Review completed hands from this table session. Automatic play waits while this is open.
Choose 2–9 seats. Each AI gets only its own hole cards and public table information.
Your default server is http://192.168.77.75:1234. Both the server address and a URL ending in /v1 work. The app adds the API path automatically and connects through its local server, so browser CORS settings are not needed.
http://192.168.77.75:1234
/v1
Start LM Studio’s server, enable access from your local network, and load your models. Click Load models, select the shared model, then Apply. Keep this app’s console window open. If you updated from an older ZIP, close its console first and start the new launcher.
A successful model list confirms connectivity; use Next AI action to test generation. HTTP 404 can also mean a selected model ID is no longer available—load the list again and reselect it.
Run a hand to see request timing and token usage.
Spark profile sends thinking-off controls, uses 160–512 output tokens, and retries at most once. Saved budgets above 512 use 256 in this profile. Use this profile manually if your Spark model has a custom alias.
For FIX5 models: turn thinking OFF in LM Studio. Start with 256 output tokens. Input context is configured in LM Studio, not here: use 4,096–8,192 tokens to hold the complete public history (8,192 for long hands). Training sequence length is separate. Existing saved budgets are preserved.
Names are picked without duplicates. If the pool runs out, unique Player names are used. Saving does not rename players in an active hand.
New bots receive a random playstyle and all bots use the shared model. Each seat keeps its playstyle until you change it. It guides the model’s decisions and changes practice-bot behavior. Models receive only their own cards, public betting history, pot size, position, hand strength and random-range equity. Formatting errors get up to two automatic retries; moves still must be legal.
Apply starts a fresh table with random $60–$200 stacks per seat. Model failures pause play; you can retry or explicitly use the built-in bot.