The Open-Source Kalshi Bot Ecosystem: 30+ Projects Worth Studying in 2026

The Kalshi bot ecosystem has grown from a handful of GitHub projects in 2023 to a sprawling landscape of dozens of open-source bots, frameworks, and tools by 2026. We catalogued them all — every public bot we could find, organized by strategy type, with honest notes on what each does well and what it doesn't.

This is a working catalog, not a competitive hit piece. If you're a developer building your own bot, this is your reading list. If you're a trader trying to understand what's out there, this is your map. We build a competing product (Bot for Kalshi), and we'll be upfront about that — but the open-source ecosystem is genuinely interesting, and many of these projects have taught us something worth knowing.

Want the raw data? Every bot in this post is available as JSON or CSV. Free to use; attribution appreciated. Each bot also has its own page — click any bot name to dive deeper.

How We Catalogued This

We searched GitHub for every public Kalshi bot, framework, and trading tool. Inclusion criteria: actually trades or analyzes Kalshi markets (not just data exploration), public source code, last meaningful commit within the past 18 months. We grouped by primary strategy approach. For each project, we link to the repo, note the author's GitHub handle (so you can find more of their work), and call out what's distinctive.

Stars are listed where we could verify them at the time of writing. They correlate with attention, not necessarily quality — some of the most interesting bots have single-digit star counts because their authors haven't promoted them. We've tried to surface what's notable about each, not just popular.

If we missed your bot, or got something wrong, tell us. We update this catalog quarterly.

AI & LLM-Driven Bots

The newest category. These bots use large language models to process unstructured information — news articles, research reports, social media — and turn it into trading signals. They tend to be experimental and resource-hungry but capable of strategies that pure quantitative approaches can't easily express.

OctagonAI / kalshi-deep-trading-bot ★206

Repo: github.com/OctagonAI/kalshi-deep-trading-bot

The most-starred AI-driven Kalshi bot, period. Performs deep fundamental research per market via the Octagon Research API, generates independent probability estimates, computes edge against the live order book, sizes via half-Kelly through a five-gate risk engine (Kelly + Liquidity + Correlation + Concentration + Drawdown). Supports JSON output for agent orchestration.

What we learned from it: the five-gate risk engine is a model for how serious risk management looks in this space. It's not just position sizing — it's a stack of independent constraints that all have to pass before a trade fires. Most bots that blow up have failed exactly one of those checks.

ryanfrigo / kalshi-ai-trading-bot ★GitHub

Repo: github.com/ryanfrigo/kalshi-ai-trading-bot

Grok-4 integration with multi-agent decision-making and portfolio optimization. Uses 0.25 Kelly sizing with a category scorer that hard-blocks markets scoring below 30 out of 100.

What we learned from it: the author's honest disclosure is unusual in this space — "AI ensemble can be 80% confident and still be wrong; no edge on economic releases." That kind of upfront refusal to overclaim is rare and worth modeling.

BEXAI / KalshiBot ★GitHub

Repo: github.com/BEXAI/KalshiBot

Dual-loop architecture: Google TimesFM 2.5 for fast quantitative tick predictions, Gemma 4 31B for slow multi-agent LangGraph debate (Bull, Bear, Volatility Forecaster, and Risk Management personas). Built without an SDK using manual RSA-PSS payload cryptography. React/Vite frontend exposes AI inference thresholds and live P&L.

What we learned from it: the dual-loop pattern (fast model + slow reasoning) is the most sophisticated architecture we've seen in open-source prediction-market bots. Trade only when both loops agree is a strong default.

ajwann / kalshi-genai-trading-bot ★GitHub

Repo: github.com/ajwann/kalshi-genai-trading-bot

"Vibe-coded" generative AI bot — system-prompts an LLM (Grok) into the role of a professional prediction-market trader. More experimental than productionized, but a useful reference for prompt-engineering approaches to trading.

alanshen421 / KalshiWeatherBot ★GitHub

A weather-focused bot that uses Claude Opus to size and approve trades. Quarter-Kelly sizing capped by both maximum contracts and maximum dollar risk per trade. The combination of LLM-as-judge with hard sizing limits is a sensible safety pattern.

