_conf_schema.json

{
  "l1_buffer": {
    "type": "object",
    "description": "L1 消息上下文缓冲",
    "hint": "将最近的对话加入 LLM 上下文，超量时自动总结为长期记忆",
    "items": {
      "enable": {
        "type": "bool",
        "default": true,
        "description": "启用 L1 上下文缓冲"
      },
      "summary_provider": {
        "type": "string",
        "default": "",
        "description": "总结模型",
        "hint": "用于 L1 缓冲总结的模型，留空则使用默认 Provider",
        "_special": "select_provider"
      },
      "inject_queue_length": {
        "type": "int",
        "default": 50,
        "description": "上下文消息条数",
        "hint": "注入 LLM 上下文的最大消息条数，超量时自动总结旧消息为长期记忆。越大上下文越完整，但 Token 消耗越多，最少20条"
      },
      "image_parsing": {
        "type": "object",
        "description": "图片解析",
        "hint": "将图片内容解析为文字描述，注入 LLM 上下文",
        "items": {
          "enable": {
            "type": "bool",
            "default": false,
            "description": "启用图片解析",
            "hint": "视觉解析会产生额外 LLM 调用，默认关闭；完成限流配置后再显式启用"
          },
          "provider": {
            "type": "string",
            "default": "",
            "description": "图片解析模型 Provider",
            "hint": "选择支持视觉能力的模型，留空则使用默认 Provider",
            "_special": "select_provider"
          },
          "mode": {
            "type": "string",
            "default": "related",
            "description": "解析模式",
            "hint": "全：所有图片；相关：触发LLM/被引用/OCR命中的图片",
            "options": [
              "all",
              "related"
            ]
          },
          "daily_quota": {
            "type": "int",
            "default": 200,
            "description": "每日解析限额",
            "hint": "全局共享计数器，超限后静默跳过，每日00:00重置"
          }
        }
      }
    }
  },
  "learning": {
    "type": "object",
    "description": "学习模块",
    "hint": "从群聊对话中学习表达风格、对话样例与圈内暗语，注入 LLM 上下文让回复更贴合群氛围",
    "items": {
      "enable": {
        "type": "bool",
        "default": false,
        "description": "启用学习模块",
        "hint": "学习群聊表达风格、对话样例与暗语，采集与审查过程会消耗少量 LLM Token"
      },
      "jargon_enable": {
        "type": "bool",
        "default": true,
        "description": "启用暗语学习",
        "hint": "通过多用户滚动统计和少量批量 LLM 审查自动发现群内暗语，命中时注入解释"
      },
      "review_provider": {
        "type": "string",
        "default": "",
        "description": "审查/推断模型",
        "hint": "用于学习模块内容审查与暗语含义推断的模型，留空则使用默认 Provider",
        "_special": "select_provider"
      },
      "review_batch_size": {
        "type": "int",
        "default": 10,
        "description": "攒批审查数量",
        "hint": "待审查的对话对与表达模式累计达到此数量时触发一次 LLM 批量审查"
      },
      "pattern_max_count": {
        "type": "int",
        "default": 300,
        "description": "表达模式上限",
        "hint": "已学习的表达模式最大数量，超量时按命中率淘汰"
      },
      "pattern_decay_days": {
        "type": "int",
        "default": 15,
        "description": "表达模式衰减天数",
        "hint": "表达模式超过此天数未被命中将衰减淘汰"
      }
    }
  },
  "persona_evolution": {
    "type": "object",
    "description": "人格自迭代",
    "hint": "根据群聊真人发言的表达风格迭代 AstrBot Persona，全程保留版本快照，支持逐字差异与回滚",
    "items": {
      "enable": {
        "type": "bool",
        "default": false,
        "description": "启用人格自迭代",
        "hint": "总开关：关闭时不采集语料、不调度迭代任务；迭代执行会消耗 LLM Token"
      },
      "provider": {
        "type": "string",
        "default": "",
        "description": "分析与生成模型",
        "hint": "用于风格归纳与人格候选生成的模型，留空则使用默认 Provider",
        "_special": "select_provider"
      },
      "review_provider": {
        "type": "string",
        "default": "",
        "description": "完整人格审查模型",
        "hint": "完整人格模式的独立审查模型，留空则复用分析与生成模型",
        "_special": "select_provider"
      },
      "sample_retention_days": {
        "type": "int",
        "default": 30,
        "description": "语料保留天数",
        "hint": "风格语料超过此天数自动删除"
      },
      "sample_max_count": {
        "type": "int",
        "default": 20000,
        "description": "全局语料上限",
        "hint": "风格语料最大条数，超量时按最旧优先删除"
      }
    }
  },
  "l2_memory": {
    "type": "object",
    "description": "L2 记忆库",
    "hint": "使用 ChromaDB 存储长期记忆向量",
