All documentation
Documentation

AI Chatgame Solver

A client-side Fabric mod that detects server chat games and uses AI to type the correct answer fast.

v1.0.14 Fabric Loader 0.17.0+Fabric APIMinecraft 1.21.11, 26.1, 26.2 or 26.3An AI provider API key (Groq and Google AI Studio are free)Mod Menu (optional)

Quick start

  1. Install Fabric and Fabric API, then put the mod jar in your mods folder.
  2. Start the game and press U to open the menu.
  3. Open AI provider, pick Groq or Google AI Studio, paste your key and press Test connection.
  4. Open Answers & typing and choose Auto-send, Fill chat or Show only.
  5. Join a server with chat games. The status overlay shows when it detects and answers one.

#Overview

AI ChatGame Solver is a client-side Fabric mod. It watches server chat for chat games (unscramble, math, trivia and similar), works out the answer, and types it for you. It needs no server plugin and nothing is installed on the server.

The AI is not called for every message. Four steps run on your client:

  1. Read. The mod reads chat from the raw message data, so custom fonts and resource packs do not get in the way.
  2. Filter. A local filter drops normal chat. No AI call, no cost.
  3. Solve. Built-in solvers handle math, reverse and some unscrambles. Anything else goes to the AI provider you picked.
  4. Type. The answer is sent after a short random delay, put in your chat box for you to send, or only shown to you.
Check the server rules

Flagged chat lines are sent to the AI provider you choose. Automated chat is not allowed on every server, so check a server's rules before you use the mod there. Use the Show only or Fill chat answer mode if you want to stay in control.

#Installation

Requirements

  • Fabric Loader 0.17.0 or newer.
  • Fabric API.
  • Minecraft 1.21.11, 26.1, 26.2 or 26.3 (Java 21 for 1.21.11, Java 25 for 26.x, which is what those Minecraft versions already require).
  • An API key from an AI provider. Groq and Google AI Studio offer free keys.
  • Mod Menu is optional. If you use it, the mod adds a config button there.

Install

  1. Install Fabric for your Minecraft version.
  2. Put Fabric API and the AI ChatGame Solver jar in your mods folder.
  3. Start the game. The same jar works on every supported version and picks the right build automatically.
  4. Press U in game to open the menu.

The mod is client only. It does not belong on a server.

Files

All files are created in the config folder of your game directory.

FileWhat it holds
aichatgamesolver.jsonAll settings, including your API key.
aichatgamesolver_stats.jsonSolve statistics and round history.
aichatgamesolver_answers.jsonAnswers learned from servers.
aichatgamesolver_updater.jsonUpdate checker settings.
Your key stays local

The API key is stored only in aichatgamesolver.json. Do not share that file.

#AI provider and API key

Open the menu (U), then AI provider.

ProviderNotesDefault model
GroqFree keys from console.groq.com. Keys start with gsk_.openai/gpt-oss-20b
Google AI StudioFree keys from aistudio.google.com. Keys start with AIza.gemini-2.5-flash
OpenAIYour own API key. Keys start with sk-. This is the default provider.gpt-4o-mini
AnthropicYour own API key.claude-haiku-5-5
CustomAny OpenAI-compatible endpoint, such as OpenRouter or LM Studio. Set the Base URL yourself.You type it
  1. Click the provider button to cycle to the one you want.
  2. Paste your key into the key box (click it and press Ctrl+V) or use Paste.
  3. Press Next model to cycle through known models, or type a model name.
  4. Press Test connection. It sends one tiny request and tells you in plain English what is wrong if it fails.

The test explains common problems: a retired model (with a one-click fix that switches to a working one), a rejected key, a rate limit or quota, no network, a provider outage, a context that is too large, or a model that used its whole token budget on hidden reasoning.

Start free

Groq with openai/gpt-oss-20b or Google AI Studio with gemini-2.5-flash works for chat games without a paid plan.

#How it detects and answers

Detection

Every chat line passes through a local filter first. By default only a line that contains one of your keywords goes on (for example unscramble, trivia, first to type, quick math). Lines that arrive in quick bursts are grouped, so multi-line announcements work.