Quantitative & Market Making

The professional side of the ecosystem. These bots use textbook quantitative finance techniques — many adapted from equity and futures markets — and apply them to Kalshi's binary contract structure.

rodlaf / KalshiMarketMaker ★195

Repo: github.com/rodlaf/KalshiMarketMaker

The most-cited reference for "real" market-making math on Kalshi. Implements multiple MM strategies in parallel, including an Avellaneda-Stoikov model. Dynamic market selection scores by volume and spread; a real-time terminal dashboard (curses) visualizes inventory. Strict deselect-cleanup invariant: stop worker → cancel all resting orders → verify cleanup before exit. Per-market 3-contract cap, 20 global. Inventory risk aversion increases as inventory approaches limits.

What we learned from it: the cleanup invariant matters more than people think. A bot that crashes with resting orders on the book can take losses you never modeled. The "verify cleanup" step isn't paranoid; it's the difference between a controlled shutdown and a financial event.

nikhilnd / kalshi-market-making ★GitHub

Repo: github.com/nikhilnd/kalshi-market-making

Probability-band MM around S&P close estimates. Originally a QuantSC student project from Spring 2023. Cleaner than its origin suggests; useful as an introductory MM reference.

orangejuicetin / kalshi_market_maker ★GitHub

Repo: github.com/orangejuicetin/kalshi_market_maker

Public-facing repo of an algorithm the author actually ran on the platform. Smaller in scope than the others, but the fact that it represents real production code (not theoretical work) makes it valuable as a "what does live MM code look like" reference.

yllvar / Kalshi-Quant-TeleBot ★GitHub

Repo: github.com/yllvar/Kalshi-Quant-TeleBot

Self-described "enterprise-grade" quant system with a Telegram control interface. Claims institutional-quality risk management. The Telegram-as-control-plane pattern shows up across many bots in this list — it's become the de facto remote management standard.

Arbitrage Bots

Cross-platform arbitrage between Kalshi and Polymarket is the largest sub-genre by repo count. The easy arbs in liquid markets are heavily competed; bots in this category increasingly target sports and niche markets where pricing efficiency is lower.

meloner3 / poly-kalshi-sports-bot ★131

Cross-platform sports arbitrage between Kalshi and Polymarket. Written in Rust. The highest-starred arbitrage bot in the ecosystem.

What we learned from it: Rust shows up disproportionately in arbitrage code. The latency margins on cross-platform arb are tight enough that the language choice matters — Python's GIL becomes a real constraint when you're racing other bots for a 3¢ spread.

ImMike / polymarket-arbitrage ★82

Repo: github.com/ImMike/polymarket-arbitrage

Watches 10,000+ markets across Polymarket and Kalshi for arbitrage opportunities. Includes a FastAPI web UI with real-time opportunities, a built-in backtesting engine, dual data mode (simulation vs real), and a kill switch.

What we learned from it: the dual data mode pattern (sim/real toggle behind one interface) is the right way to ship a bot. You can validate the same code path in simulation before flipping the switch — fewer "well it worked in dev" failures.

Rezzecup / polymarket-kalshi-arbitrage-bot ★16

Sentence-transformer semantic matching to find equivalent markets across platforms. Jaccard similarity + date/entity hints + optional LLM verification (12 parallel workers via Chutes.ai). PM2 process management; FastAPI + SSE for live updates.

What we learned from it: semantic matching for cross-platform market identification is the under-appreciated half of arbitrage. Most arb bots match on string equality and miss obvious pairs. The semantic approach generalizes much better.

CarlosIbCu / polymarket-kalshi-btc-arbitrage-bot ★GitHub

Repo: github.com/CarlosIbCu/polymarket-kalshi-btc-arbitrage-bot

Specifically targets Bitcoin 1-Hour Price markets between Polymarket and Kalshi — the most cross-listed contract type with the highest gap frequency.

realfishsam / prediction-market-arbitrage-bot ★GitHub

Repo: github.com/realfishsam/prediction-market-arbitrage-bot

Auto-detects and executes arbitrage between Polymarket and Kalshi. Built on the pmxt.dev framework. The author has a useful companion blog post: "How I built a risk-free arbitrage bot for Polymarket-Kalshi."

vladmeer / kalshi-arbitrage-bot ★GitHub

Repo: github.com/vladmeer/kalshi-arbitrage-bot

General Kalshi arb bot. Smaller scope, useful as a reference implementation for arb basics.

Asset-Class Specific Bots

The largest category, organized around specific market types. These bots embed domain knowledge — weather forecasting, crypto microstructure, sports analytics — into trading logic.

