Skip to content

Advanced settings ​

The 高级 page groups six cards of low-frequency timing parameters and system-level switches.

读取范围 (Driver read scope) ​

Controls which conversations the driver syncs from the client database into its mirror—forming the outer boundary enforced before the bot's allowlist:

  • 全部会话 (All conversations, default): the driver mirrors every conversation, and you choose which ones the bot participates in via 参与 / 只记录 on Conversations and policies.
  • 仅选中的会话 (Selected conversations only): the driver mirrors only the selected conversations. Switching from 全部会话 to 仅选中的会话 automatically seeds the selection with the conversations currently in scope so existing mirrored history is not dropped; individual conversations are toggled in the 状态 column on Conversations and policies.

发送节奏 (Send pacing) ​

Every outbound message goes through the driver's single serialized action queue. The minimum interval spaces consecutive actions apart, while the per-conversation character budget charges each send at the configured typing rates (Chinese and other characters share one sliding window budget per conversation, with ±15 % random jitter), holding back bursts until room frees up in the window:

SettingDefaultMeaning
最小间隔 (Min interval)3 sMinimum delay between any two actions
中文速度 (Chinese rate)240 chars/minTyping rate charged for CJK characters and fullwidth punctuation
其他字符速度 (Other rate)960 chars/minTyping rate charged for Latin letters, digits, emoji, and whitespace
图片和贴纸 (Image / sticker cost)10 charsChinese-character equivalent charged for one image, sticker, or file
小黄脸 (Built-in face cost)2 charsChinese-character equivalent charged for one built-in QQ face
统计窗口 (Budget window)60 sSliding time window over which a conversation's character budget is measured (0 disables the character budget and keeps only the minimum interval)
最长等待 (Max rate wait)900 sMaximum time a single send may wait for budget room before failing with an error

Keep the default pacing: when automating a desktop client, natural send rates matter as much as the integration method.

合并消息 (Message batching) ​

Controls how long consecutive messages arriving in one conversation are collected into a batch before a turn starts:

SettingDefaultMeaning
静默等待 (Quiet wait)2 sWait after the last message in an idle conversation
轮次进行中 (Turn running)5 sWait while a turn is already running
同一人连续发 (Same sender)2 sWait while the same sender keeps sending messages
换人发 (Sender change)0.5 sWait when a different sender replies
被叫到的上限 (Mention cap)1 sMaximum wait once a message in the batch addresses the assistant
一批的上限 (Batch cap)8 sMaximum total wait measured from the first message in the batch

排队 (Turn queue) ​

Controls which waiting conversation runs next when several are queued (mentions and private chats always run ahead of group join-ins and timer wakes):

  • 饿等上限 (Starvation guard, default 120 s): any conversation that has waited longer than this limit moves to the front regardless of Jev's scores.
  • 需要程度 (Urgency via Jev): scores waiting conversations (0–2 threshold) by how urgently they need a reply now. Supports 关闭 (off), 只记录不生效 (shadow, the default), and 启用 (active), plus an optional 额外判断 (custom criterion).

工具 (Tool exposure) ​

Toggles code_exec (enabled by default). When enabled, secondary and edge tools are placed behind the code_exec JavaScript runner so the assistant can chain lookups and filtering in a single script call; turning it off exposes all edge tools directly as a fallback for models that struggle with code_exec. Changes take effect at the next window rollover.