Releases: MindAgency/mind-agency
Releases · MindAgency/mind-agency
v0.9.0: Validation-Optimization Complete
v0.9.0 发布说明
核心改进
- ✅ Workflow 依赖步骤调度修复
- ✅ API Key 配置修复(mimo-v2.5)
- ✅ Jest Worker 崩溃修复
- ✅ 事件循环阻塞修复
- ✅ Agent 指令遵从性修复
- ✅ RAG 集成修复
- ✅ Dashboard Web UI 新增
- ✅ 146 个单元测试
- ✅ API 文档
- ✅ Metrics 系统
- ✅ 清理 39 个僵尸 Agent/Group
测试结果
- 引导式测试:10 个角色,平均 4.5/5
- 自主测试:19 个角色,平均 2.5-3/5
- 依赖步骤调度:✅ 通过
核心功能验证
- Workflow 引擎 DAG 执行
- 真实 AI(mimo-v2.5)集成
- SimulatedStepExecutor 降级
- Agent 自动响应
- Token 经济系统
- Event Bus
下载
- Mind-Agency-Setup-0.9.0.exe (NSIS 安装包)
Full Changelog: v0.8.0...v0.9.0
v0.8.0: Performance + Code Simplification
What's New
⚡ Performance
- Consolidated 4 polling intervals → 1 (15s)
- Consolidated 4 WebSocket connections → 1
- LogoCanvas: canvas animation → static SVG (eliminates 3 rAF loops)
- ChatPanel: React.memo on MdText, Think, Tool, Result, Err
- Notion theme: removed universal selector with !important
- Google Fonts: removed render-blocking link
🧹 Code Simplification
- Removed dead code (orphaned MdText duplicate)
- Added missing React import
- Fixed missing closing parens
- Extracted patchLastMsg helper (DRY)
- Consolidated scroll tracking
- Made refresh return Promise
- Formatted load() with reduce()
🐛 Bug Fixes
- Dynamic Node.js path detection (was hardcoded)
- Fixed i18n key 'pending'
- Updated theme whitelist (4 → 8 themes)
- Standardized Solarized Light colors
v0.7.0: Desktop App + Modular AgentProxy
What's New
🖥️ Desktop App (Electron)
- Standalone Next.js server extracted to writable directory
- System Node.js for ES module support
- Custom icon and window frame
🧩 Modular AgentProxy
- Split into 8 focused modules
- Typed error hierarchy
🧹 Repository Cleanup
- Removed 100+ runtime/output files
- Clean .gitignore rules
📚 README
- Live demo examples, feature table, architecture diagram
🔧 Fixes
- Hydration mismatch in TitleBar
- NaN in StatCard
- Logo sprite sheet restored
v0.6.0: Workflow Automation & Economy
Mind Agency v0.6.0
New Features
- Workflow Automation: EventBus instant triggering, auto-retry, timeout monitoring
- Token Economy: Deposit/transfer/reward/budget per agent
- Task Claiming: Post/claim/select workflow with rewards
- Learning System: AI evaluation, reviewer scoring, learning records
- Orchestrate API: Goal decomposition → workflow creation → trigger
- Transformer-style Architecture Diagram: NLP paper-style workflow visualization
Bug Fixes
- Notification path consistency (AGENTS_DIR)
- Engine singleton (shared instance)
- ESM require() issues
- signal-collector emailCheck bug
- _group type safety (removed 7x 'as any')
- getStats() retry count bug
Architecture Improvements
- EventBus replaces file communication for instant agent triggering
- Priority bucket queue O(1) insertion/dequeue
- Adaptive heartbeat (2min active / 5min idle)
- Auto-save agent output to files
Frontend
- Dashboard: token leaderboard + open tasks
- Group page: task board + orchestrate button
- Agent page: token balance + transaction history
- Learning page: scores, approval rates, feedback
- Workflow: Transformer-style architecture diagram
Download
- 📦 Mind Agency.exe (~795 MB) - No Node.js required
v0.5.1 - Event-driven Scheduler + Global Dispatch
v0.5.1 Release Notes
🎯 New Features
- Event-driven Scheduler: Signal collector replaces 30s polling with event-driven dispatch
- Global Scheduler: Priority queue + agent state tracking + capability matching
- Failure Alchemy: Learning framework for task failures (created, not yet integrated)
🐛 Bug Fixes
- group_send: Invalidate groupChat cache after sending