The filter drops these before any AI call:

  • Player chat. Lines in <Name> text or [Rank] Name » text form, lines where an online player's name (from the tab list) comes before :, » or >, and anything matching your player-chat patterns. A player cannot trigger the solver by typing a fake announcement.
  • Your ignore phrases, which by default include vote and purchase broadcasts and result lines such as "time's up", "the answer was" and "winner:".
  • Finished rounds. If the server announces a result, any answer that is still waiting or being worked out is dropped.
  • Server notices that only explain chat games.
  • A round that was already answered.

Game formats

The mod is not tied to one chat-game plugin. It handles:

  • Unscramble: uses every letter once and keeps the same number of words. Capitals from the scramble are kept.
  • Reverse: flips the whole text back, spaces included.
  • Math: types the final number only.
  • Trivia and riddles: types the shortest correct answer.
  • Emoji decode: turns emojis or icons into the word or phrase.
  • Type this phrase: types the exact phrase shown.
  • Fill in the gaps: solved from the game's item and mob names. An AI answer must fit the gaps or it tries again.
  • Guess the number: see Built-in answers below.

Quick detect

When a flagged burst clearly matches a known format ("Unscramble the word", "Solve the equation"), the AI check is skipped and the round goes straight to solving. This is on by default.

Fast mode

Fast mode (on by default) combines classification and solving into one AI call and opens the connection early. It roughly halves the time to an answer. With it off, the mod uses a classic two-step pipeline.

Status overlay

A small panel in the top-left corner shows what the solver is doing: chat game detected, AI is working, answering, or an AI error (see the Activity log). Turn it off in the menu for nothing on screen.

#Answers and typing

Menu, Answers & typing.

Answer mode

ModeWhat happens
Auto-send (default)The answer is typed and sent for you after a short random delay.
Fill chatChat opens with the answer filled in. You check it and press Enter.
Show onlyThe answer appears in your chat (only you see it) and is copied to the clipboard. Nothing is typed.

Settings

SettingDefaultWhat it does
Min confidence0.80Answers the AI is less sure about are dropped. Lower it to about 0.60 if it never answers. Raise it if it types wrong answers.
Max answer length48Longer answers are discarded instead of typed.
Lowercase answersOffTypes everything in lowercase. Many quiz servers ignore capitals.
Delay (min and max)250 to 900 msA random wait before typing, measured from when the round was detected.
Give up after10 sAnswers older than this are discarded.
Max solve tokens600Output budget for the solving call. Reasoning models need more (1500 to 3000). If the log says "truncated by the token cap", raise it.
AI requests per hourNo limitProtects free-tier quotas and costs. When the limit is hit, rounds are skipped and the Activity log says why. The menu shows how many were used in the last hour.

Safety checks

  • A round that mentions a command such as /pay or /msg is skipped.
  • An answer that starts with /, . or ! is never sent.
  • An answer that contains the name of an online player is never sent.
  • Answers are dropped if they get too old, if the round ends first, or if you change server.

#Built-in answers (no AI)

Menu, Built-in answers. These cost nothing and are checked before the AI.

  • Math: plain sums like 144 ÷ 12 = ? are worked out locally. Equations with an unknown still go to the AI. On by default.
  • Unscramble and reverse: reverse rounds, and unscrambles that spell exactly one Minecraft word. The word list includes items, mobs, blocks and potion effects, and a phrase is solved word by word when the whole phrase is not a known name. On by default.
  • Guess the number: in rounds like "guess a number between 1 and 20" it keeps guessing until it wins or the round ends. It uses higher and lower hints when the server gives them and skips numbers other players already tried. It needs Auto-send and stops at the max guesses (15 by default).
  • Learn answers: when a server reveals the answer after a round ("Answer: 12", "Word: COMPUTER"), it is saved and reused the next time. You can clear the saved answers with one button.
  • Custom answers: your own list, written as trigger => answer. If a chat game contains the trigger, that answer is used. This list is checked first.
what is the server name => MyServer

Turn on Instant built-in in Speed & retries to send these answers without the random delay.

#Chat filter

Menu, Chat filter.

  • Trigger mode. Keywords only (default, recommended): a line reaches the AI only if it contains one of your keywords. Smart also uses formatting and message bursts to catch games with no clear label, but reacts to more normal chat and uses more API calls.
  • Keywords. Words that mark a chat game, such as your server's game name.
  • Player-chat patterns. Regexes (one per entry, case-insensitive) that recognize player chat on servers that rebroadcast it as system messages. If fake games still get through, add your server's exact chat format here.
  • Ignore phrases. Lines containing any of these are never sent to the AI.