    "items": {
      "enable": {
        "type": "bool",
        "default": true,
        "description": "启用 L2 记忆库"
      },
      "embedding_source": {
        "type": "string",
        "default": "provider",
        "description": "嵌入模型来源",
        "hint": "Provider：使用 AstrBot 配置的 Embedding 服务（推荐，无需下载模型）；本地：使用内置 sentence-transformers 模型（需下载约 96MB 模型文件）",
        "options": [
          "provider",
          "local"
        ]
      },
      "embedding_provider": {
        "type": "string",
        "default": "",
        "description": "Embedding Provider ID",
        "hint": "填写 AstrBot 中配置的 Embedding Provider ID（可在 AstrBot 管理面板的「模型」页面查看），留空则自动使用第一个可用的"
      },
      "embedding_model": {
        "type": "string",
        "default": "BAAI/bge-small-zh-v1.5",
        "description": "本地嵌入模型",
        "hint": "仅当嵌入来源为「本地」时生效。更换后需重建记忆库",
        "options": [
          "BAAI/bge-small-zh-v1.5",
          "moka-ai/m3e-small",
          "moka-ai/m3e-base",
          "BAAI/bge-base-zh-v1.5",
          "shibing624/text2vec-base-chinese",
          "all-MiniLM-L6-v2"
        ]
      },
      "top_k": {
        "type": "int",
        "default": 10,
        "description": "检索 Top-K",
        "hint": "从记忆库检索的相关记忆数量"
      },
      "relevance_threshold": {
        "type": "float",
        "default": 0.3,
        "description": "相关性阈值",
        "hint": "低于此分数的检索结果将被过滤，0 表示不过滤，推荐 0.2-0.5"
      },
      "enable_hybrid_retrieval": {
        "type": "bool",
        "default": true,
        "description": "启用混合检索",
        "hint": "同时使用向量语义与关键词（BM25/FTS5）双路召回并以 RRF 融合排序，提升专名、术语、数字等关键词的召回率"
      }
    }
  },
  "l3_kg": {
    "type": "object",
    "description": "L3 知识图谱",
    "hint": "使用 SQLite 存储实体关系图谱",
    "items": {
      "enable": {
        "type": "bool",
        "default": true,
        "description": "启用 L3 知识图谱"
      },
      "extraction_provider": {
        "type": "string",
        "default": "",
        "description": "L3 实体提取模型 Provider",
        "hint": "选择用于知识图谱实体提取的模型，留空则使用默认 Provider",
        "_special": "select_provider"
      }
    }
  },
  "profile": {
    "type": "object",
    "description": "画像系统",
    "hint": "记录群聊画像和用户画像",
    "items": {
      "enable": {
        "type": "bool",
        "default": true,
        "description": "启用画像系统"
      },
      "analysis_provider": {
        "type": "string",
        "default": "",
        "description": "画像分析模型 Provider",
        "hint": "选择用于画像分析的模型，留空则使用默认 Provider",
        "_special": "select_provider"
      },
      "enable_auto_injection": {
        "type": "bool",
        "default": true,
        "description": "启用画像自动注入",
        "hint": "在每次LLM请求前自动注入画像到system_prompt"
      },
      "favorability_enable": {
        "type": "bool",
        "default": true,
        "description": "启用好感度功能",
        "hint": "记录用户与bot的好感度（0-100），随互动演化并影响回复语气"
      }
    }
  },
  "isolation_config": {
    "type": "object",
    "description": "隔离配置",
    "hint": "(谨慎修改)控制记忆和图谱的隔离策略",
    "items": {
      "enable_group_memory_isolation": {
        "type": "bool",
        "default": false,
        "description": "启用群聊记忆隔离",
        "hint": "各群聊记忆独立存储，关闭则全局共享"
      },
      "enable_group_isolation": {
        "type": "bool",
        "default": false,
        "description": "启用群聊用户画像隔离",
        "hint": "各群聊用户画像独立存储，关闭则全局共享，切换会导致用户画像重建"
      },
      "enable_persona_isolation": {
        "type": "bool",
        "default": false,
        "description": "启用bot人格隔离",
        "hint": "不同bot人格的记忆与画像按人格ID逻辑隔离（共享底层存储、命名空间互不可见）；切换人格仅改变可见数据，不删除或重建"
      }
    }
  },
  "scheduled_tasks": {
    "type": "object",
    "description": "梦境任务",
    "hint": "记忆的离线增量加工：时间锚定、共享扫描的记忆协调、知识归纳、persona 级 L2 清洗与每轮一次的全局 L3 维护",
    "items": {
      "enable_dream": {
        "type": "bool",
        "default": false,
        "description": "启用梦境任务",
        "hint": "高成本离线任务，默认关闭"
      },
      "provider": {
        "type": "string",
        "default": "",
        "description": "梦境任务模型 Provider",
        "hint": "选择用于执行梦境任务的模型，留空则使用默认 Provider",
