The hardest part of a group chat bot is not writing one reply. It is deciding whether a reply should happen at all. Too much activity interrupts the room; too little makes the bot feel absent. Group Chat Plus treats participation as a decision pipeline that can be observed, tuned, and tested.
Project Overview
Repository: RhoninSeiei/astrbot_plugin_group_chat_plus.
The plugin is organized around group message intake, probability control, AI decision-making, context buffering, image descriptions, memory adapters, proactive messages, and a web management panel. It does not send every message straight to generation. It first checks whether the current situation deserves a reply.
Decision Chain
The first layer is rules and probability. Command-like messages, unrelated messages, cooldowns, and reply-density limits filter part of the input. The probability system then adjusts candidate weight based on recent replies, conversation activity, and message quality.
The second layer is AI judgment. Messages that reach this layer carry enough context for a model to decide whether a response would be useful. This turns keyword triggering into scene judgment and reduces mechanical interruptions.
The third layer is session preference. Recent fixes bring session persona and provider rules into the decision, so the same plugin does not accidentally use the wrong persona or model setup in a different session. Reply timing and generation choices can follow the session boundary.
Context and Memory
The plugin buffers messages that did not trigger a reply and can fold them into context later when a reply becomes appropriate. Images are reduced to short descriptions before they enter the decision path, avoiding repeated media processing. Long-term memory is connected through adapters so character and session boundaries remain separate.
The useful shift is from immediate response to delayed participation. The bot can observe first and answer when the conversation leaves room for it, instead of treating every message as a request.
Maintenance Lesson
For this kind of plugin, the important boundaries are stable input selection, context eligibility, and session-specific generation settings. Once those boundaries are testable modules, reply frequency, image handling, and memory integration can change without rewriting the full reply path.