AutoLearn Plugin

Autonomous learning plugin for LangBot. Learns from real conversations to adapt bot behavior over time.

Features

• Conversation Style Learning — Tracks user expression patterns, common phrases, and sentence length from real chats
• Group Slang Detection — Counts high-frequency words in group chats and infers meanings (optional LLM)
• Social Relationship System — Records interactions, favorability (0–100), and mood per user
• Personality Evolution — Gradually adjusts warmth, humor, formality, empathy, and curiosity traits
• Memory Graph — Builds a knowledge/memory network with interactive visualization in WebUI

Components

Component	Description
EventListener	Captures messages, responses, and injects learned context into prompts
Command !learn	View stats, slang, relationships, and trigger slang inference
Tool query_learned_context	Lets the LLM Agent query learned data during conversations
Page Memory Graph	Visual dashboard with force-directed graph and personality traits

Configuration

Field	Type	Description
llm_model	llm-model-selector	Optional. Used for slang meaning inference via !learn infer
enable_prompt_injection	boolean	Inject learned context into system prompt (default: true)

Commands

!learn              — Overview
!learn qfx          — Group analysis chart image (group chat only)
!learn slang        — Top group slang words
!learn relation     — Relationship & favorability for current user
!learn infer        — Infer slang meanings via LLM (requires llm_model config)
!learn graph        — Memory graph statistics

Auto Learning

• Every group message is automatically learned (no command required)
• Per-speaker style, relationship and slang context is injected into LLM prompts
• Use !learn qfx in a group to get a visual analysis chart

Development

cd AutoLearnPlugin
cp .env.example .env
# Edit DEBUG_RUNTIME_WS_URL to your LangBot Plugin Runtime debug address
python -m langbot_plugin.cli.__init__ run

Then enable the plugin in your LangBot pipeline configuration.

Debug WebSocket

Set in .env:

DEBUG_RUNTIME_WS_URL=ws://your-host:port/plugin/debug/ws
PLUGIN_DEBUG_KEY=          # if your runtime requires authentication