Weather

suislanchez / polymarket-kalshi-weather-bot ★116

Trades Kalshi KXHIGH temperature markets plus Polymarket. Uses 31-member GFS ensemble forecasts via Open-Meteo plus BTC 5-min microstructure signals. 15% Kelly sizing capped at 5% of bankroll and $75–100 per trade. React 3-column dashboard (signals, weather, trades) with FastAPI backend; Brier score tracking; simulation mode with virtual bankroll and equity curves. Reported max profit of $1.8k.

What we learned from it: ensemble forecasts crush single-model forecasts on Kalshi weather markets. The 31-member GFS ensemble is free via Open-Meteo. The Brier score tracking inside the dashboard — explicit measurement of forecast calibration — is what separates serious weather bots from beginners.

cobra88giga / kalshi-weather-bot ★71

NWS/NOAA weather edge trading. Forecast horizon weighting (near-term 1.0, 3-day 0.75, 7-day 0.45) with regional NWS accuracy discount. Fixed $25 per position; Telegram alerts.

What we learned from it: the horizon-weighting scheme (closer forecast = more confidence) approximates the empirical accuracy decay of point forecasts. It's a simple heuristic that captures most of what a more complex Bayesian update would do.

EddieTGH / kalshi-weather-predictor ★GitHub

Custom temporal transformer ensemble with XGBoost. XGBoost achieves 1.55°F MAE; the custom temporal transformer achieves 1.0°F MAE. Optuna Bayesian hyperparameter search. Converts continuous forecasts to integer probability mass functions across bracket boundaries. Fractional Kelly with a 2% bankroll cap. Full frontend dashboard with auth.

What we learned from it: the accuracy ceiling on day-of weather forecasts is around 1°F MAE for the best ML models. That's tight enough to often pick the right 2°F bracket. If you're going to compete with this kind of bot, you need either a better model (hard) or different markets (easier).

akshatgurbuxani / Kalshi-Weather-Forecasting-Financial-Trading ★GitHub

Repo: github.com/akshatgurbuxani/Kalshi-Weather-Forecasting-Financial-Trading

Daily climate event forecasting that maps directly into trades.

Economics

cobrashadow88 / kalshi-economic-trader ★107

Pre-positions 15–60 minutes before CPI, Fed, NFP, and GDP releases using Bloomberg consensus deviation signals. Exits within 10 minutes after each release. $40 per position. Telegram alerts before entry, at release, and on exit. Historical resolution pattern matching.

What we learned from it: the pre-position-then-exit-fast pattern avoids the post-release reprice race entirely. You're trading the gap between consensus drift and market positioning, not racing the print itself. The discipline of exiting within 10 minutes is what makes this strategy work — most traders hold too long and give back the edge.

Crypto

Bh-Ayush / Kalshi-CryptoBot ★GitHub

Repo: github.com/Bh-Ayush/Kalshi-CryptoBot

BTC trader on 15-minute and 1-hour intervals. Self-described as "risk-averse and profitable." Smaller in scope than the dual-strategy bots below, but the focused approach is refreshing.

danielsilvaperez / kalshi-trading-bot ★GitHub

XGBoost model trained on 70k historical BTC candles via Kraken WebSocket. Dual SWING/SCALPER strategy. Kelly with volatility-regime overlay. Concurrent monitoring for stop-loss, take-profit, flash-crash, and time-to-expiry. systemd units plus a custom watchdog. Circuit-breaker events via Telegram.

What we learned from it: the dual-strategy pattern — different time horizons, different sizing, same execution — is a good way to extract value from a single underlying without doubling infrastructure. The circuit-breaker layer is non-optional for any bot that touches real money.

quantgalore / kalshi-trading ★36

S&P 500 daily bracket trading. Tight scope, well-executed.

Razzleberryss / AstroTick ★10

BTC 15-minute momentum + orderbook skew + OpenClaw AI enrichment. DRY_RUN mode; configurable stop-loss and take-profit cents per position. The orderbook skew signal is under-explored in this space and worth studying.

Sports

5000neoliobtc / kalshi-sports-bot ★95

NFL, NBA, MLB, NHL game outcomes. Compares Kalshi prices vs sportsbook implied probabilities. Injury feed integration; line movement detection. $30 per entry; $300 daily cap; take-profit at 0.85; stop-loss at 0.08; max 8 open positions.