        "_special": "select_provider"
      },
      "dream_stage_temporal_anchor_enabled": {
        "type": "bool",
        "default": true,
        "description": "时间锚定",
        "hint": "在合并前以确定性规则将相对时间表达（昨天、上周等）转换为绝对日期，不调用 LLM"
      },
      "dream_stage_reconciliation_enabled": {
        "type": "bool",
        "default": true,
        "description": "记忆协调",
        "hint": "共享一次向量近邻扫描，批量完成重复合并与矛盾消解"
      },
      "dream_stage_knowledge_induction_enabled": {
        "type": "bool",
        "default": true,
        "description": "知识归纳",
        "hint": "增量发现行为模式，并从 L2 提取实体关系写入 persona 隔离的 L3 图谱"
      },
      "dream_stage_l2_pruning_enabled": {
        "type": "bool",
        "default": true,
        "description": "L2 遗忘清洗",
        "hint": "按 persona 淘汰低价值向量记忆"
      },
      "dream_stage_l3_maintenance_enabled": {
        "type": "bool",
        "default": true,
        "description": "L3 全局维护",
        "hint": "每轮执行一次图谱去重、无主节点清理和低价值节点淘汰"
      },
      "dream_max_llm_calls_per_run": {
        "type": "int",
        "default": 20,
        "description": "单轮 LLM 调用硬上限",
        "hint": "达到上限立即停止后续 persona/阶段，下一轮继续"
      },
      "dream_max_runtime_minutes": {
        "type": "int",
        "default": 20,
        "description": "单轮最长运行分钟数"
      },
      "dream_max_groups_per_stage": {
        "type": "int",
        "default": 5,
        "description": "单阶段最多处理的分组数"
      },
      "dream_min_call_interval_ms": {
        "type": "int",
        "default": 500,
        "description": "Dream 两次 LLM 调用最小间隔(ms)"
      }
    }
  },
  "context_control": {
    "type": "object",
    "description": "上下文接管",
    "hint": "控制 AstrBot 内置对话历史的管理策略，确保上下文完全由本插件的 L1/L2/L3 记忆系统接管",
    "items": {
      "enable_conversation_cleanup": {
        "type": "bool",
        "default": true,
        "description": "启用上下文接管",
        "hint": "开启后，每次 LLM 请求前自动清空 AstrBot 内置对话历史(req.contexts)，上下文完全由本插件的记忆系统提供。保留对话 ID 以兼容 AstrBot 主动回复功能。关闭后 AstrBot 将保留并使用内置对话历史"
      }
    }
  },
  "proactive": {
    "type": "object",
    "description": "主动回复",
    "hint": "统一决策主动回复（跟话/跟进/发起）。更多高级参数在插件页面「隐藏参数」中调整",
    "items": {
      "enabled": {
        "type": "bool",
        "default": true,
        "description": "启用主动回复",
        "hint": "总开关。关闭后不进行任何主动发言评估（管理指令仍可用）"
      },
      "proactive_enabled": {
        "type": "bool",
        "default": false,
        "description": "启用主动发起话题",
        "hint": "开启后 Iris 会在群冷场或话题结束时主动开启新话题"
      },
      "stats_enabled": {
        "type": "bool",
        "default": false,
        "description": "启用决策统计",
        "hint": "开启后在插件页面记录每次决策的 prompt 与结果，便于调优"
      },
      "provider_id": {
        "type": "string",
        "default": "",
        "description": "决策模型 Provider",
        "hint": "主动回复统一决策使用的模型，留空则使用默认 Provider",
        "_special": "select_provider"
      }
    }
  },
  "extras": {
    "type": "object",
    "description": "辅助功能",
    "hint": "低成本的消息触发与输出优化功能",
    "items": {
      "pure_at_reply": {
        "type": "object",
        "description": "纯 @ 回复",
        "hint": "让单独 @ 机器人的消息进入标准 LLM 管道，并仅使用已有 L1 上下文",
        "items": {
          "enable": {
            "type": "bool",
            "default": false,
            "description": "启用纯 @ 回复接管",
            "hint": "启用前需在 AstrBot 平台设置中关闭「只 @ 机器人是否触发等待」（platform_settings.empty_mention_waiting），否则纯 @ 会先被 AstrBot 内置处理器拦截"
          }
        }
      },
      "error_friendly": {
        "type": "object",
        "description": "错误消息友好化",
        "items": {
          "enable": {
            "type": "bool",
            "default": true,
            "description": "启用错误消息友好化",
            "hint": "将框架发送的技术性错误消息替换为友好的提示语"
          }
        }
      },
      "markdown_stripper": {
        "type": "object",
        "description": "Markdown 格式去除器",
        "items": {
          "enable": {
            "type": "bool",
            "default": true,
            "description": "启用 Markdown 去除功能",
            "hint": "当文本以纯文本形式发送时，自动去除 Markdown 格式标记（如 **粗体**、*斜体* 等），提升可读性"
          }
        }
      }
    }
  }
}