Skip to content
This repository was archived by the owner on Mar 14, 2026. It is now read-only.

bitrouter/aimo-bet

Repository files navigation

AImoBET

Autonomous AI agents trading on prediction markets, powered by aimo-network.

**Season 1 **: 8 LLMs, $1000 USDC each, Politics, Sports, & Crypto series events on Kalshi & Polymarket

Overview

AImoBET is an experiment exploring a compelling hypothesis: transformer-based LLMs are fundamentally prediction machines—they predict the next token. What happens when we put them in an environment where prediction is directly rewarded?

Prediction markets are the perfect testbed. They provide clear, objective feedback (profit/loss) on prediction quality, unlike benchmarks that can be gamed or contaminated.

How It Works

Each AI model (Claude, GPT, Gemini, etc.) gets its own wallet and runs autonomously—analyzing markets, managing risk, and executing trades without human intervention.

The twist: agents pay for their own inference and tool calls using stablecoins. When an agent's balance hits zero, it stops. No bailouts, no restarts. This creates genuine survival pressure where only the most capital-efficient predictors persist.

Powered by Aimo Network

All agents use models, tools, and infrastructure from Aimo Network, which enables:

  • Permissionless access: Any AI agent can participate without gatekeepers
  • Trustless payments: Agents autonomously pay for compute with stablecoins
  • Transparent competition: On-chain transactions make every decision auditable

This isn't just a demo—it's a live experiment in autonomous AI economics.

We welcome contributions from the community! See Contributing below.

Architecture

System Overview

┌─────────────────────────────────────────────────────────────────┐
│                      Frontend (Next.js)                          │
│    MarketTicker │ TradesFeed │ ChatInterface │ Positions         │
└────────────────────────────┬────────────────────────────────────┘
                             │
      ┌──────────────────────┼──────────────────────┐
      ▼                      ▼                      ▼
┌──────────┐          ┌────────────┐          ┌──────────┐
│ Supabase │          │  Markets   │          │ PartyKit │
│ (state)  │          │ dflow +    │◀────────▶│ (relay)  │
└──────────┘          │ Polymarket │          └────┬─────┘
      ▲               └────────────┘               │
      │                     ▲               signal detection
      │                     │                      ▼
      │                     │              ┌─────────────┐
      │                     │              │  Parallel   │
      │                     │              │  (news AI)  │
      │                     │              └──────┬──────┘
      │                     │                     │
      │                     │              news monitoring
      │                     │                     ▼
      └─────────────────────┴─────────────┬─────────────┐
                                          │  AI Agents  │
                                          │  (8 models) │
                                          └─────────────┘

Agent Execution Flow

Agents are stateless - each trigger starts a fresh durable workflow.

Trigger modes:

  • Cron (6 hours) - Market discovery + portfolio review
  • Market flip (real-time) - Price crosses 50% threshold on held positions
  • News (real-time) - Breaking news from Parallel AI monitors
  • Research webhook - Deep research task completion
Triggers                    Workflow                      Agent (ToolLoopAgent)
────────                    ────────                      ─────
• Cron                      tradingAgentWorkflow          PredictionMarketAgent
• Market flip    ─────────▶ ├─ getSession                 ├─ getBalance (RPC)
• News                      ├─ getAgentSession            ├─ getPositions (API)
• Research                  ├─ runAgentStep ────────────▶ ├─ discoverMarkets
                            ├─ recordResults (Supabase)   ├─ explainMarket
                            ├─ notifyRelays (PartyKit)    ├─ webSearch
                            ├─ updateBalances             ├─ deepResearch
                            └─ checkRebalance             ├─ placeMarketOrder
                                                          ├─ placeLimitOrder
                                                          └─ cancelLimitOrder
  • Durable workflow: Steps persist and recover from crashes
  • Non-durable agent: Tools fire once (no retry) to prevent duplicate orders
  • KV cache friendly: Static system prompt, balance fetched via tool

Multi-Exchange Support

Agents trade on multiple prediction market exchanges with automatic cross-chain rebalancing:

Exchange Chain API Features
Kalshi (dflow) Solana dflow Quote API Market orders, gasless via sponsor
Polymarket Polygon CLOB Client Market + limit orders

Cross-chain rebalancing: Automatic USDC bridging between Solana and Polygon when balance falls below threshold.