What we learned from it: the sportsbook-as-reference pattern is the canonical sports edge. Sportsbook lines move faster than Kalshi for many markets; if you can detect a line move and reprice on Kalshi within seconds, you have a real edge — at least until the bot population catches up.

Multi-strategy / general

DeweyMarco / simple-kalshi-bot ★GitHub

Repo: github.com/DeweyMarco/simple-kalshi-bot

Seven distinct paper-trading strategies running simultaneously on BTC markets. Designed as an educational reference, and that's exactly what it does well.

What we learned from it: seven strategies in paper mode, simultaneously, is a great way to figure out which approaches generate which kinds of P&L profiles before you commit real money. We use a similar pattern in our own framework.

allengeer / kalshihub ★GitHub

Repo: github.com/allengeer/kalshihub

General Python algo bot — market analysis, position management, risk controls. A reasonable starting framework if you're building from scratch.

Sentiment & Reinforcement Learning

Smaller category, but the most experimental approaches live here. These bots try to extract edge from harder-to-quantify signals — social sentiment, market-microstructure anomalies, learned policies.

foxstacker5000 / kalshi-sentiment-bot ★26

Monitors Twitter/X, Reddit, and Telegram with five-dimension scoring (source count, velocity, authority, keyword match depth, momentum duration). NLP transformer classifier tuned on prediction-market language. Author claims 5–14 minute lead time before Kalshi repricing. Instant Telegram alerts with score breakdown.

What we learned from it: the multi-dimensional scoring (vs. a single sentiment number) is the right shape for sentiment signals. A 5-source spike with high authority is genuinely different from a 50-source spike from low-authority accounts; collapsing both to "high sentiment" loses the distinction.

andrewkni / bayesian-spike-detector ★GitHub

Bayesian model that distinguishes "fake" YES price spikes (low volume + widening spread) from "real" repricings (high volume + tightening spread), then fades the fakes. Enters NO when posterior mu > 0.7. Includes a CSV-based backtester tracking price/volume/spread deltas.

What we learned from it: the volume-and-spread joint signal is much stronger than either alone for distinguishing manipulation from real moves. This is the kind of strategy that works only because most participants react to price alone.

Pajamajoker / reinforcement-learning-for-kalshi-trading ★GitHub

DQN agent for hourly BTC threshold markets. Gymnasium-compatible RL environment; multi-day backtesting with policy comparison (random vs. baseline vs. DQN); Streamlit GUI.

What we learned from it: the RL framing for prediction markets is interesting but reward-shaping is hard. The author's policy comparison framework (always benchmark against a random baseline) is a useful discipline that many ML approaches skip.

Tools & Infrastructure

Not bots themselves, but infrastructure that bots and traders depend on.

alexandermazza / kalshi-trading-mcp ★2

Model Context Protocol server with 20+ tools for Claude Code/Desktop integration. Weather forecasting, ensemble analysis, position drift monitoring, safety controls. Designed for conversational trading with Claude.

What we learned from it: the MCP-as-bot-control-plane pattern is genuinely new. It lets you describe a strategy in natural language and have the LLM operate the trading tools. The safety controls inside the MCP are non-negotiable — a hallucinating LLM with raw API access is exactly as dangerous as it sounds.

hackingthemarkets / prediction-market-assistant ★50

Kalshi API + Perplexity Sonar API for research. Streamlit UI. Useful as a "research assistant" rather than a fully automated bot.

carllman13 / Kalshi_Trading ★17

Fed implied rate curve visualization and historical comparisons. Jupyter-based; not a bot, but a strong analytics reference for anyone trading Fed markets.

cameronbrock4-arch / kalshi-dashboard ★GitHub

NBA Kalshi vs. DraftKings backtest covering 876 games and 10 strategy tests. The methodology of running multiple strategy variants over the same historical window is exactly the kind of comparative work this space needs more of.

abudnick8 / prop-edge ★GitHub

Cross-platform scanner across Kalshi, Polymarket, and DraftKings for NFL, NBA, MLB, and NHL props. Confidence-scored alerts. The three-platform comparison (vs. the more common two) catches mispricings the bilateral arb bots miss.

API Client Libraries

The foundation everyone builds on. Most serious bots either use the official Kalshi Python SDK directly or wrap their own thin client.

Choice of language matters more for arbitrage and high-frequency strategies than for slower directional bots. If you're trading hourly markets, Python is fine. If you're racing other bots for cross-platform arb, Rust or Go pays for itself.

