Skip to content

Releases: hrygo/divinesense

v0.101.0

21 Feb 17:56

Choose a tag to compare

What's Changed

🚀 Features

  • feat(frontend): add streaming progress bar for schedule creation (#276)

🔧 Refactoring

  • refactor(agents): upgrade hotplex to v0.7.4 and restructure System Prompt (#277)
    • TaskSystemPrompt renamed to TaskInstructions
    • Session-level persistence for TaskInstructions
    • HotPlexClient interface instead of concrete *Engine
    • Namespace isolation: divinesense-geek and divinesense-evolution
  • refactor(ai/runner): CCRunner 多路复用架构升级与优雅关闭 (#272)
  • refactor(api/v1): DRY/SOLID 违约技术债务清理 (#270)

📚 Documentation

  • docs: update documentation for CCRunner v2.0 Hot-Multiplexing architecture (#273)
  • docs: add optimized refactoring plan for API v1

🐛 Bug Fixes

  • fix(ccrunner): add stderr scanner error logging
  • fix: resolve AIService redeclaration in tests and update API v1 report

Dependencies

  • hotplex: v0.6.3 → v0.7.4

Full Changelog: v0.100.2...v0.101.0

v0.100.2

19 Feb 16:04

Choose a tag to compare

v0.100.2 - 配置重构与调试增强版本

🐛 Bug Fixes

  • CCRunner 调试增强: stderr 日志采样从 10% 提升到 100%,添加 CLI 执行详情 debug 日志
  • 启动脚本: 修复后端进程退出等待问题
  • 日志级别: 移除 mode-based 日志级别覆盖,改用环境变量 LOG_LEVEL 控制

✨ New Features

  • @ 指令限制: @ 提及专家代理功能限制在普通模式,避免在特殊模式下误触发
  • LOG_LEVEL 环境变量: 支持通过 DIVINESENSE_LOG_LEVEL 控制日志级别

♻️ Refactoring

  • 环境变量重构: 统一使用 DIVINESENSE_ 前缀,移除 MEMOS_ 前缀兼容
  • Explore 页面重构: Memo UI/UX 优化与页面重构

Full Changelog: v0.100.1...v0.100.2

v0.100.1

17 Feb 15:58

Choose a tag to compare

新功能

  • 通用任务直接响应: Orchestrator 现在支持通用任务直接响应,无需调用专家代理 (#252, #253)

性能优化

  • 路由延迟优化: 添加数据库索引 idx_ai_block_conversation_round_desc 优化 GetLatestAIBlock 查询
    • 路由延迟从 ~284ms 降至 <5ms
    • 提升多轮对话响应速度

Bug 修复

  • 多轮对话上下文丢失: 修复 Orchestrator 模式下多轮对话上下文丢失问题 (#258)
  • 通用代理问题: 修复 GeneralParrot 代理相关问题 (#258)
  • 标题生成问题: 修复标题生成相关问题 (#258)
  • Block UI 未渲染: 修复 Block UI 未渲染和刷新后卡在初始化问题
  • 多轮对话 3 个问题: 修复多轮对话的多个关联问题 (#254)
  • Orchestrator 多轮对话 bug: 修复 Orchestrator 模式多轮对话的多个 bug (#250)
  • 路由置信度计算: 修复路由置信度计算和 Agent 工具描述问题

数据库迁移

现有部署需执行以下迁移:

psql -d divinesense -f store/migration/postgres/migrate/20260217000001_optimize_get_latest_block.up.sql

完整变更日志

v0.100.0...v0.100.1

v0.100.0 - 喜迎马年春节 🐴

16 Feb 14:24

Choose a tag to compare

v0.100.0 - 喜迎马年春节 🐴

这是一个重要的功能增强版本,带来了 AI 搜索、UI 体验和架构的全面升级。

✨ New Features

🤖 AI 智能增强

  • AI Memo 摘要生成:实现 AI Memo 摘要生成与内容增强架构,自动为笔记生成智能摘要
  • Schedule 日程删除:实现 ScheduleParrot 日程删除功能,完整日程管理
  • Expert Agent Handoff:实现专家代理之间的高效切换机制
  • Context Engineering:完整实现 Context Engineering 三阶段架构,增强长期记忆能力

🔍 智能搜索升级

  • 混合检索架构 (Phase 1 & 2):向量检索 + 全文检索 + Rerank 重排三阶段架构
  • 智能质量过滤:修复 Low 质量结果被 minScore 过度过滤的问题
  • topK 截断修复:修复 Rerank 后 topK 正确截断
  • 移除分数列:搜索结果直接使用 Rerank 排序,移除分数显示

