Skip to content

ALTERNATIVES.md

Latest commit

 

History

History
248 lines (161 loc) · 11 KB

ALTERNATIVES.md

File metadata and controls

248 lines (161 loc) · 11 KB

- https://github.com//mikhailtal-manifold-agent Python agent trading exclusively on MikhailTal markets using calibrated probabilities and Kelly sizing.

Alternatives / Derivative Manifold Bots

This repository (microprediction/manifoldbot) is a general-purpose Python package for building Manifold trading bots, with examples such as an LLM trading bot that compares GPT-generated probabilities to market prices and bets using fractional Kelly sizing.

Below are community-built bots that take this pattern in different directions. For each one, the focus is on what’s novel relative to manifoldbot rather than just re-implementing the same thing.

manifold_mikhail_bot

Repo: https://github.com/mostafazhrn/manifold_mikhail_bot

What it is:
A Python-based, GUI-enabled Manifold trading bot that targets markets created by MikhailTal,
combining classical machine-learning models with LLM-based reasoning (local models or OpenAI API),
optionally augmented with web-search analysis, to produce probability estimates and automated trades.

Novelty vs manifoldbot:

  • Explicit ML training pipeline on resolved Manifold markets with per-option probability outputs
  • Probability calibration that respects and incorporates live market probabilities rather than ignoring them
  • Risk-aware final decision layer that blends ML estimates and LLM reasoning to gate and size trades conservatively
  • Optional reasoning layer using either local LLMs (via Ollama) or hosted APIs (e.g. OpenAI)
  • Web-search–augmented context gathering for markets with external information
  • GUI-first design for strategy configuration, live monitoring, and experimentation
  • Designed for extensibility beyond a single creator (default: MikhailTal)

Best for:
Users who want a flexible, GUI-driven Manifold bot that blends traditional machine learning
with optional local or API-based LLM reasoning.

eliteBot (jinanmh123)

Repo: https://github.com/jinanmh123/eliteBot/

What it is: A fully agentic hybrid Manifold trading bot, cleanly separating LLM-based semantic reasoning from quantitative decision-making, with regime-aware execution, volatility-adjusted Kelly sizing, realistic backtesting. Addressing key failure modes: (1) overconfident LLM probabilities, (2) purely technical strategies that miss semantic signals, (3) naive fractional Kelly overbetting in thin or unstable markets. Trading MikhailTal markets.

Novelty vs manifoldbot:

  • Hybrid, but with hard safety boundaries: LLMs handle semantics only (language, evidence, uncertainty, blind spots), Technical models handle behavior only (price, volume, volatility)
  • LLMs are forbidden from outputting probabilities, prices, or bet sizes. Agentic semantic reasoning without numeric authority Six-stage LLM pipeline (normalization, classification, base-rate context, evidence decomposition, uncertainty, constraints) Outputs are ordinal, schema-locked, cached, and deterministically mapped downstream, preventing hallucinated confidence and feedback loops
  • Regime-aware signal fusion: Markets are classified as information-driven, noise-driven, or mixed, and this regime dynamically bounds how much semantic (LLM) versus technical evidence is trusted, preventing runaway conviction in ambiguous or noisy markets.
  • Volatility- & uncertainty-aware Kelly sizing: explicitly scales bet size with Realized volatility, Liquidity, Aggregate model uncertainty and Market regime directly targeting the overbetting failure mode common in thin or ambiguous markets.
  • Execution and backtesting realism: Regime-dependent execution; Time-ordered replay backtester with slippage and delay modeling; Reports ROI, drawdown, Sharpe, and Brier calibration (no hindsight)

Best for: Users who want a robust hybrid system that captures market semantics without trusting LLMs with money, and sizes risk using volatility and uncertainty rather than edge alone.

better_manifold_bot (sachin-detrax)

Repo: https://github.com/sachin-detrax/better_manifold_bot

What it is: A “souped-up” Manifold bot using a multi-signal ensemble (historical performance, market microstructure, and an OpenAI LLM) to produce a single “true probability,” plus serious tooling for performance analysis.

Novelty vs manifoldbot:

  • Multi-signal ensemble combining historical bias, order-book microstructure, and structured LLM forecasts
  • Variance-aware LLM usage (multiple runs + disagreement penalty)
  • Dedicated ensemble decision layer with edge calculation
  • Advanced fractional Kelly with bankroll caps
  • Built-in performance graphs and detailed rationale logging
  • Configurable creator targeting (default: MikhailTal)

Best for: Users who want a research-grade ensemble bot with excellent observability.

ppbot-ai (prathameshpatrawale)

Repo: https://github.com/prathameshpatrawale/ppbot-ai

What it is: Contest-oriented bot built for the Manifold Featured Challenge, trading only MikhailTal markets.

Novelty vs manifoldbot:

  • Pure technical strategy (no LLM): mean-reversion + momentum + liquidity filters
  • Automatic simulation/real mode switching based on API key presence
  • Simple PnL logging and one-click cumulative profit plotting
  • Very clean, minimal modular structure

Best for: Lightweight non-LLM bots with quick setup and basic analytics.

