Java · AI-native · Enterprise

What's available today

Every item states what it gives you.

Foundation

Sign in / sign out / captcha

An authentication loop you can use as is

Sign in with account, email or phone; BCrypt is the only password scheme, no weak-hash option is offered

Users / roles / menus

The main admin path

Three tables tell one story; column comments match the implementation; menus and front-end routes share one source

Button-level permissions

Control down to each button

Permission codes are defined centrally, literals are banned; the build verifies all three places agree

Departments / data scope

People see what they should

By department / self / custom range / all; injected by the data access layer, business code is unaware

Dictionaries / parameter config

Everyday config is not hardcoded

Cached, and the cache is observable rather than a naked query

Operation log / sign-in log

Traceable when something goes wrong

Written asynchronously so the main path is not blocked; correlatable with traceId

File upload

Generic attachment support

Type and size allowlists are enforced on the server, not in the browser

API documentation

Front-end and back-end stop negotiating by word of mouth

OpenAPI is exported at build time from the same source as the permission-code list

Dynamic routes and permission directives

Zero authorization logic in the front end

The front end only shows and hides; it never decides permissions

Scheduled tasks / message consumption

Non-HTTP entry points are observable too

They share the same traceId as HTTP requests

Engineering quality

Architecture guardrails

Conventions enforced by machine

Layering and dependency direction are mandatory; six banned patterns (field injection, controller reaching straight to Mapper, System.out, new Date() and more); no dependency cycles

Contract consistency

Docs never drift from code

Registered endpoint count, permission codes and error-code ranges are compared against source in real time; a mismatch fails the build

Deny-by-default authorization

Authorization is never forgotten

An endpoint without an auth annotation fails at startup; anonymous access returns a fixed error code

End-to-end traceId

Locate a problem in five minutes

HTTP, scheduled tasks, message consumption and thread pools all read the same value

Structured logging

Logs are searchable and aggregatable

traceId and userId are top-level fields; phone numbers and passwords are masked by default

Operation audit

Sensitive actions leave a trail

Permission changes, deletes, exports and sign-ins trace back to person, time, IP and traceId

Query and write quality

It does not do the things that cause incidents

Condition allowlist plus wildcard escaping; a type mismatch throws instead of silently dropping the condition; a stable secondary sort is added when none is given; unified pagination with deep-page protection

Transaction and migration discipline

One source of truth for data

Writes go through one exit; transaction boundaries are held by architecture assertions; the soft-delete column is NOT NULL DEFAULT 0; migrations are append-only

Skeleton generator

A new module in 30 minutes

Generates directories, migration scripts and test skeletons, and registers the module, permission codes and menus; the output ships with tests and passes the build on the first run

Approval instance layer (capability pack)

Outsourced OA essentials, ready now

Flow definition plus fixed-chain nodes, instances and tasks, approval trails with traceId, joint and any signing, rejection; pluggable, and the core has no dependency once it is removed

One command to boot the environment

A new machine is sign-in ready in 10 minutes

Migrations create tables, seed data and the initial administrator; you only need JDK 21 and MySQL 8, no intranet address and no manual table creation

CI and quality gates

Every commit gets a check-up

Build, tests, guard rules, formatting and commit hooks, all in one command

What you get

A sign-in-ready admin console, a reproducible set of engineering-quality metrics, one command to boot the environment, and a generator that produces code which passes the build as is.

By the way: code that fails the architecture check cannot be merged.

What we deliberately don't do

Not doingWhy notIf you need it
Workflow engine and visual designerA ten-year project; today we ship the instance layer onlyCover fixed chains with the approval instance layer; extend complex flows through a capability pack
Multi-tenancyThe core repository does not introduce a tenant dimensionIsolate by department with row-level data scope; tenancy is a later direction
Field-level permissions and a masking frameworkToday only log masking is on by defaultHandle sensitive fields in the serialization layer; a capability pack later
Performance benchmark baselines"Easier to maintain" shows up in conventions and quality, not in benchmark numbersRead the query and write quality section; benchmarks are a later direction
Calling large models inside the frameworkCost, keys and model turnover are all traps, and it means owning the model's qualityWrite code with any AI tool and let this framework's build gates hold the line
Transport / storage encryption and national cryptographyThe discipline of keeping keys external comes firstSolve it at the gateway and database layers; it does not enter the core repository
Multiple databases and domestic-platform adaptationWe accept the "MySQL first" trade-offAdapt through a capability pack when needed; the core stays untouched
Low-code and online form buildersOnce the pitch is "zero code", complex scenarios will breakUse the skeleton generator plus hand-written business code
Built-in business modules (shop / CRM / membership)The core repository holds zero business code; shipping business is exactly what locks users inKeep business in your own repository and extend through capability packs
Container images and orchestration filesThe delivery shape is a jar plus a database; choosing operations tooling for users only widens our promise surfacePackage it to your own standards; the deployment guide covers the non-container path only
(Website) any competitor comparisonOur public story is about ourselves onlyWe publish the dimensions and the methods; you run the numbers
(Website) customer logos and case wallsWe have no authorized endorsements yetLook at engineering quality and build evidence; we will add cases when they are real