readme

Author: linyuan.yang

README.md

@@ -52,9 +52,12 @@ Configuration and data are persisted in `~/.sbot` on the host.
 - **Multiple LLM providers** — OpenAI, Anthropic Claude, Google Gemini, Ollama, and any OpenAI-compatible API (Azure OpenAI, Groq, Mistral, DeepSeek, etc.)
 - **Multi-agent orchestration** — ReAct recursive task decomposition + Generative multimodal, agents can be nested and composed
 - **Knowledge base** — Built-in wiki system with automatic document extraction and semantic search, referenced by agents during conversations
-- **Long-term memory** — Full extract → compress pipeline with vector-embedding semantic search
-- **MCP tools** — Standard MCP protocol (stdio/HTTP), connect to any MCP tool ecosystem
-- **Multiple channels** — Web UI, CLI, Lark/Feishu, Slack, WeCom, WeChat, REST API, WebSocket
+- **Long-term memory** — Vector-embedding semantic search for persistent context recall
+- **Conversation compaction** — Automatic conversation summarization when token usage exceeds threshold, preserving continuity while reducing consumption
+- **Insight system** — Automatic knowledge extraction from conversations into wiki and memory
+- **ACP agent support** — Agent Client Protocol integration with persistent and transient agent modes
+- **MCP tools** — Standard MCP protocol (stdio/SSE), connect to any MCP tool ecosystem
+- **Multiple channels** — Web UI, CLI, Lark/Feishu, Slack, WeCom, WeChat, OneBot (QQ), XiaoAI, REST API, WebSocket
 - **Built-in tools** — Shell execution, file system, archive operations, media file read, Python/PowerShell inline execution, cron scheduler, todo tasks
 - **Skills** — Installable prompt modules, remote install from skills.sh / Clawhub
 - **Token usage tracking** — Built-in consumption statistics with real-time visibility
@@ -104,12 +107,10 @@ Choose a mode:
 
 **5. (Optional) Enable Wiki Knowledge Base** — sidebar → **Wiki** → New
 
-Built-in knowledge base. Create pages manually (title + content + tags) or let the agent auto-extract knowledge from conversations. Auto-extraction requires an extractor model. Assign wikis to a session or channel; agents can search, read, and create pages via built-in tools.
+Built-in knowledge base. Create pages manually (title + content + tags) or let the insight system auto-extract knowledge from conversations. Assign wikis to a session or channel; agents can search, read, and create pages via built-in tools.
 
 | Field | Description |
 |-------|-------------|
-| Extractor | LLM model for auto-extracting knowledge from conversations |
-| Auto-extract | When enabled, the agent extracts valuable knowledge after each reply |
 | Shared | Off = per-thread wiki; On = shared across all threads |
 
 ---
@@ -123,8 +124,6 @@ Requires an embedding model first (sidebar → **Embeddings** → New). Then ass
 | Mode | `read_only` / `human_only` (user messages) / `human_and_ai` (full conversation) |
 | Max age (days) | Auto-expire memories after N days |
 | Embedding | Embedding model for semantic search (OpenAI, Azure, Ollama) |
-| Extractor | Extracts key facts from conversation |
-| Compressor | Merges similar memories to reduce redundancy |
 | Shared | Off = per-thread memory; On = shared across all threads |
 
 ---
@@ -135,7 +134,7 @@ Sidebar → **MCP** → New
 
 Add a server:
 - **stdio** — command + args (e.g. `npx -y some-mcp-package`)
-- **http** — remote URL + optional headers
+- **sse** — remote URL + optional headers
 
 Supports global servers shared across all agents and per-agent overrides. Servers auto-restart on failure. Then open an agent → MCP tab to attach the servers you want it to use.
 
@@ -268,4 +267,4 @@ WeChat integration connects via the iLink Bot API, with file and image support.
 
 ## Keywords
 
-`ai agent` `self-hosted` `llm server` `open source` `mcp` `model context protocol` `multi-agent` `react agent` `openai` `claude` `anthropic` `ollama` `chatbot` `lark` `feishu` `long-term memory` `vector search` `typescript` `node.js`
+`ai agent` `self-hosted` `llm server` `open source` `mcp` `acp` `model context protocol` `multi-agent` `react agent` `openai` `claude` `anthropic` `gemini` `ollama` `chatbot` `lark` `feishu` `onebot` `qq` `long-term memory` `vector search` `typescript` `node.js`

README.zh.md

@@ -52,9 +52,12 @@ docker run -d \
 - **多 LLM 供应商** — OpenAI、Anthropic Claude、Google Gemini、Ollama，以及任何 OpenAI 兼容接口（Azure OpenAI、Groq、Mistral、DeepSeek 等）
 - **多 Agent 编排** — ReAct 递归任务分解 + Generative 多模态生成，Agent 可嵌套组合
 - **知识库** — 内置 Wiki 系统，支持文档自动提取和语义搜索，Agent 对话中自动引用
-- **长期记忆** — 完整的提取 → 压缩流水线，基于向量 Embedding 进行语义检索
-- **MCP 工具** — 标准 MCP 协议（stdio/HTTP），接入任意 MCP 工具生态
-- **多渠道接入** — Web UI、CLI、飞书/Lark、Slack、企业微信、微信、REST API、WebSocket
+- **长期记忆** — 基于向量 Embedding 的语义检索，持久化上下文记忆
+- **对话压缩** — Token 用量超阈值时自动摘要压缩，保持上下文连续性的同时降低消耗
+- **洞察系统** — 自动从对话中提取知识到 Wiki 和 Memory
+- **ACP Agent 支持** — Agent Client Protocol 集成，支持持久化和临时两种 Agent 模式
+- **MCP 工具** — 标准 MCP 协议（stdio/SSE），接入任意 MCP 工具生态
+- **多渠道接入** — Web UI、CLI、飞书/Lark、Slack、企业微信、微信、OneBot (QQ)、小爱音箱、REST API、WebSocket
 - **内置工具** — Shell 执行、文件系统、归档操作、媒体文件读取、Python/PowerShell 内联执行、Cron 调度、待办事项
 - **技能系统** — 可安装的 Prompt 模块，支持从 skills.sh / Clawhub 远程安装
 - **Token 用量追踪** — 内置消耗统计，实时可见
@@ -104,12 +107,10 @@ docker run -d \
 
