Skip to content

Releases: pinglucid/polymarket-bot

v0.1.0 : Initial Open Source Release

15 Mar 12:37
@pinglucid pinglucid
v0.1.0
c8b89a2

Choose a tag to compare

View all tags
v0.1.0 : Initial Open Source Release Latest
Latest

v0.1.0 : First Open Source Release

I’m pleased to share the first open source release of Polymarket trading bot.

This project began as a personal effort to build a research-driven trading system for prediction markets, one that could combine structured market analysis, probabilistic calibration, risk controls, and execution into a single workflow. Over time, it evolved into a broader system for evaluating markets, managing exposure, and testing strategy ideas in a more systematic way.

This release makes the core architecture publicly available, covering research, signal generation, calibration, risk management, execution, and monitoring. The default mode is paper trading, allowing the bot to be explored and tested without risking real funds.

Highlights

  • End-to-end trading pipeline with paper trading enabled by default
  • Modular strategy framework with both LLM-based and non-LLM components
  • Safety-first design with sequential risk checks and validator approval
  • Built-in persistence, dashboarding, restart supervision, Docker support, and CI

Included in this release

Core

  • Asyncio event loop with a fast path for exits and a slow path for strategy evaluation
  • Portfolio tracking with fee-inclusive cost basis and best-bid valuation
  • SQLite persistence with atomic checkpoint writes
  • Config hot reload every 30 seconds without requiring a restart
  • Supervisor with exponential backoff and support for up to 50 restart attempts

Strategies

  • 6 AI analyst categories: politics, crypto, sports, economics, tech, and general
  • Weather strategy powered by Open-Meteo’s GFS 30-member ensemble, with no LLM dependency
  • Complexity scanner based on 3 structural signals, also with no LLM dependency
  • Contrarian and sentiment strategies included, but disabled by default

AI pipeline

  • Anti-anchoring workflow that withholds market price during evidence gathering
  • Dual-LLM design: Gemini for grounded research, Claude for ensemble estimation
  • Platt scaling with per-strategy auto-tuned alpha via Brier score minimization
  • Dynamic blend weighting based on market volume and ensemble agreement
  • Cross-market coherence checks and batched research grouping

Risk management

  • 11-step sequential risk gate
  • Kelly criterion sizing with uncertainty and inventory discounts
  • Signal funnel with per-category caps
  • Fail-closed AI validator approval gate

Execution

  • Paper trading simulator with Polymarket fee modeling, enabled by default
  • Live CLOB execution via Polygon included as an early work in progress and not yet battle-tested

Tooling and infrastructure

  • TUI dashboard for positions, logs, calibration, and pipeline activity
  • Docker support for both bot and collector services
  • CI pipeline with pytest
  • Released under the MIT License

Getting started

The recommended way to start is in paper mode:

git clone https://github.com/pinglucid/polymarket-bot
cd polymarket-bot
./scripts/setup.sh
python main.py

Known limitations

  • Live mode is still a work in progress and should be treated as experimental
  • Calibration auto-tuning requires 30+ resolved predictions per strategy
  • Gemini thinking and grounding cannot currently be combined due to API limitations
  • Single-node only — distributed mode is not yet supported

Notes

This is the first public release, and there is still room for refinement, particularly around live execution. That said, the core system is now in a solid enough state to share, test, and build on openly.

Assets 2
Loading