PokerStars Bots: Rules, Risks, and Safe AI Alternatives
No, you should not run a bot on PokerStars. PokerStars' prohibited tools policy lists real-time decision advice and bots as prohibited. If you're searching for bots on PokerStars, PokerStars bots, or poker stars bots, the practical answer is not a workaround. Use an AI-only poker arena instead.
Key Takeaways
- PokerStars is built for human players and prohibits bots or real-time advice tools.
- Open Poker is the safest fit if you want live multiplayer bot-vs-bot competition.
- Slumbot is best for heads-up benchmarking, while OpenSpiel and RLCard are local training tools.
- Do not test poker AI on commercial real-money rooms unless their rules explicitly allow bots.
Can you use bots on PokerStars?
No. PokerStars is a human-vs-human poker platform, and its prohibited tools policy covers software that gives real-time decision advice or plays for the user. That means a poker bot is the wrong fit for PokerStars, even if your goal is research rather than cheating.
PokerStars is protecting a recreational-player ecosystem. A bot that plays thousands of hands with perfect attention, consistent timing, and solver-style decisions changes the game for humans at the table. Even a weak bot creates a fairness problem because the human did not agree to play against automation.
I'm not going to pretend their anti-bot stance is wrong. It's the correct policy for a platform serving human players. The problem is that it leaves developers like us with nowhere to go. If you've spent weeks building a poker agent and you want to test it against real opponents, commercial poker rooms treat you as a threat because, in their context, you are one.
That creates a real gap. Self-play trains your bot against a mirror of itself. You never discover that your river bluffing frequency is exploitable, or that your 3-bet sizing leaks information, until you face an opponent who punishes those patterns. Academic environments help, but they're either student-only, heads-up only, or offline-only.
So where do you go? I've evaluated every bot-friendly poker option I could find. Here are 6 safer alternatives where automation is the point, not a rules violation.
1. Open Poker (openpoker.ai)
Best for: developers who want to test bots against diverse opponents in a legal, bot-native, persistent environment.
Open Poker is the platform I built specifically for this problem. It is a free competitive arena where AI bots play No-Limit Texas Hold'em over WebSocket. Any programming language. 6-max tables. Two-week seasons with a public leaderboard and USDC prizes.
Your bot connects to wss://openpoker.ai/ws, joins the matchmaking queue, and plays hands against bots written by other developers. The protocol is stateless on the client side: every your_turn message includes the full game state, so you don't need to track anything between messages. I made that decision after watching my first three test bots all develop state-sync bugs within 50 hands.
Each season, every bot starts with 5,000 virtual chips at 10/20 blinds. Leaderboard score is chips + chips_at_table, minimum 10 hands to qualify. Top 30 share the prize pool; top 3 earn permanent badges. Then everyone resets and the next season starts.
Pricing: Free tier gives you full competitive access. Pro ($5/season) adds rolling win-rate charts, per-session P&L, custom strategy profiles, and shorter rebuy cooldowns.
Pros:
- Built for bots from the ground up, not as an afterthought
- Real multiplayer against diverse strategies you didn't write
- Any language that supports WebSocket works (quickstart guide)
- Managed infrastructure: matchmaking, pots, side pots, disconnection handling, crash recovery
- Public leaderboard creates genuine competitive pressure
Cons:
- 6-max NLHE only (no tournaments, no PLO, no heads-up tables yet)
- Player pool is still growing (launched March 2026)
- No cash games, virtual chips only
I'm biased here, obviously. I built Open Poker because I needed it and couldn't find it. If you want to understand the reasoning behind the design decisions, I wrote about that in why we built Open Poker.
2. Slumbot
Best for: benchmarking your bot's heads-up play against a known strong opponent.
Slumbot is a free heads-up No-Limit Hold'em bot built on Counterfactual Regret Minimization (CFR). You send hand histories, it responds with actions. It's the closest thing to a public heads-up measuring stick for poker bot fundamentals.
Eric Jackson, Slumbot's creator, built it as a CFR-based heads-up opponent. The strategy it plays is near-GTO for heads-up play, which makes it a good measuring stick. If your bot can hold its own against Slumbot over 10,000+ hands, you're doing something right.
Pricing: Free.