Manifoldbot-Ultra (Sbha8282)

Repo: https://github.com/Sbha8282/Manifoldbot-Ultra

What it is: Packaged, modular bot designed as a proper Python package with dry-run by default.

Novelty vs manifoldbot:

  • Modern src/ layout with tests (Python 3.10+)
  • Session-cookie betting support (MANIFOLD_SESSION_COOKIE)
  • Safe-by-default (DRY_RUN=true)
  • Optional LLM layer (falls back to heuristics)
  • Strict creator targeting via env var

Best for: Anyone wanting a clean, production-style package skeleton.

TalOS-Manifold-Bot (rodriguezramirezederdominic-web)

Repo: https://github.com/rodriguezramirezederdominic-web/TalOS-Manifold-Bot

What it is: Agentic bot using GPT-4 for reference-class reasoning and fractional Kelly sizing.

Novelty vs manifoldbot:

  • Fully agentic LLM workflow (not just prompt → probability)
  • Explicit fractional Kelly module for risk-of-ruin control
  • Clear separation: trading loop, brain, and money-management files
  • Hard-coded focus on a single creator

Best for: Studying readable “LLM-as-brain + proper bankroll management” designs.

Best-Open-Source-Judgmental-Prediction-Python-Repository (“Afaq”)

Repo: https://github.com/H-tech-AFAQ-CEO/Best-Open-Source-Judgmental-Prediction-Python-Repository

What it is: Polished modular bot with both simple and LLM strategies.

Novelty vs manifoldbot:

  • Full type hints, strong error handling, and logging
  • Strategy base class with SimpleStrategy and LLMStrategy
  • Integrated backtesting script
  • Risk controls and cooldown logic
  • Designed to be PR-ready for upstream manifoldbot

Best for: Clean, typed codebases suitable for contribution or extension.

bot- (“Joshi-Bot”, 101jayjoshi-sudo)

Repo: https://github.com/101jayjoshi-sudo/bot-

What it is: Minimal scaffold focused on MikhailTal markets.

Novelty vs manifoldbot:

  • Extremely lightweight — meant as a clean starting template
  • Heuristics-first with hooks for local models (not tied to OpenAI)
  • Real bets disabled by default (PLACE_BETS=true required)
  • Explicit username → user-ID resolution step

Best for: Quick hacking and plugging in custom/local models.

Manifold-Markets-Trading-Bot (blackXmask)

Repo: https://github.com/blackXmask/Manifold-Markets-Trading-Bot

What it is: Full-featured trading dashboard with Streamlit GUI.

Novelty vs manifoldbot:

  • Rich Streamlit app with live monitoring, portfolio analytics, correlation heatmaps
  • Portfolio-level optimization and diversification logic
  • Built-in arbitrage detection
  • Advanced AI ensemble strategies
  • Comprehensive backtesting suite

Best for: Users who prefer a visual trader’s dashboard over CLI scripts.

mikhailtal-s-market-master (Djmon007)

Repo: https://github.com/Djmon007/mikhailtal-s-market-master

What it is (currently): A modern Vite + React + Tailwind front-end template (Lovable.dev scaffold). No bot logic yet.

Novelty vs manifoldbot: Front-end-first approach instead of Python library.

Best for: Starting point for a custom web UI or dashboard around a Manifold bot.

ContestMikhailBot (tanyat29)

Repo: https://github.com/barbiet503-bot/Strategy_contest.git

What it is:
A contest-focused Python trading bot built for the Manifold Featured Challenge, trading exclusively in markets created by MikhailTal.

Novelty vs manifoldbot:

  • Strict creator-only market filtering (contest rules enforced)
  • Edge + soft momentum confirmation before trading
  • Risk-aware, capped position sizing
  • Per-market cooldowns and duplicate-trade protection
  • Clean, contest-grade logging and CSV outputs

Best for:
Contest participants seeking a disciplined, explainable, and stable Manifold trading bot rather than aggressive or overfitted strategies.

manifoldbot (faizalmy)

Repo: https://github.com/faizalmy/manifoldbot

What it is: Multi-agent trading bot built on Google ADK (Agent Development Kit) with specialized agents for market analysis, risk assessment, and decision-making using consensus mechanisms.

Novelty vs manifoldbot:

  • Multi-agent architecture with specialized agents (MarketAnalysisAgent, RiskAssessmentAgent, CoordinatorAgent) using Google ADK
  • Chain-of-thought reasoning with transparent, auditable decision-making
  • External data source integration (Exa, Perplexity, Tavily) for real-time information access
  • Consensus-based decision-making through weighted voting across agents
  • Production-ready features: continuous operation mode, graceful shutdown, configuration validation
  • Comprehensive performance tracking (realized/unrealized P&L, win rate, risk-adjusted metrics)
  • Model-agnostic LLM access via LiteLLM (OpenAI, Ollama) through Google ADK

Best for: Users wanting a production-grade multi-agent system with external data integration and transparent reasoning.

Full list of alternatives (feel free to add more)