- email_send: Invalidate emails cache for sender and recipient
- MCP memory tool: Proper cache invalidation on write/delete
📊 Performance
- Fallback polling reduced from 30s to 5min (event-driven primary)
- Signal queue with priority-based dispatch
- Agent state tracking prevents duplicate processing
🏗️ Architecture
- signal-collector.ts: Centralized signal queue with priority
- global-scheduler.ts: Agent state machine (idle/busy/blocked)
- scheduler.ts: Event-driven with fallback polling
📦 Files Added
- src/lib/signal-collector.ts
- src/lib/global-scheduler.ts
- src/lib/failure-alchemy.ts
v0.5.0 - Unified Cache Layer + 71 Bug Fixes
v0.5.0 Release Notes
🎯 Major Features
- Unified Cache Layer (cache.ts): Centralized cache management with LRU eviction, region isolation, and TTL support
- Event-driven Workflow Triggers: Workflow can now be triggered by EventBus events
- Event-driven waitForCompletion: Workflow bridge uses events instead of polling
🐛 Bug Fixes (71 issues)
- Consensus: Fixed empty handlers, permission bypass, cache invalidation
- Caching: Fixed cache bypass in MCP tools, memory leaks, TTL mismatches
- Validation: Added path traversal protection, input validation
- Performance: Removed embed double-call, aligned stateCache TTL with polling
- Concurrency: Fixed session race condition, autoRespond queue serialization
📊 Performance Improvements
- stateCache TTL: 5s → 30s (aligned with polling)
- searchCache: MD5 hash keys (reduced memory)
- baseOptions: Proper invalidation on memory changes
- groupChat: Cache invalidation on new messages
🏗️ Architecture
- MCP tools now invalidate caches after writes
- Unified cache invalidation via agentCache.invalidateAgent()
- LRU eviction with 80% threshold for batch cleanup
📦 Dependencies
- Next.js 15.5.18
- TypeScript strict mode
v0.4.0 — Multi-Agent Orchestration Overhaul
v0.4.0 Changes
Core
- Task-report-centric workflow model (agent reports via MCP task tool)
- Dynamic CRUD on running workflows (add/delete/modify steps)
- Auto-review steps (reviewer field)
- Cycle detection in DAG
- Provider abstraction (Claude + Codex)
- Multi-round adversarial review
- Consensus timeout (24h auto-expiry)
Security & Reliability
- CSRF protection, API key hiding, path traversal prevention
- Atomic file writes, audit log race condition fix
- Graceful shutdown, workflow cancellation
UI/UX
- Settings page: clean list layout
- Group page: workflow editor + run history
- Agent page: task panel
- DAG visualization: topological sort with dependency lines
- Chat: file operation diff display
- 5 themes with proper color tokens
- Auto-update via GitHub Releases
- Installer (Inno Setup)
Infrastructure
- EventBus in-process Pub/Sub
- SimHash semantic memory search
- Priority queue for auto-respond
- Workflow triggers (file change, schedule)
Mind Agency v0.3.0
From Agent to Agency
一个 AI 做不了的事,一群 AI 可以。
✨ Features
- Multi-Agent Collaboration — Alice writes, Bob reviews, Charlie tests
- Consensus Voting — AND/OR/threshold logic with adversarial review
- DAG Workflow Engine — YAML pipelines with human approval gates
- Agent Memory — Short-term, long-term, and entity memory with search
- Crash Recovery — Checkpoint-based recovery, never lose progress
- Audit Logging — Every action tracked, every decision auditable
- Multi-Theme — Notion, Minimal White, Warm Wood, Deep Space, Nord
📦 Download
- Windows exe:
Mind-Agency-Setup-0.3.0.exe— double-click to install - Source: Clone and run
npm install && npm run dev
🔧 Requirements
- Node.js >= 18 (for source build)
- API Key: Claude / DeepSeek / GPT-4o
📄 License
Apache 2.0