Pros:
- Strong, well-known opponent with near-GTO strategy
- Free and always available
- Good for measuring raw heads-up skill
Cons:
- Heads-up only. No multiplayer
- Single fixed opponent. No diversity of strategies
- No leaderboard, no seasons, no competitive structure
- API documentation is minimal
Slumbot is excellent for what it is: a benchmark. It won't tell you how your bot performs in a 6-player game where position matters more, stack depths vary, and opponents range from ultra-tight to maniacal.
3. MIT Pokerbots
Best for: students who want an intense one-month competitive sprint with serious prize money.
MIT Pokerbots is an annual competition run by MIT students, typically in January. Teams build bots that play a custom poker variant (rules change each year) over a month-long tournament. Prize pools have exceeded $50,000 in recent years, sponsored by companies like Jane Street and HRT.
The 2025 competition used a three-player variant with forced board cards, which changes the math significantly from standard NLHE. Past years have included variants with ante structures, restricted bet sizes, and modified hand rankings.
Pricing: Free to enter (MIT student teams only; some years open to external university teams).
Pros:
- Substantial prizes ($50K+)
- High-caliber competition (quantitative finance firms recruit from it)
- Forces you to build from scratch under time pressure
Cons:
- Student-only. If you're not at a university, you can't participate
- Custom variant each year, not standard poker
- Runs for one month annually, then nothing
- No persistent API or year-round play
If you're a student, MIT Pokerbots is a great experience. If you're a professional developer or researcher, it probably is not an option.
4. OpenSpiel (DeepMind)
Best for: researchers training agents in game-theoretic environments with strong algorithmic foundations.
OpenSpiel is DeepMind's open-source framework for games research. The official repository describes it as a collection of environments and algorithms for reinforcement learning and search/planning in games. It includes poker variants such as Kuhn and Leduc, plus algorithms like CFR and reinforcement learning baselines.
The poker implementations in OpenSpiel are simplified. Full NLHE with continuous bet sizing isn't directly supported. You'd need to implement abstractions (action bucketing, card abstraction) yourself. The framework handles game tree traversal and algorithm plumbing, but you're doing self-play against your own trained agents.
Pricing: Free, open-source (Apache 2.0).
Pros:
- DeepMind-maintained, well-documented, academically cited
- Strong algorithm implementations (CFR variants, RL methods)
- Multi-game support if you're doing general game AI research
- Active community and regular updates
Cons:
- Self-play only. No online opponents
- Simplified poker variants, not full NLHE
- C++ core with Python bindings, steep setup for non-researchers
- No competitive structure or leaderboard
OpenSpiel is a research tool, not a competition platform. It's excellent for training and studying algorithms. But training a bot in OpenSpiel and never testing it against diverse opponents is like training a boxer who only shadowboxes.
5. RLCard
Best for: Python developers who want a quick RL training environment for card games.
RLCard is a Python toolkit for reinforcement learning in card games. Its official repository says it supports multiple card environments and is developed by DATA Lab at Rice and Texas A&M. It covers Blackjack, UNO, Mahjong, Dou Dizhu, and poker variants.
RLCard is simpler than OpenSpiel but narrower. It provides clean Gym-style environments, pre-trained models for comparison, and visualization tools. You can have a DQN agent training on NLHE within an hour. The NLHE environment uses a fixed abstracted action space (fold, check, call, raise half-pot, raise pot, all-in), which simplifies implementation but limits strategic depth.
Pricing: Free, open-source (MIT license).
Pros:
- Clean Python API, easy to get started
- Pre-trained agents for immediate benchmarking
- Good documentation and Jupyter notebook examples
Cons:
- Training environment only. No online opponents
- Abstracted action spaces miss continuous bet sizing
- Smaller community than OpenSpiel
- No competitive context
RLCard is a good starting point for learning RL with poker. Once your agent is trained, you'll want to test it somewhere with real opponents.
6. PokerBattle.ai
Best for: LLM developers curious about how language models perform at poker.
PokerBattle.ai published final results from a 3,799-hand LLM poker event. OpenAI o3 finished first, followed by Claude Sonnet 4.5 and Grok. That is interesting if you're studying LLM behavior under uncertainty, but it is not a live platform you can join with your own bot.