 **5. （可选）开启 Wiki 知识库** — 侧栏 → **Wiki** → 新建
 
-内置知识库系统，可手动创建词条（标题 + 内容 + 标签），也可由 Agent 在对话中自动提取知识。自动提取需配置一个提取模型（Extractor）。创建后将 Wiki 分配给会话或渠道，Agent 对话时可通过内置工具搜索、读取和创建词条。
+内置知识库系统，可手动创建词条（标题 + 内容 + 标签），也可由洞察系统在对话中自动提取知识。创建后将 Wiki 分配给会话或渠道，Agent 对话时可通过内置工具搜索、读取和创建词条。
 
 | 字段 | 说明 |
 |------|------|
-| 提取模型 | 从对话中自动提取知识的 LLM 模型 |
-| 自动提取 | 开启后 Agent 回复时自动提取有价值的知识 |
 | 共享 | 关闭 = 每 thread 独立；开启 = 所有 thread 共享 |
 
 ---
@@ -123,8 +124,6 @@ docker run -d \
 | 模式 | `read_only` 只读 / `human_only` 仅记用户消息 / `human_and_ai` 记录双方 |
 | 最大保留天数 | 到期自动清理 |
 | 向量模型 | 用于语义检索（支持 OpenAI、Azure、Ollama） |
-| 提取模型 | 从对话中提取关键事实 |
-| 压缩模型 | 合并相似记忆，减少冗余 |
 | 共享 | 关闭 = 每 thread 独立；开启 = 所有 thread 共享 |
 
 ---
@@ -135,7 +134,7 @@ docker run -d \
 
 添加工具服务器：
 - **stdio** — 填写命令和参数（如 `npx -y some-mcp-package`）
-- **http** — 填写远程 URL 和可选请求头
+- **sse** — 填写远程 URL 和可选请求头
 
 支持全局共享服务器和 Agent 级别独立配置，故障自动重启。然后打开 Agent 编辑页 → MCP 标签页挂载所需服务器。
 
@@ -268,4 +267,4 @@ docker run -d \
 
 ## 关键词
 
-`AI Agent` `自托管` `大模型服务` `开源` `MCP` `模型上下文协议` `多智能体` `ReAct` `OpenAI` `Claude` `Ollama` `聊天机器人` `飞书` `Lark` `长期记忆` `向量检索` `TypeScript` `Node.js`
+`AI Agent` `自托管` `大模型服务` `开源` `MCP` `ACP` `模型上下文协议` `多智能体` `ReAct` `OpenAI` `Claude` `Gemini` `Ollama` `聊天机器人` `飞书` `Lark` `OneBot` `QQ` `长期记忆` `向量检索` `TypeScript` `Node.js`

packages/desktop/src-tauri/icons/icon.icns

@@ -8,6 +8,7 @@
 - **XiaoAI Channel**: New channel for Xiaomi AI speakers — account login, device discovery, TTS playback, and polling-based conversation
 - **Token Usage Tracking**: Per-model token usage statistics with admin dashboard visualization
 - **Session Search Tool**: New tool allowing agents to search across session history
+- **Heartbeat System**: Configurable periodic agent self-activation — supports custom intervals, prompt templates, and target channels; managed via admin panel
 - **Middleware Pipeline**: New `MiddlewarePipeline` with intent-based filtering for session message processing
 - **Timer Executor**: Extracted `TimerExecutor` utility for unified scheduled task execution (heartbeat, scheduler, wiki indexing)
 - **Workspace Prompt Discovery**: `ContextFileDiscovery` supports loading workspace-level prompt files

packages/sbot/ReleaseNote.md

@@ -8,6 +8,7 @@
 - **小爱渠道**: 新增小米 AI 音箱渠道 — 账号登录、设备发现、TTS 播放和轮询式对话
 - **Token 用量统计**: 按模型统计 Token 使用量，管理后台提供可视化面板
 - **会话搜索工具**: 新增工具允许 Agent 搜索历史会话记录
+- **心跳系统**: 可配置的 Agent 定时自激活 — 支持自定义间隔、提示词模板和目标渠道，通过管理后台统一管理
 - **中间件管道**: 新增 `MiddlewarePipeline`，支持基于意图的会话消息过滤
 - **计时器执行器**: 抽取 `TimerExecutor` 统一调度工具（心跳、定时任务、Wiki 索引）
 - **工作区提示词发现**: `ContextFileDiscovery` 支持加载工作区级别的 prompt 文件