🎨 UI/UX 优化

  • 彩色便签设计:全新的彩色便签样式,视觉体验更丰富
  • Zen Kanban 布局:支持禅意看板布局,专注写作
  • 统一侧边栏折叠:侧边栏折叠状态全局统一
  • FixedEditor 折叠:编辑器支持折叠/展开
  • MemoBlockV3 展开状态:修复多个展开/收起状态问题
  • 全屏模式滚动:修复全屏模式滚动问题
  • 笔记列表性能:优化笔记列表加载性能与用户体验

🏗️ 架构改进

  • 路由系统重构:配置驱动路由,从代码迁移到 YAML 配置,支持 HILT 反馈优化
  • Memory Extension Point:抽象记忆生成扩展点
  • Memory 架构重构ai/services/memory/ai/memory/simple/
  • SOLID/DRY 重构:多轮代码重构,提升代码质量和可维护性

🧪 测试增强

  • E2E 测试基础设施:搭建完整的端到端测试体系
  • 手动验收测试:完善手动验收测试用例文档

🐛 Bug Fixes

  • 修复日志描述"三层"改为"两层"
  • 修复 LLM JSON 解析失败 & 超时配置化
  • 修复多个搜索质量和路由问题
  • 修复 Orchestrator 并发、依赖注入和弹性问题
  • 修复 golangci-lint 问题

祝大家马年吉祥,万事如意! 🐴🎊

v0.99.0 - Orchestrator-Workers Multi-Agent Architecture

12 Feb 15:30
a1c7b94

Choose a tag to compare

🎯 Major Architecture Upgrade

This release introduces a complete Orchestrator-Workers architecture that replaces the previous single-agent model with a coordinated multi-agent system. This is a significant architectural evolution enabling better task decomposition, parallel execution, and expert agent coordination.

🏗️ New Architecture

用户输入
    ↓
┌─────────────────────────────────────────┐
│            Orchestrator                 │  ← LLM 驱动任务分解
│  ┌─────────────────────────────────┐   │
│  │  Decomposer (任务分解)           │   │
│  │  Executor  (并行执行)            │   │
│  │  Aggregator (结果聚合)           │   │
│  └─────────────────────────────────┘   │
└────────────────┬────────────────────────┘
                 │
        ┌────────┴────────┐
        ↓                 ↓
┌───────────────┐ ┌───────────────┐
│ MemoParrot    │ │ ScheduleParrot│  ← Expert Agents (配置化)
│ (灰灰)        │ │ (时巧)        │
└───────────────┘ └───────────────┘

✨ New Features

  • Time Context Injection: Automatic temporal context for better scheduling
  • DAG Dependency Support: Tasks can declare dependencies and execute in correct order
  • Externalized Prompts: All prompts moved to config/orchestrator/*.yaml for easy tuning
  • Structured Agent Protocols: Schedule and Memo agents enhanced with structured protocols

🧹 Code Cleanup

Removed Lines
Legacy scheduler (scheduler_v2.go) -300
Old prompts (prompts.go) -755
Memory service (services/memory/) -1,100
Fast create handler -136
Habit learning module (ai/habit/) -300

📦 Installation

Docker (Recommended):

docker pull ghcr.io/hrygo/divinesense:v0.99.0

Binary: Download from assets below

📚 Documentation


Full Changelog: v0.98.0...v0.99.0

DivineSense v0.98.0

10 Feb 17:49
b261cc8

Choose a tag to compare

DivineSense v0.98.0

🚀 AI Support Enabled

This release includes full AI support with sqlite-vec:

  • Vector similarity search with vec0 virtual tables
  • Conversation persistence and episodic memory
  • Multi-agent AI system (MemoParrot, ScheduleParrot, AmazingParrot)

📦 Available Platforms

  • macOS Intel: divinesense-darwin-amd64
  • macOS ARM64 (Apple Silicon): divinesense-darwin-arm64
  • Linux AMD64: divinesense-linux-amd64
  • Linux ARM64: divinesense-linux-arm64

🔧 Build Details

  • Go version: 1.25
  • sqlite-vec version: v0.1.7-alpha.2
  • Build tags: sqlite_vec
  • CGO enabled for static linking

📚 Documentation


Full Changelog: ...v0.98.0

DivineSense v0.97.0

10 Feb 10:52

Choose a tag to compare

DivineSense v0.97.0

🚀 AI Support Enabled

This release includes full AI support with sqlite-vec:

  • Vector similarity search with vec0 virtual tables
  • Conversation persistence and episodic memory
  • Multi-agent AI system (MemoParrot, ScheduleParrot, AmazingParrot)