Advanced tuning

SettingDefaultWhat it does
Symbol ratio0.25Flags a line as decorated when this share of its characters are symbols.
Burst count2How many fast non-player lines count as a game burst.
Debounce350 msHow long to wait after chat goes quiet. Raise it if the AI gets half-finished questions.
Chat buffer20 linesRecent lines kept for burst detection and context.
Context lines2Extra lines sent around the question. Set 0 to use fewer tokens.

The default keywords are: chat game, chatgame, chat-game, unscramble, scramble, reverse, unreverse, trivia, riddle, first to type, first to answer, first one to type, guess the, type this, quick math, quick maths, decode, solve the equation, answer the question, word scramble.

#Speed and retries

Menu, Speed & retries.

SettingDefaultWhat it does
Answer retries1Extra AI attempts, with stricter rules, when an answer is rejected (wrong letters, placeholder, too long).
Network retries1Extra attempts after a network or server error.
Retry wait1.0 sWait before a network retry. Longer helps with rate limits.
AI timeout10 sHow long to wait for the AI before giving up.
Fast modeOnOne AI call instead of two.
Quick detectOnSkip the AI check for obvious game formats.
Instant built-inOffSend built-in answers without the random delay.
Debounce350 msLower starts sooner.

#Servers: context, allow and block lists

Server context

Menu, Server context. Add a server (a full or partial address, for example play.example.net or just example), then write a short note about it and add memories. The note and memories are sent with every AI request on a matching server, which helps with questions about the server itself.

  • Note: what the server is, for example "a skyblock server where rewards are coins".
  • Memories: facts about the server. Each is sent to the AI in full as a bullet point.
  • Forget server deletes a server's note and memories.

Allow and block lists

  • Allow list: if it is not empty, the solver runs only on servers whose address contains one of the entries.
  • Block list: the solver never runs on a server whose address contains one of the entries.

Matching is a case-insensitive "contains" check on the server address.

#Stats, history and activity log

  • Stats & history shows solves, rounds, solve rate and errors, with charts for the last 7 days, 8 weeks and 12 months, and the average time from detection to typed answer. Click any recent round to see the question, the answer and the step timings. Clear all stats wipes everything.
  • Activity log lists every AI failure and pipeline event, newest first. If a round was not answered, the reason is here. The button also shows how many errors there are.
  • Debug messages (a toggle in General) prints every pipeline decision as gray chat text. Use it only while diagnosing a problem.

#Controls and menu

There are no commands and no server permissions.

ControlDefaultWhat it does
Open configUOpens the menu. Rebind it in Options, Controls, Key Binds, AI ChatGame Solver.
Turn AI ChatGame Solver On/OffUnboundToggles the master switch and shows "AI ChatGame Solver: on" or "off" in the action bar.
Menu key button-In the menu, click it and press a key to use a different open key. Esc cancels, Backspace resets to U. This key only works when no screen is open.
Mod Menu-If Mod Menu is installed, the mod adds a config button there.

Menu pages

SectionPage
GeneralSolver on/off, Fast mode, Status overlay, Debug messages, Update checks, Auto-update, an update banner when one is available.
SetupAI provider, Answers & typing, Chat filter, Server context, Speed & retries, Built-in answers.
Servers & activityAllow list, Block list, Activity log, Stats & history.

Hover any setting for a tooltip with its description and default value.

#Configuration file

Everything in the menu is saved to config/aichatgamesolver.json and applies right away. You can edit the file while the game is closed. The main fields and their defaults:

FieldDefault
masterEnabledtrue
hudEnabledtrue
fastModetrue
debugLoggingfalse
providerOPENAI (also ANTHROPIC, GROQ, GOOGLE_AI_STUDIO, CUSTOM_OPENAI_COMPATIBLE)
modelgpt-4o-mini
requestTimeoutMs10000
keywordOnlyModetrue
quickDetecttrue
debounceMs350
minConfidence0.8
maxAnswerLength48
solveMaxTokens600
responseDelayMinMs / responseDelayMaxMs250 / 900
stalenessLimitMs10000
autoSend / showOnlytrue / false
maxAiCallsPerHour0 (no limit)
localMath, localWordGames, learnAnswers, numberGuessingtrue
maxGuesses15
answerRetries / networkRetries1 / 1
serverAllowList / serverBlockListempty

Custom endpoints use customBaseUrl (default http://localhost:1234/v1). Your key is the apiKey field.

