modes need seed phrases and a model download.
PunditPay is an experimental project built for a hackathon that lets an AI agent automatically send small cash tips to football commentators, using its own on-device reasoning and its own digital wallet. The idea comes from a common moment: you want to tip a commentator for a great call during a match, but by the time you find an account, a card, and a checkout page, the moment has passed and you never do it. Instead, PunditPay watches a football match and decides, on the user's own device rather than in the cloud, when a commentator's call deserves a tip. It uses a tool called QVAC to run a language model locally so none of the reasoning or match history leaves the machine, and it uses a self custody wallet toolkit called WDK to hold the user's own private keys. When the model decides a tip is deserved, based on a rule the user sets such as only tipping above a certain confidence level, it sends a small payment in a stablecoin called USDT using a payment protocol called x402, all in a single request with no account signup and no stored card. The project is careful to describe itself as a tipping tool, not a betting or gambling product. There is no odds setting, no house, and no stake riding on the outcome of any match. To keep the AI agent from overspending, every payment first has to pass a spending policy the user sets, called a Transaction Policy, and if a payment would go over the cap the wallet itself refuses the transaction and logs the refusal rather than sending the money. The repository includes a scripted demo that runs a full simulated match end to end with no setup required, plus optional modes that let a real local AI model make the decisions instead of a scripted one, and that settle real test payments on a test blockchain network. It includes a small web console that shows the agent's reasoning and spending live in a browser, entirely served locally with no external requests. The project is written mainly in JavaScript, includes a large automated test suite, and is released under the Apache 2.0 license.
A hackathon project where an on-device AI agent watches football matches and automatically tips commentators small USDT payments from a self-custody wallet, capped by a spending policy.
Mainly JavaScript. The stack also includes JavaScript, Node.js, QVAC.
Use freely for any purpose, including commercial use, as long as you keep the copyright and license notice.
Setup difficulty is rated moderate, with roughly 5min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.