# Chat Moderator \u2013 AI-Assisted Chat Moderation for Paper 1.21<\/p>\n
A lightweight, free, AI-assisted chat moderation plugin for Paper 1.21 Minecraft servers. ChatMod combines fast regex\/language filtering with an optional offline ONNX-based classifier to accurately moderate toxic, hateful, sexual, and harassing messages without paid APIs.<\/p>\n
## Features<\/p>\n
– Hybrid moderation: regex + optional RoBERTa ONNX AI
\n– Identity-aware hate detection (no API keys required)
\n– Language detection and blocking
\n– Tunable thresholds, punishments, and strict mode
\n– Admin tools for review, training, and status
\n– Persistent stats with autosave (SQLite\/MySQL) and auto-migration
\n– File\/DB logging and review tools
\n– Performance-optimized with async classification and caching
\n– Works fully offline (no paid APIs)
\n– Highly configurable and extensible<\/p>\n
## Installation<\/p>\n
1. Drop `ChatModer.jar` into your server `plugins\/` directory
\n2. Restart the server to generate `plugins\/ChatModer\/config.yml`
\n3. (Optional) Add an ONNX model folder under `plugins\/ChatModer\/models\/` and point the config to the ONNX file<\/p>\n
## AI Model Setup (Hugging Face RoBERTa)<\/p>\n
ChatModer supports Hugging Face RoBERTa ONNX models, e.g. `protectai\/unbiased-toxic-roberta-onnx`.<\/p>\n
Create a folder like:
\n“`
\nplugins\/ChatModer\/models\/unbiased-toxic-roberta-onnx\/
\n“`
\nPut these files inside:
\n– `model_quantized.onnx` (recommended) or `model.onnx`
\n– `config.json` (contains id2label\/label2id)
\n– `tokenizer.json` (preferred), or `vocab.json` + `merges.txt`
\n– Optional: `tokenizer_config.json`, `special_tokens_map.json`<\/p>\n
Config snippet:
\n“`yaml
\nmoderation:
\n ai:
\n enabled: true
\n model_path: plugins\/ChatModer\/models\/unbiased-toxic-roberta-onnx\/model_quantized.onnx
\n max_sequence_length: 128
\n async_threads: 2
\n timeout_ms: 500
\n“`
\nThe tokenizer\/config are auto-loaded from the same folder as the ONNX.<\/p>\n
### Label Mapping
\n– `obscene` -> `profanity`
\n– `insult`\/`threat` -> `harassment`
\n– `sexual_explicit` -> `sexual`
\n– `identity_attack` -> `hate`
\n– `toxicity`\/`severe_toxicity` boost profanity\/harassment slightly
\n– Identity-only labels (e.g., `black`, `muslim`) are not flagged by themselves; they serve as context when combined with insult\/toxicity<\/p>\n
## Modes<\/p>\n
– __regex__: Only regex\/wordlist heuristics
\n– __ai-only__: Only ONNX model (if enabled); still uses language blocking and whitelist
\n– __hybrid__ (recommended): Combine regex + AI; AI score can be weighted via `moderation.ai.weight`<\/p>\n
“`yaml
\nmoderation:
\n enabled: true
\n mode: hybrid # regex | ai-only | hybrid
\n ai:
\n enabled: true
\n weight: 0.8
\n“`<\/p>\n
## Database & Autosave<\/p>\n
Stats are persisted to SQLite by default or MySQL if configured. The plugin can auto-migrate existing SQLite stats to MySQL.<\/p>\n
“`yaml
\ndatabase:
\n type: sqlite
\n sqlite_file: plugins\/ChatModer\/data.db
\n mysql:
\n host: 127.0.0.1
\n port: 3306
\n database: chatmoder
\n user: root
\n password: “secret”
\n useSSL: false
\n sqlite_migrate_file: plugins\/ChatModer\/data.db<\/p>\n
stats:
\n autosave_seconds: 300
\n“`<\/p>\n
## Thresholds & Punishments<\/p>\n
Fine-tune moderation sensitivity and actions per category.<\/p>\n
“`yaml
\nthresholds:
\n profanity: 0.7
\n harassment: 0.6
\n hate: 0.6
\n sexual: 0.7
\n spam: 0.8<\/p>\n
