For AI agents: the complete documentation index is available at /en/llms.txt, the full documentation bundle is available at /en/llms-full.txt, and this page is available as Markdown at /en/guide/what-is-yearning.md.
  • English
  • Overview & What Changed in Yearning v4

    Public Beta Notice

    Yearning v4.0 is currently in Public Beta. Core workflows, AI rules, and migration tooling are actively evolving. Please evaluate in staging or test environments first, maintain full backups before production cutover, and report feedback via GitHub Issues.

    Yearning v4 is a ground-up rebuild of the open-source Yearning platform as an AI Agentic SQL Review & Database Governance Platform.

    Built as a single-binary Go modular monolith paired with a React 19 + TypeScript + shadcn/ui web workspace and a PostgreSQL metadata plane, Yearning v4 retires legacy RPC services and MySQL metadata storage while putting an autonomous metadata-probing AI Agentic SQL Review Engine and a Database AI Assistant at the heart of frozen multi-stage governance, in-process gh-ost online DDL, binlog-free cross-engine rollback, governed read-only queries, and Top-SQL performance diagnostics.

    Yearning v4 Workbench: personal queues, 30-day execution metrics, 6-component health dots, and the scoped Database AI Assistant
    Yearning v4 Workbench: personal queues, 30-day execution metrics, 6-component health dots, and the scoped Database AI Assistant

    Product Positioning

    Yearning v4 is an AI-powered database change governance platform, never an unbounded chat shell:

    • AI analyzes risk and synthesizes evidence: combining dialect AST parsing, dual-fingerprint deduplication, 5 read-only metadata probes, custom prompt skills, and enterprise DBA knowledge rules.
    • Structured governance controls approvals and execution: AI never approves orders, never directly executes SQL, and never bypasses the Submission Gate.
    • Privacy by architecture: cloud LLM traffic (default track) strips literals and comments and anonymizes every identifier in memory before dispatch; self-hosted internal models (san track) run on isolated RFC1918 LAN contracts with zero cloud fallback leakage.

    Yearning v4 vs. v3 Architecture Comparison

    CapabilityYearning v3 (Legacy)Yearning v4 (Rebuilt)
    Control PlaneLegacy RPC couplingSingle-binary Go modular monolith, zero RPC, stateless multi-replica Kubernetes ready
    Frontend WorkspaceVue (gemini-next)React 19 + TypeScript + Vite + Tailwind CSS v4 + shadcn/ui
    Metadata StoreMySQL + AutoMigratePostgreSQL 14–18 + explicit versioned migrations + 3-role least privilege
    Target EnginesPrimarily MySQLMySQL 5.7/8.0, PostgreSQL 14–18, TiDB, OceanBase, PolarDB
    SQL Review EngineStatic rulesDeterministic rules + Dual-Track AI Precheck (up to 100k statements / 32 MiB, 5 metadata tools)
    Workflow & FreezingMutable routing & AutoTaskFrozen at submission (1..N stages, 1..10 steps); no admin override; AutoTask retired
    Online Schema ChangeExternal binariesIn-process Go gh-ost (18 typed parameters, live progress); pt-osc retired
    Automatic RollbackMySQL Binlog scrapingBinlog/WAL-free AST inverse + in-transaction before-images across MySQL & PostgreSQL
    Query & DiagnosticsBasic query & maskingTime-bounded Query Grants, read-only AST, PII masking, SSE lock, 6-chart Top-SQL diagnostics
    AI Agent IntegrationNoneFull OpenAPI v4 + SSE event stream + zero-dependency skills/yearning Agent Skill