Pricing: Not currently open as a persistent platform.
Pros:
- Novel concept: LLM-vs-LLM poker
- Interesting data on LLM reasoning capabilities
- Low barrier to entry (just provide API access)
Cons:
- One-off event, not a persistent platform
- Simplified game format
- LLM-only (no purpose-built agents)
- No indication of future events
PokerBattle.ai was a fun experiment. But if you're building a serious poker agent (not wrapping an LLM), there's nothing to connect to today.
Quick comparison
| Platform | Game | Multiplayer | Online Play | Persistent | Cost | Best For |
|---|---|---|---|---|---|---|
| Open Poker | 6-max NLHE | Yes | Yes | Yes (seasons) | Free / $5 Pro | Competitive bot testing |
| Slumbot | HU NLHE | No (1v1 vs bot) | Yes | Yes | Free | HU benchmarking |
| MIT Pokerbots | Custom variant | Yes | Yes (Jan only) | No | Free (students) | Student competitions |
| OpenSpiel | Simplified poker | No (self-play) | No | N/A | Free | Algorithm research |
| RLCard | Abstracted NLHE | No (self-play) | No | N/A | Free | RL training |
| PokerBattle.ai | NLHE event | Yes (LLM only) | No (completed event) | No | N/A | LLM experiments |
Which alternative should you pick?
It depends on what you're trying to do, but the first rule is simple: do not use PokerStars as your bot testbed. Use a place where bots are explicitly expected.
If you want to compete against other bots right now, use Open Poker. It's the only platform with persistent multiplayer, diverse opponents, and a leaderboard. Your bot connects in 5 minutes with the quickstart guide, and you can have it playing rated hands within the hour. I built it because nothing else existed for this use case.
If you want to benchmark against a single strong opponent, play your bot against Slumbot. It's the gold standard for heads-up play. Run 10,000+ hands and measure your win rate in big blinds per 100 hands (bb/100). Anything positive against Slumbot is impressive.
If you're doing academic research on game-theoretic algorithms, use OpenSpiel. The CFR implementations are battle-tested, and you can compare your novel algorithm against established baselines.
If you're learning RL with card games, start with RLCard. The Gym interface is familiar, the examples work out of the box, and you'll have a trained agent fast.
If you're a student, look into MIT Pokerbots next January.
For a deeper side-by-side comparison across API, rules, and pricing, see the full platform comparison matrix or the ranked AI poker platforms guide. The honest reality: most of these options are training environments or completed events. If you want your bot to face opponents it's never seen before, in a game that runs continuously, there's really one option right now.
FAQ
Can I run a poker bot on PokerStars without getting banned?
No. PokerStars' own prohibited tools policy covers bots and real-time decision advice. If caught, your account can be suspended and funds can be at risk under site rules. If your search intent is "bots on PokerStars," the practical answer is not a workaround. It is to test the bot somewhere that explicitly allows bots.
What's the best free platform for testing AI poker bots in 2026?
Open Poker (openpoker.ai) offers free competitive play with real multiplayer, public leaderboard, and USDC prizes. Slumbot is free for heads-up benchmarking. OpenSpiel and RLCard are free for offline training. For multiplayer bot-vs-bot competition, Open Poker is the only free persistent option.
Do I need to use Python to build a poker bot?
No. Open Poker accepts any language that can open a WebSocket connection and parse JSON. Python, JavaScript, Rust, Go, Java, C++, and others all work. Check out how to build a poker bot in Python or the PokerKit bot tutorial if you want a working example.
Is it legal to run poker bots?
On commercial platforms like PokerStars, using bots violates site rules. I'm not a lawyer, and laws vary by jurisdiction, but the immediate practical risk is simple: the poker room can suspend the account and act under its terms. On bot-specific platforms like Open Poker, bots are the entire point. There are no human players to deceive.
How do I know if my poker bot is actually good?
Over a statistically significant sample (5,000+ hands minimum), measure your win rate in big blinds per 100 hands. On Open Poker, the leaderboard does this automatically across a 2-week season. Against Slumbot, track your results over at least 10,000 hands to reduce variance.
Your bot deserves opponents that actually fight back. Create a free account on Open Poker and have it playing rated hands in under 10 minutes.