punishments: ## Logging & Review<\/p>\n – __File logs__: JSONL written to `plugins\/ChatModer\/logs\/flags.log` ## Commands<\/p>\n – `\/chatmod toggle` \u2014 Enable\/disable moderation. ## Training Workflow (Optional)<\/p>\n – __Capture__: record recent messages as context for training All tools are available under `\/chatmod train …`.<\/p>\n ## Strict Mode<\/p>\n For immediate blocks (e.g., high-severity tokens), enable strict rules and reload them without a restart:<\/p>\n “` ## Permissions<\/p>\n – `chatmod.admin` \u2014 Access to admin commands ## Performance<\/p>\n – Regex-only mode for minimal overhead ## Privacy<\/p>\n All processing is local. No external API calls are made unless you enable webhooks.<\/p>\n ## Compatibility<\/p>\n – Paper 1.21 (most modern forks should work) ## Troubleshooting<\/p>\n – __ONNX model not loading__: Check `moderation.ai.model_path` and ensure `config.json` and `tokenizer.json` (or `vocab.json` + `merges.txt`) are in the same directory. ## Support<\/p>\n Open an issue on the project repository with logs, configs, and steps to reproduce. Provide model details if you use a custom model.<\/p>\n","protected":false},"excerpt":{"rendered":" ChatModer is a lightweight AI chat moderation plugin for Paper 1.21 that combines regex filtering with optional offline ONNX models to block toxic and harmful messages all without paid APIs.<\/p>\n","protected":false},"featured_media":38742,"template":"","meta":{"_minecraft_slug":"chat-moderator-local-ai","_minecraft_project_id":"uIdj1nvj","_minecraft_author":"Bobcat","_minecraft_license":"LicenseRef-All-Rights-Reserved","_minecraft_downloads":154,"_minecraft_follows":4,"_minecraft_client_side":"unsupported","_minecraft_server_side":"required","_minecraft_storage_type":"zip","_minecraft_archive_name":"chat-moderator-local-ai.zip","_minecraft_size_bytes":103602494,"_minecraft_version_count":1,"_minecraft_updated_at":"2026-05-16T13:24:43.776240Z","_minecraft_date_created":"2025-09-08T08:18:26.583521Z","_minecraft_date_modified":"2025-09-07T15:02:17.446501Z","_minecraft_project_url":"https:\/\/modrinth.com\/mod\/chat-moderator-local-ai","_minecraft_icon_attachment_id":38742,"_minecraft_imported_at":"2026-06-01T20:08:24+00:00","_minecraft_import_hash":"2d7698fbd8f7e0cdfe9654b427b76e6a","_minecraft_versions":"","_minecraft_links":"","_minecraft_gallery":"","_minecraft_api_metadata":""},"mod_category":[127],"mod_loader":[132],"minecraft_version":[62,37,73,74,38,78,79,80,81],"class_list":["post-38741","mod","type-mod","status-publish","has-post-thumbnail","hentry","mod_category-paper","mod_loader-paper","minecraft_version-1-21","minecraft_version-1-21-1","minecraft_version-1-21-2","minecraft_version-1-21-3","minecraft_version-1-21-4","minecraft_version-1-21-5","minecraft_version-1-21-6","minecraft_version-1-21-7","minecraft_version-1-21-8"],"yoast_head":"\n
\n profanity: block # warn | tempmute | kick | tempban | block | command:
\n harassment: warn
\n hate: tempmute
\n sexual: block
\n spam: block
\n“`<\/p>\n
\n– __DB logs__: SQLite\/MySQL `flags` table (optional)
\n– __Review__: `\/chatmod review
\n– `\/chatmod status` \u2014 Live stats overview.
\n– `\/chatmod review
\n– `\/chatmod whitelist add|remove
\n– `\/chatmod strict reload` \u2014 Reload strict list.
\n– `\/chatmod train add|capture|fromlog|words|stats|export` \u2014 Dataset tools.
\n– `\/chatmod stats reset confirm` \u2014 Clear in-memory stats and persist immediately.<\/p>\n
\n– __From log__: mine flagged logs into a dataset
\n– __Words__: word frequency analysis
\n– __Stats__: training dataset stats
\n– __Export__: outputs `dataset.jsonl` ready for fine-tuning<\/p>\n
\n\/chatmod strict reload
\n“`<\/p>\n
\n– `chatmod.bypass` \u2014 Bypass moderation
\n– `chatmod.review` \u2014 Access review tools<\/p>\n
\n– AI runs asynchronously and uses a quantized ONNX model on CPU
\n– Token and message caching to reduce repeated work<\/p>\n
\n– Java 17+
\n– SQLite included; MySQL requires valid credentials<\/p>\n
\n– __Classification timeout__: Increase `moderation.ai.timeout_ms` or reduce `max_sequence_length`.
\n– __High CPU__: Use `model_quantized.onnx` and keep `max_sequence_length` at 128\/256.
\n– __Review shows no entries__: Ensure logging is enabled (file or DB) and that your logs aren\u2019t empty.<\/p>\n