Edit with the game closed

The mod rewrites this file when you change settings in the menu. Edit it by hand only while Minecraft is not running.

#Updater

The mod checks for new versions on launch, when you join a server (at most every 10 minutes) and every 3 hours by default. Notices appear as popups in game, and an update banner shows in the menu.

  • Copies downloaded from BuiltByBit are checked against BuiltByBit.
  • Copies from Modrinth or CurseForge are found through the jar's hash on Modrinth.
  • Click the banner in the menu to download an update. A downloaded update is verified (valid jar, right mod and version, supports your Minecraft version) and installs when you close the game. The old jar is kept in a backup folder at .aichatgamesolver-update/backup inside your game directory.
  • Auto-update (off by default) downloads and installs new versions by itself.

Settings live in config/aichatgamesolver_updater.json:

SettingDefaultWhat it does
enabledtrueTurn update checks on or off. Applies on next launch.
checkOnStartuptrueCheck when the game starts.
checkOnServerJointrueCheck when you join a server.
autoDownloadfalseDownload and install updates automatically.
notifyInGametrueShow popups.
notifyWhenCurrentfalseAlso tell you when you are up to date.
checkIntervalHours3Hours between checks (1 to 168).
modrinthProjectblankModrinth slug, used only when the jar's hash is not known.

#Troubleshooting and FAQ

Nothing happens when a chat game appears.

Check in order: the Solver toggle is on, an API key is set (the menu header says "Setup needed" if not), the server is not on your block list (or is on your allow list if you use one), and the chat line contains one of your keywords. Open the Activity log: every skipped or failed round has a reason there. Turning on Debug messages shows each filter decision in chat.

Test connection fails.

Read the message under the button. A rejected key means the key is wrong or belongs to another provider. A model error means the provider retired the model: press Next model, or use the one-click "Use ... and retest" button. A rate limit means wait a minute or check your usage. A network error means check your internet and the Base URL.

The log says "truncated by the token cap".

Your model spent its whole token budget on hidden reasoning. Raise Max solve tokens in Answers and typing to 1500 to 3000, or use a non-reasoning model.

It answers too slowly.

Keep Fast mode on, keep Quick detect on, use a fast model such as Groq or Gemini Flash, and lower the delay range. Turn on Instant built-in so math and word answers skip the random delay. Lower Debounce if rounds start late, but not so low that half-finished questions are sent.

It never answers, or the confidence is too low.

The AI may be less sure than Min confidence (0.80). Lower it toward 0.60. Also check Max answer length and the "Give up after" time.

It reacts to normal chat.

Leave Trigger mode on Keywords only, remove broad keywords, and add noisy phrases to Ignore phrases. Smart mode is more eager and costs more API calls.

A player's fake announcement fooled it.

Add your server's exact chat format to Player-chat patterns as a regex. The mod also ignores lines where an online player's name appears before the chat separator.

It stopped after one guess in a guess-the-number round.

Guess the number needs Auto-send. Check that the option is on and that Show only is off.

Is this allowed on my server?

Not everywhere. Automated chat is against the rules on some servers. Read the rules first. Show only and Fill chat give you control over what is sent.

Where is my API key stored? Who sees my chat?

The key is saved only in config/aichatgamesolver.json. Chat lines that pass the local filter are sent to the AI provider you chose, and nothing is sent for normal chat.

The mod does not load.

Make sure you have Fabric Loader 0.17.0 or newer, Fabric API, and a supported Minecraft version (1.21.11, 26.1, 26.2 or 26.3). Check the game log: the first line from the mod names the build it picked.

Update checks never find anything.

Checks need an internet connection and enabled must be true in the updater file. Use the banner's try again button in the menu. Modrinth and CurseForge copies are looked up through Modrinth by file hash.

#Updating

  1. Close the game, replace the old jar in mods with the new one, and start the game. Your settings, stats and learned answers stay in config.
  2. Or let the mod do it: when the menu shows "Update is available", click the banner to download it, then close the game to install. Enable Auto-update to do this without clicking.

If something goes wrong after an update, the previous jar is in .aichatgamesolver-update/backup in your game directory.