HomeAbout
About

Among all the places you could compromise, find the one direction that holds

Zhenbei builds enterprise infrastructure: unified identity and Agent IAM (UIAM), a multi-tenant AI agent platform (Uniclaw), private-domain SCRM for WeCom (Uniscrm), an online learning platform (Unilearning), and a hardcore toolkit for developers — legdger, NewTool and pxc.

What Zhenbei is

True North is the direction the compass actually points — not the one the needle drifts to under magnetic interference. We build infrastructure in pursuit of the same thing: among countless places where compromise is possible, find the direction that stays correct, then pin it down.

What we build

Four main business lines — UIAM (unified identity and authentication), Uniclaw (enterprise AI agent platform), Uniscrm (WeCom private-domain SCRM) and Unilearning (online learning platform) — plus a hardcore toolkit built for developers themselves: legdger, an encrypted local-first ledger; NewTool, driven by a Rust algorithm kernel; and pxc, a traffic-capture proxy benchmarked against Proxyman.

What we do not do

We do not sacrifice the data model for demo effect, do not make uploading user data to the cloud the default, and do not re-implement business logic in the UI layer. Each is a hard constraint written into repository conventions, not marketing copy.

How we decide something is done

We look at whether the numbers moved. Every product page has a “code facts” column: endpoint counts, table counts, module counts, migration counts. These are countable in the repository, and they are how we judge progress and quality internally. Adjectives do not count as done.

Facts
Company
真北科技
English name
Zhenbei
Domain
zhenbei.tech
Namespace
tech.zhenbei.*
Location
Shanghai, China
Business lines
UIAM / Uniclaw / Uniscrm / Unilearning
Tools
legdger / NewTool / pxc
Runtime languages
Kotlin · Java · TS · Go · Rust · Swift
Portfolio

Seven product lines, one engineering standard

Four business lines plus three tools. Different languages, different forms, but one shared set of constraints: contract first, boundaries enforced by the compiler or framework, every action auditable.

What Makes Us Different

Four things we are willing to be checked on

These are not values posters — they are practices you can verify in the product and the code.

Boundaries guarded by the compiler, not by documents

Go’s internal wall makes cross-domain imports fail at compile time; jOOQ’s access listener intercepts missing tenant predicates at the query layer. Any constraint that can be hardcoded is not left to a document to remember.

The data boundary is not a configuration option

On the device-side product (legdger), neither the books nor AI inference ever leave the device; in private deployments the models, vectors and business data all stay inside the customer network. That is the product form, not a switch.

AI is a first-class citizen, not a plugin

UIAM gives agents MACHINE principals and tool-level grants; Uniclaw splits the cognitive loop into four independent engines; NewTool exposes its algorithm kernel to AI, humans and the browser alike.

End-to-end delivery, no half-finished handoffs

Migration tooling, legacy data import, rollback plans and runbooks ship with the delivery. Unilearning’s legacy-system migration preserves IDs and original passwords, so business teams do not start over.

What We Are

What kind of company Zhenbei is

The four lines look unrelated — identity, agents, private-domain CRM, online learning. What connects them is not an industry but a way of working.

One engineering standard, four business lines

The four lines differ in language and shape, but share one set of constraints: contract first, boundaries enforced by the compiler or framework, every action traceable.

UIAM's api/biz layering and Unilearning's api/internal layering are the same rule expressed in two languages

AI is a first-class citizen, not a plugin

Every product reserves an identity, credential and read-only interface slot for agents. AI gets structured output, not screenshots.

UIAM's four Agent IAM pillars; Uniclaw's 19 built-in tools; legdger's read-only ledger-cli token

Boundaries are not a matter of discipline

Any constraint that can be baked into the compiler, the framework or the query layer does not go into a document for people to remember.

Tenant filtering is enforced by a jOOQ VisitListener in the query layer; Go’s internal wall makes cross-domain imports fail to compile

No compromise on the data boundary

Sensitive data stays on the user’s own device or inside the customer network by default; the cloud only does what it must. Every business line supports private deployment.

legdger runs AI on-device with local encryption; Uniclaw can self-host models and memory; Unilearning ships as a single container

By The Numbers

Scale is counted, not described

Every number below maps to a module, table or endpoint that actually exists in the code. It is also how we decide whether something is done.

209
HTTP controllers
UIAM 77 + Uniscrm 132
56
UIAM data tables
16 data domains, 82 migration files
25
Uniclaw business modules
including cognition, memory, persona and collaboration
19
built-in agent tools
search / SQL / charts / code execution
8
Unilearning domains
6 microservices folded into 1 binary
6
runtime languages
Kotlin · Java · TS · Go · Rust · Swift
Runtime Territory
Kotlin

UIAM identity kernel · Uniscrm business modules

Java

Uniscrm WeCom server side

TypeScript

Uniclaw agent runtime · console

Go

Unilearning modular monolith · pxc kernel

Rust

NewTool algorithm kernel · CLI · WASM

Swift

legdger on iOS and macOS · pxc macOS shell

Timeline

From one identity system to four business lines and a toolkit

The timeline only records milestones backed by deliverables. We do not write “aspires to be” sentences — only what was actually built.