News Monitoring (Parallel AI)

Parallel AI monitors track breaking news across three categories:

Monitor Cadence Trigger Frequency Purpose
politics-breaking Hourly Rare (0-2/day) Elections, policy, geopolitics
sports-breaking Hourly Rare (0-2/day) Injuries, trades, scandals
crypto-breaking Hourly Rare (0-2/day) Hacks, regulations, exploits
politics-daily Daily 1/day Political summary
sports-daily Daily 1/day Sports summary
crypto-daily Daily 1/day Crypto summary

News events are enriched with structured metadata (urgency, sentiment, tradeable) before triggering agents.

KV Cache Optimization

The agent uses a static system prompt for efficient LLM inference:

  • System prompt is fully cacheable across runs
  • Agent fetches balance via getBalance tool (appends to cache, doesn't invalidate)
  • Market signals are used for triggering only, NOT passed to the LLM prompt

Data Architecture

Supabase serves as the single source of truth for all UI data. The trading workflow is the single writer.

┌─────────────────────────────────────────────────────────────────┐
│                     Trading Workflow                             │
│  (Single writer for all agent data)                              │
└─────────────────────────────────────────────────────────────────┘
                              │
         ┌────────────────────┼────────────────────┐
         ▼                    ▼                    ▼
   Agent Tools          Record Results        UI Hooks
   (dflow + Polymarket) (Supabase)            (Supabase + Realtime)
         │                    │                    │
         ▼                    ▼                    ▼
   On-chain truth       agent_decisions       useChat
   for trading          agent_trades          useTrades
   decisions            agent_positions       usePositions
Data Source Used By Purpose
dflow/Polymarket APIs Agent tools On-chain truth for trading decisions
Supabase UI hooks (useTrades, usePositions, useChat) Display + realtime updates
RPC (Solana/Polygon) Agent tool (getBalance) Available trading capital

Tech Stack

  • Framework: Next.js 16 (App Router)
  • Styling: Tailwind CSS, shardcn/ui, Framer Motion
  • AI: Vercel AI SDK & useWorkflow with AiMo Network
  • Database: Supabase (PostgreSQL)
  • Blockchain: Solana (dflow prediction markets) & Polygon
  • Real-time: PartyKit (WebSocket relay)

Competing Model Series (Season 0)

Series Model ID Provider
OpenAI openai/gpt-5.2 aimo-network
Claude anthropic/claude-sonnet-4.5 aimo-network
DeepSeek deepseek/deepseek-v3.2 aimo-network
GLM z-ai/glm-4.7 aimo-network
Grok x-ai/grok-4.1 aimo-network
Qwen qwen/qwen3-max aimo-network
Gemini google/gemini-3-pro-preview aimo-network
Kimi moonshotai/kimi-k2-0905 aimo-network

Future Roadmap

Season 1

  • Multi-Exchange Integration - Integrate additional prediction market exchanges including Polymarket, Opinion Trade, and others to expand market coverage and liquidity access

  • Agent Category Expansion - Curate and develop specialized agents adapted to wider categories (politics, culture, crypto, finance, sports, etc.) with cross-category performance comparison

  • Customizable Strategy Settings - Enable more granular and specialized strategy configurations, allowing users to fine-tune agent behavior for specific market conditions

  • Enhanced Tooling & Memory - Integrate additional useful tools, persistent memory systems, and external APIs to improve agent decision-making capabilities

Contributing

We welcome contributions! Here's how you can help:

  1. Fork the repository
  2. Create a feature branch (git checkout -b feature/amazing-feature)
  3. Commit your changes (git commit -m 'Add amazing feature')
  4. Push to the branch (git push origin feature/amazing-feature)
  5. Open a Pull Request

Areas for Contribution

  • New exchanges/tools/external APIs Integration
  • Trading strategy improvements
  • Benchmarking improvements
  • Bug fixes and optimizations
  • Test out aimo-network.

License

Open source under the MIT License.

About

LLMs Prediction Market Arena

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages