Analysis updated 2026-05-18
Run a serverless friend group prediction pool where no one holds everyone else's money.
Let an on-device AI stake its own cryptocurrency and compete against human predictions.
Study a working example of tamper-evident, peer to peer consensus using Autobase and Hypercore.
| edycutjong/treble | 0xmukesh/docusaurus-tutorial | 1tsmejp/palworld-docker-wine | |
|---|---|---|---|
| Stars | 1 | 1 | 1 |
| Language | JavaScript | JavaScript | JavaScript |
| Last pushed | — | 2021-12-27 | — |
| Maintenance | — | Dormant | — |
| Setup difficulty | moderate | easy | moderate |
| Complexity | 5/5 | 2/5 | 4/5 |
| Audience | developer | developer | ops devops |
Figures from each repo's GitHub metadata at analysis time.
Running the real AI brain downloads a roughly 1GB Qwen3 model via @qvac/sdk on first use, a heuristic fallback runs offline instantly.
The Treble is a hackathon project that lets a small group of friends run a football prediction betting pool without a server, a treasurer, or any central authority holding the money, and it lets an AI join the pool as a real participant with its own money and its own set of keys. The project targets a common problem with informal prediction pools: someone has to hold everyone's cash, and everyone has to trust that whoever wrote down each person's pick did not later change it. Instead, every participant, human or AI, holds their own cryptocurrency wallet and stakes their own funds directly. Picks are written to a shared, tamper evident, append only log using peer to peer technology, so once the match starts, no one, including the AI, can edit their prediction. If someone tries, every other peer in the pool rejects the change. The AI participant, nicknamed the Gaffer, reasons about the match using a small AI model that runs directly on the device rather than in the cloud, produces a short written rationale for its prediction, and then stakes its own money using a wallet it controls, all governed by a spending policy that caps how much it is allowed to risk. Importantly, the AI has deliberately less power than the humans: it cannot vote on the final score or lock in the result, only humans can do that, so it cannot manipulate the outcome in its own favor. Once the match result is confirmed by a majority of the human participants, payouts are calculated automatically so that the total amount paid out always equals the total amount staked, with no cut taken by anyone. Every peer involved ends up with an identical copy of the final state, which is how the system stays trustworthy without a central server checking anyone's work. The project is built with JavaScript and Node.js version 20 or later, using peer to peer networking tools alongside a cryptocurrency wallet toolkit and an on device AI toolkit, includes nearly two hundred automated tests, and is released under the Apache 2.0 license. It was built for the Tether Developers Cup hackathon.
A serverless football prediction pool where an on-device AI stakes its own cryptocurrency against human players, with no treasurer or cheating possible.
Mainly JavaScript. The stack also includes JavaScript, Node.js, Pear.
Use freely for any purpose, including commercial use, as long as you keep the copyright notice and any stated changes.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Verify against the repo before relying on details.