Local · AI-native IDE · Open source (MIT)

Tame AI inside a DAG.

VL-Code is the local, AI-native IDE that turns plain English into a full-stack app — generate, compile, preview, ship. Creativity lives in the nodes; determinism lives in the edges. You stop babysitting AI and start orchestrating it.

An early preview for friends — videos & articles updated over time.
Full walkthrough

Describe a project → watch it build itself

One prompt kicks off an AI workflow that generates a complete VL app — pages, components, services, data — then lints, compiles, and previews it. This run builds a course-enrollment platform end to end.

Product demo · describe → generate → code & visual → preview · ~11 min 🎧 Chinese narration English version coming soon
÷10
tokens used
×10
build speed
10×
project scale
~100KB
the whole language
1300
1 line VL ≈ 200–400 LOC
0
vendor lock-in
Why VL-Code

A language built for AI — not the other way around

Everyone is making AI write Python & JavaScript — languages built for humans. VL takes the opposite bet: a deterministic, low-entropy representation an LLM can generate reliably and a machine can verify.

01 · AI-NATIVE

Not AI-assisted. AI-native.

VL = Language + IDE + AI, designed together from day one. Low-code hides complexity; VisualLogic structures it.

02 · POINT → LINE

AI is brilliant at a point. It can't hold a line.

So VL-Code puts AI inside a DAG — creativity in the nodes, determinism in the edges. Reproducible, rerunnable, auditable runs.

03 · THE MOAT

VL is AI's Intermediate Representation.

Intent → VL (IR) → executable React/Node. Whoever controls the representation layer controls the ceiling. Representation efficiency > model size.

04 · ONE DOCUMENT

One ~100 KB doc. Infinite apps.

The whole VL syntax fits in one file — “The Bible.” No React/Node/TS/Postgres/Tailwind/Prisma sprawl. One source of truth.

05 · VERIFIABLE

From “looks right” to verified.

Every DAG node can lint / compile / preview / test / human-gate. AI debugs explicit structure, not loose text — debugging becomes structural.

06 · LOCAL-FIRST

Local-first. Cloud optional. 0 lock-in.

Run the IDE locally — local read/write + AI, the smoothest path. Deploy to managed cloud, or export React+Node and self-host.

The IDE is a projection of the language

You reason in pictures — code ↔ graph, 1:1

Architecture becomes a Map, process becomes a DAG, data becomes an ER diagram, UI becomes a canvas. Every view is two-way bound to the same VL source.

Flow / DAG mode

Flow — the build as a verifiable workflow

generate → check → fix → preview, every node inspectable and rerunnable

Map mode

Map — architecture at a glance

the whole app as a navigable component graph

Visual mode

Visual — design the UI directly

edit the canvas; the VL source updates in lockstep

Code mode

Code — the readable VL source

small, single-responsibility files under ~100 lines each

Data mode

Data — model it as an ER diagram

tables, relations and virtual data, visually

80-second intro

Short on time? Start here

A quick English overview of the idea behind VL-Code.

VL-Code — quick intro · English · ~80 sec
Resources & reading

Go deeper

The concept brief, the code, and the community. More articles & demos are added here over time.

Build a full-stack app from one sentence.

Describe it in plain English. Watch the DAG generate, verify, and preview it — all on your machine.