The problem 0xBet solves

0xBet

Autonomous intelligent betting, at your fingertips.

---

The Problem

Sports betting rewards research. Almost no one has the time, tools, or discipline to do it properly.

The edge comes from understanding team form, injury reports, historical head-to-heads, line movement, and market mispricing. Professional bettors ("sharps") spend hours every day on this before placing a single bet.

For everyone else, three things get in the way:

• Time — thorough research takes hours per fixture, most people don't have it
• Consistency — humans cut corners, miss signals, and make assumptions
• Emotion — humans tilt, chase losses, and bet with their heart not their head

The result: retail bettors lose. Not because the edge doesn't exist — it does — but because they lack the infrastructure to find and act on it.

---

What 0xBet Does

0xBet is an autonomous AI betting agent that operates as its own on-chain business.

You fund it once. It does everything else.

The agent continuously researches upcoming sports fixtures, identifies mispriced markets on Polymarket, places bets with mathematically sound stake sizing, collects winnings, and reinvests profits into more research — compounding its own intelligence over time.

It runs 24/7, without emotion, without fatigue, and without taking a rake.

---

How It Works

User deposits USDC → 0xBet's icpay merchant account
        ↓
Agent allocates budget: bets / compute / reserve
        ↓
Agent purchases research compute via ClawRouter - pay-per-request USDC, no API keys
        ↓
Agent uses Bird (OpenClaw skill) to pull recent X activity around upcoming fixtures
        ↓
AI analyses sentiment, narrative, and social signals to identify which result to bet on
        ↓
Agent sizes stakes using Kelly Criterion based on confidence and bankroll
        ↓
Agent deposits USDC to Polymarket and places bets via CLOB API
        ↓
Winnings swept back to treasury → reinvested automatically

---

What People Use It For

🧠 Replacing manual research

Bettors who currently spend hours reading match previews, checking injury lists, and scanning X for breaking news can offload all of that to 0xBet. The agent uses Bird to pull real-time X activity — journalist reports, insider leaks, fan sentiment, team announcements — and synthesises it into a betting decision faster and more consistently than any individual could.

📈 Compounding a betting bankroll

Rather than letting winnings sit in a withdrawal queue, 0xBet automatically reinvests profits into the next research cycle and the next set of bets. The bankroll compounds. The more it wins, the more compute it buys, the better its research gets.

🔒 Removing emotional bias

The biggest edge killer in sports betting is the human mind — chasing losses, overweighting recent form, betting on your favourite team. 0xBet has no favourite team. It reads the signals, sizes the bet, and executes.

💸 Accessing decentralised betting markets

Polymarket offers peer-to-peer sports markets with tighter margins than traditional books. 0xBet plugs directly into Polymarket's CLOB API, giving users access to better odds without the friction of navigating the interface themselves.

👁️ Full transparency, on-chain

Every deposit, every research spend, every bet placed, and every settlement is recorded on-chain and visible in the 0xBet dashboard. No black box. Users can see exactly what the agent is doing with their funds in real time.

🤖 The future of agentic commerce

0xBet is the first example of a fully autonomous business built on icpay — accepting payments, purchasing its own compute, and generating returns, all without human intervention. It's not just a betting tool. It's a proof of concept for what AI agents can do when given economic autonomy.

---

The Stack

---

Built at ETH Denver 2026.

Challenges I ran into

Challenges We Ran Into

---

1. Agent Hallucinations on Trade Confirmation

One of the most frustrating issues we encountered was the AI agent confidently reporting that it had placed a bet on Polymarket — when it hadn't.

The agent would complete its research cycle, identify a pick, and then state that the trade had been executed. Checking our database told a different story: no trade record existed. The agent had hallucinated the confirmation.

How we fixed it: We implemented a verification step where the agent is explicitly told to check the database for confirmation of the trade before reporting success. If no record exists, it's informed directly that the trade was not placed and must retry. Grounding the agent in on-chain and database state — rather than trusting its own output — was the key shift.

---

2. Geo-Block on Polymarket Access

When we deployed our agent to an AWS-hosted server, we hit an immediate wall: Polymarket is geo-blocked in certain regions, and our server instance fell into one of them. The agent couldn't reach the Polymarket CLOB API at all.

How we fixed it: We routed the server traffic through a VPN to a permitted region. Not the most elegant solution, but it unblocked us quickly during the hackathon. A production version of this would use a server hosted in a non-restricted jurisdiction from the start.

---

3. Lack of Visibility Into Agent Activity

Early on, we had no reliable way to tell whether the agent was actively working or had silently stalled. There were no live indicators — we were essentially flying blind, relying on manually prompting the agent to check in on its status.

How we fixed it: We built the live activity feed in the 0xBet dashboard. Every action the agent takes — research calls, compute purchases, bets placed, settlements received — is logged in real time and surfaced in the UI. The agent status indicator (Researching / Placing Bets / Settling / Idle) gives an at-a-glance view of what it's doing at any moment, removing the dependency on manual check-ins entirely.

Use of AI tools and agents

Use of AI Tools and Agents

0xBet is not a tool that assists a human bettor. It is the bettor — an autonomous AI agent that researches, decides, and executes entirely on its own.

---

The Agent

At the core of 0xBet is an AI research agent powered by a large language model. The agent accesses two distinct tools to do its work:

• ClawRouter — a pay-per-request compute layer that allows the agent to purchase LLM intelligence on demand, in USDC on Base, with no pre-funded API keys
• OpenClaw — used to access real-time X (Twitter) sentiment and activity, giving the agent a live pulse on public opinion, injury news, team morale, and breaking developments around upcoming fixtures

This is the critical design decision: the agent earns money, then spends that money on its own thinking. Compute is not a fixed cost — it's a variable investment that scales with the agent's bankroll and performance.

---

What the Agent Does

Each research cycle, the agent:

1. Fetches upcoming fixtures across football, basketball, tennis, and other supported sports via live data APIs
2. Pulls structured data — team form, head-to-head records, injury reports, recent results, and current Polymarket odds
3. Reads X sentiment via OpenClaw — scans real-time posts, fan sentiment, journalist reporting, and insider activity around each fixture to surface signals the structured data might miss
4. Reasons over the combined data — the LLM synthesises structured stats and live social sentiment to identify where the market may be mispriced, producing a structured output: pick, confidence score, and expected value estimate
5. Sizes stakes using Kelly Criterion — mathematically optimal bet sizing based on perceived edge and current bankroll
6. Places bets autonomously via the Polymarket CLOB API
7. Settles and reinvests — winnings are swept back to treasury and the cycle repeats

---

How AI and Onchain Work Together

The AI layer and the onchain layer are tightly coupled — neither works without the other:

The agent uses its icpay balance to decide how much compute to buy each cycle. If the bankroll grows, it buys more research calls. If it shrinks, it conserves. The AI's budget is dynamically managed onchain.

---

Why This Matters

Most AI agents are given a fixed compute budget by a human operator. 0xBet's compute budget is determined by its own performance. A winning agent gets smarter. A losing agent cuts back. And by combining structured sports data with live X sentiment via OpenClaw, the agent has access to the same signals a professional sharp would — updated in real time, processed at machine speed.

---

AI is not a feature of 0xBet. It is the operator.