📦 Available Platforms

  • macOS Intel: divinesense-darwin-amd64
  • macOS ARM64 (Apple Silicon): divinesense-darwin-arm64
  • Linux AMD64: divinesense-linux-amd64
  • Linux ARM64: divinesense-linux-arm64

🔧 Build Details

  • Go version: 1.25
  • sqlite-vec version: v0.1.7-alpha.2
  • Build tags: sqlite_vec
  • CGO enabled for static linking

📚 Documentation


Full Changelog: ...v0.97.0

v0.97.0

10 Feb 02:55
4c6744d

Choose a tag to compare

What's Changed

SQLite 向量搜索支持 (PR #131)

为 SQLite 数据库添加向量搜索支持,使用 sqlite-vec 扩展实现高效的语义检索功能。

✅ 已实现

  • 向量搜索: 使用 sqlite-vec 扩展实现 O(log n) 的 KNN 搜索
  • 向量存储: memo_embedding 表支持 BLOB (vec0) 和 TEXT (JSON) 双格式
  • Go Fallback: 当 sqlite-vec 不可用时,应用层 cosine 相似度计算
  • 安全验证: SQL 注入防护 (isValidTableName, SHA-1 临时表名)
  • 静态链接: 自动下载 sqlite-vec 静态库,支持交叉编译

⚠️ 重要变更

  • SQLite 驱动从 modernc.org/sqlite 切换到 mattn/go-sqlite3 (AI 模式需要 CGO)
  • 删除了已废弃的 ai_conversationai_message 表定义
  • SQLite 当前仅支持向量搜索,完整 AI 功能请使用 PostgreSQL

🚧 待实现

详见 #134

  • 对话持久化 (AIBlock)
  • 情景记忆 (EpisodicMemory)
  • 用户偏好 (UserPreferences)
  • 代理指标 (AgentMetrics)

构建

# 默认模式(无 AI,纯 Go)
go build ./cmd/divinesense

# AI 模式(需要 CGO + sqlite-vec)
CGO_ENABLED=1 go build -tags sqlite_vec ./cmd/divinesense

Full Changelog: https://github.com/hrygo/divinesense/commits/v0.97.0

Resolves #9

v0.96.0 - AI Caching Optimization

07 Feb 14:26
e1a2fc3

Choose a tag to compare

AI 缓存优化 (Issues #91, #92, #93)

  • 语义缓存层 (L2): 基于 Embedding 相似度的两层查找

    • SHA256 精确匹配 + 余弦相似度语义匹配(阈值 0.95)
    • LRU 驱逐策略,默认 TTL 24 小时
  • 工具级结果缓存 (L3): 减少重复数据库查询

    • 按工具类型配置 TTL(schedule_query: 30s, memo_search: 5m)
    • 写操作不缓存,自动过期管理
  • 动态 Token 预算分配: 按意图类型自适应调整

    • 7 种预设配置:memo_search, schedule_create, schedule_query, amazing, geek, evolution, default
    • 环境变量覆盖支持 (DIVINESENSE_BUDGET_*)
    • GEEK/EVOLUTION 模式零预算(由 Claude Code CLI 管理)

性能提升

缓存层 延迟 命中率
L1 路由缓存 ~0ms 70%
L2 语义缓存 ~10ms 30%
L3 工具缓存 ~1ms 40%

配置示例

# 语义缓存
DIVINESENSE_SEMANTIC_CACHE_MAX_ENTRIES=1000
DIVINESENSE_SEMANTIC_CACHE_SIMILARITY_THRESHOLD=0.95

# 工具缓存
DIVINESENSE_TOOL_CACHE_MAX_ENTRIES=100

# 预算覆盖
DIVINESENSE_BUDGET_MEMO_SEARCH_RETRIEVAL=0.70

完整变更: v0.95.0...v0.96.0

v0.95.0

05 Feb 17:53

Choose a tag to compare

What's Changed

Features

  • Unified Block Model (Phase 4): Frontend Block integration (Issue #71)
  • Block API hooks with React Query integration
  • Block status management (pending/streaming/completed/error)
  • Conversation block persistence with PostgreSQL

Fixes

  • UUID namespace collision in cc_runner.go (P0)
  • React Query cache keys unified with blockKeys factory
  • Logo animation selector performance optimization
  • Type-safe error handling with ApiError interface
  • appendUserInput type definition fixed

Technical Details

  • Added web/src/config/chat.ts for centralized chat configuration
  • Added web/src/config/errors.ts for type-safe error handling
  • Updated Block hooks to use consistent cache keys

Full commit history: https://github.com/hrygo/divinesense/commits/main