Commercial Platforms

Not open-source, but part of the landscape any developer should know.

Bot for Kalshi

Our own platform. Web-based no-code bot builder, signal system, encrypted credential storage, sub-50ms order execution. We built it because every open-source option requires maintaining infrastructure, and many traders want the automation without the engineering. Honest disclosure: we're listed here so the catalog is complete; pick the option that fits your needs.

Predict & Profit

Weather markets specifically. Runs trades through a 62-member hybrid ensemble (NOAA GFS plus AI models). The ensemble size is roughly twice what most open-source weather bots use; it's also a paid product, which is the tradeoff.

Kalshi Weather Edge

NOAA-data-driven weather market signals. Data tool rather than a bot — you still need execution.

Weather Edge Finder

Data-backed prediction-market analysis for weather. Similar profile.

TradingVPS

VPS hosting plus Kalshi bot setup tutorial. Useful if you're running open-source bots and don't want to manage your own infrastructure.

Patterns Across the Ecosystem

After cataloging this many bots, several patterns become clear:

Telegram is the default control plane. At least eight of the bots above use Telegram for entry alerts, exit alerts, P&L summaries, and circuit breakers. It's not because Telegram is technically superior — it's because it's the lowest-friction way to push notifications to a phone. Slack and Discord show up occasionally but Telegram dominates.

Fractional Kelly is universal among serious bots. Every bot above $50★ uses some form of Kelly sizing reduced by a constant factor (typically 0.25 or 0.5). Full Kelly shows up only in toy examples. Hard caps stack on top: per-position, daily, drawdown.

Paper / simulation modes are non-optional. Every bot with serious traction includes a paper or dry-run mode. The bots that skip this stage and go straight to live trading don't accumulate stars — partly because they accumulate losses and the authors abandon them.

Risk engines are stacks, not single checks. The five-gate engine in OctagonAI's bot, the volatility-regime overlay in danielsilvaperez's, the multiple stop conditions in 5000neoliobtc's — all reflect a shared insight that no single risk check is sufficient. You need a stack of independent constraints.

Cross-platform arb is the mature sub-genre. Six dedicated arbitrage bots, multiple commercial offerings, and Rust adoption in the category all point to it being the most-competed segment. The easy edges here are gone; the remaining ones live in long-tail markets and require sophisticated semantic matching.

Weather is the most-loved category. Five dedicated weather bots in the open-source list, plus three commercial offerings. The reason is structural: NWS data is free and authoritative, the markets settle on the same data, and forecast quality varies enough between providers that better forecasts genuinely produce edge.

The dual-loop pattern is emerging. Fast quantitative model plus slower LLM reasoning, with trades only firing when both agree. BEXAI's KalshiBot is the cleanest example, but variants show up in other AI-driven bots. We expect this pattern to become standard within a year.

Our Honest Take

If you're building from scratch in 2026:

Start by reading rodlaf/KalshiMarketMaker. Even if you're not building a market maker, the cleanup invariants and inventory management patterns are the most reusable code in this catalog. They define what "production-ready" looks like.

Pick your category before your stack. A weather bot looks nothing like an arbitrage bot. Decide what you're trading first; the language and architecture follow from that. Python is fine for most categories. Rust if you're racing other bots on tight latency margins.

Steal the risk engine pattern. The five-gate model from OctagonAI's bot is the right default. Write your strategy logic separately from your risk logic. Strategy says "I want to enter this trade." Risk says yes or no. The separation lets you swap strategies without rewriting safety.

Paper trade for at least 30 days. Every serious bot above does this. Strategies that look brilliant in backtest often fall apart in paper mode because your backtest didn't model spread, fade, or correlated bot reactions. Paper mode catches these.

Don't build what already exists. If you want a sports arbitrage bot, fork meloner3's. If you want a weather bot, study suislanchez's. If you want a market maker, study rodlaf's. The novelty in this space is in the categories nobody has built yet — political-mention markets, FDA approval markets, niche cultural markets — not in re-implementing what's already public.

And if you'd rather skip the infrastructure entirely and focus on strategy, that's exactly why we built Bot for Kalshi. Visual builder, encrypted credentials, signal system, kill switch — all the things you'd build yourself from scratch, ready to use.

Catalog last updated 2026-05-10. Found a bot we missed, or have an update on one above? Send us a note — we update this guide quarterly.