StartUIAM

It began with one identity system

Zhenbei’s first business line was unified identity. The starting point was concrete: customers ran several business systems, each maintaining its own accounts, with mismatched permission semantics and no audit view to show. We started from the identity kernel, getting OAuth 2.1 / OIDC, tenant isolation and fine-grained authorization right.

Identity kernel and tenant isolation
OAuth 2.1 / OIDC authorization server
RBAC roles plus data permissions on two tracks
Homogeneous MySQL / PostgreSQL dialects
ExpansionUIAM · Agent IAM

The identity boundary moved from people to AI

As agents entered enterprise systems, the existing identity model could not hold them: agents worked with human tokens, and incidents could not be attributed. So machine principals, delegation chains, multi-hop propagation and MCP tool authorization became first-class capabilities of the kernel.

MACHINE principals and service account modelling
RFC 8693 delegation chains with permission intersection
agent_tools registry and tool-level authorization
Agent call quotas and end-to-end audit
FormationFour business lines

One business line became four

Beyond identity, customers raised needs in private-domain operations, online learning and AI applications — all sharing the same underlying judgements: where data should live, and who should guard the boundary. Four business lines took shape, in different languages but under one engineering standard.

Uniscrm WeCom SCRM (16 modules / 132 endpoints)
Uniclaw multi-tenant AI agent platform (25 modules / 19 tools)
Unilearning online learning platform (8 domains / modular monolith)
All four lines sharing the UIAM identity kernel
RebuildUnilearning · Uniscrm

Turning “good enough” into “holds up”

Once products ran in production, the real pressure came from operations and iteration. We did two structural rebuilds: Unilearning folded 6 microservices into 1 modular monolith, and Uniscrm rebuilt customer relationships from “friends inside an employee’s WeChat” into a company asset that can be inherited.

Unilearning: 6 microservices → 1 binary, internal wall enforced by the compiler
Uniscrm: customer assetisation, conversation archiving and channel attribution
Unilearning: legacy-import migration preserving IDs and passwords
Migration proceeds domain by domain, each independently rollback-able
NowToolkit

The toolkit: built for ourselves, used by ourselves, then shipped

Across four business lines in production we accumulated a set of internal tools sharing one trait: kernel separated from shell, with AI treated as a first-class citizen. We polished them into products — running internally first is what made us willing to ship them.

legdger encrypted local-first ledger (iOS / macOS / CLI)
NewTool Rust algorithm kernel toolbox (desktop / CLI / WASM / MCP)
pxc capture proxy and team collaboration (Go kernel + three shells)
Tools and business lines under one engineering standard
Engineering Principles

How we build products

Not methodology slogans — hard constraints written into each repository’s AGENTS.md and enforced on every commit.

Kernel first

Product capability lives in a testable, reusable kernel; the UI is just a shell. Algorithms are written once and shared by people and AI alike.

Local first

Sensitive data stays on the user’s device by default; the cloud only handles what it must — sync and collaboration.

Contract driven

Kernel and shell agree on exactly one contract — a /v1 API, a CLI envelope or an algorithm manifest — so behaviour is identical across platforms.

AI native

Every product reserves an integration point for agents: identity, credentials, a read-only CLI and structured output, so AI can work safely.

Century data model

Once a data model is settled it is never overturned. Load-bearing walls are fixed; everything else is free to iterate quickly.

Observable

From identity audit to captured traffic, every action is traceable, replayable and shareable. Debugging stops relying on word of mouth.

How we work

How we deliver

From the architecture call to production operations, we turn every step into a verifiable artifact.

01
Architecture call

Thirty minutes to map out the current state, constraints and goals, ending with a written problem list.

02
Solution design

Architecture layers, data model and migration path — with the trade-offs stated explicitly.

03
Delivery

Private-deployable, infrastructure as code, identical environments.

04
Long-term evolution

Stable kernel, backward-compatible interfaces, audited and rollback-able changes.

Milestones

A few milestones backed by deliverables

The full evolution lives on the timeline; only the three most recent phases are shown here.

FormationFour business lines
One business line became four

Beyond identity, customers raised needs in private-domain operations, online learning and AI applications — all sharing the same underlying judgements: where data should live, and who should guard the boundary. Four business lines took shape, in different languages but under one engineering standard.

RebuildUnilearning · Uniscrm
Turning “good enough” into “holds up”

Once products ran in production, the real pressure came from operations and iteration. We did two structural rebuilds: Unilearning folded 6 microservices into 1 modular monolith, and Uniscrm rebuilt customer relationships from “friends inside an employee’s WeChat” into a company asset that can be inherited.

NowToolkit
The toolkit: built for ourselves, used by ourselves, then shipped

Across four business lines in production we accumulated a set of internal tools sharing one trait: kernel separated from shell, with AI treated as a first-class citizen. We polished them into products — running internally first is what made us willing to ship them.

Also

Other directions

Historical experience and new exploration — both judged by what we can actually deliver.

Blockchain +Blockchain industry solutions

Deep blockchain experience across industries, from advisory through to delivered systems.

Private-domain growth advisorAdvisory & coaching

Process, product, engineering and management working together as end-to-end advisors for private-domain operations.

Get in touch

Hand the complexity of identity, agents and private domain to one governable kernel

Whether you are replacing an existing IAM, building an agent platform, or trying to make private-domain operations actually work — start with a 30-minute architecture call. We will first judge whether this is the kind of problem we are good at, and say so plainly if it is not.