Your Yujin Forge subscription is live.
Your subscription is live. One click and you are in -- nothing to type.
From chat to professional software.
Start without writing a line of code: chat what you want and Forge builds it. But under the hood it has everything a dev needs to take complex software to a professional level. And you manage the development on your laptop from your phone.
USD 10 / month after trial · cancel anytime · BYOK (you pay your LLM provider directly)
Real software, shipped.
Two systems designed, built and deployed with Yujin Forge -- from the chat panel to production. Not demos: software in daily use.
Medical order approval
A platform that downloads, validates, approves with human judgment and files medical-service orders across three disconnected portals -- four coordinated RPA bots, a state machine and a custom web app. It ships with Yujin Forge embedded as a resident AI assistant: it reads the operation, tunes the configuration and diagnoses issues by voice or chat. The client owns the software, with no per-bot licensing.
How we built it → Email -- agentic AIYuemail
An email client where an AI agent reads, triages, drafts and acts on your inbox by voice and chat -- the same NAC3 agentic layer Forge generates, applied to everyday mail.
See it live →記 Documentacion viva
Que hace Forge y como se comporta -- todo verificable
No te lo decimos en bullets de marketing. Estos dos documentos son el manual operativo + la matriz de comportamiento real. Cada capability con su archivo + cada escenario con su SSE log + diff filesystem + tokens consumidos. Generados de la corrida 2026-06-06.
全Funcionalidad total -- Desktop →
Inventario exhaustivo de las capacidades de Forge: 56 subcomandos CLI, 3 modos UI, 119 verbos NAC-3 internos, 10 providers LLM, 10 idiomas, 10 formatos de documento, vault de 27 slots, doctrines server-side, brain tier routing, QA harness, MCP bridge, license model.
Funcionalidad total -- Mobile →
Companion movil en dos formas: la PWA (web app, sin tienda) para chat de texto, tablero vivo (grafo / workflows / pendientes) y control remoto del escritorio -- sin microfono; y la app nativa (APK) que agrega voz on-device con el reconocedor del sistema operativo. Pairing por QR + 6 digitos. Las claves quedan en el vault del escritorio -- nunca viajan al telefono.
験 Las 17 pruebas a las que sometimos a Forge
Submitimos Forge a 17 tipos de pedidos distintos. Cada card abre una pagina con el pedido literal, el comportamiento esperado, el comportamiento real (tokens, modelo, tier, diff), la respuesta del chat y el gap clasificado por severidad.
Corrida 2026-06-06: 17 escenarios + 6 re-runs validatorios. Dos fixes shipped en el mismo dia: alpha.59z.157 (pilot setup heuristic) cerro 4/5 gaps P1; alpha.59z.158 (SQ 9 verify factual claims) cerro el ultimo. Resultado final: 14 OK, 3 P2, 0 P1, 0 P0 (como llegamos hasta aca).
What Forge does for you.
You install one command-line tool, paste a Claude or OpenAI or Gemini key once, and Forge gives you a chat panel in your browser that can read your repo, your specs (PDF, DOCX, MD, Figma, etc), generate NAC3-compliant React components, run the test suite, and ship the app -- guided by voice if you want.
Every app Forge scaffolds is NAC3-compliant by default, which means it is also addressable by AI agents, voice runners, screen readers and RPA bots on day one. No extra annotation pass. No second framework.
Pricing
- CLI + browser chat panel (3 modes: globito / mini / full pizarra)
- Voice loop -- STT + TTS through Whisper / Google Cloud / ElevenLabs
- BYOK multi-provider routing (Anthropic / OpenAI / Google AI / DeepSeek)
- Paired mobile companion -- PWA (text + live board + remote control) and a native Android app that adds on-device voice
- 10-format document reader (PDF, DOCX, XLSX, MD, HTML, EPUB, CSV, RTF, TXT)
- NAC3 scaffolding + migration auditor + auto-generated e2e tests
- Updates + telemetry-free
BYOK: the subscription pays for the tooling. The LLM calls are made with your own key and billed by your LLM provider (Anthropic, OpenAI, Google). Forge never sees, persists, or proxies your key through its servers in any logged form.
Download the desktop app
The Windows installer bundles everything -- no Node, no terminal. Double-click and the chat panel opens.
Prefer a package manager? winget install Yujin.Forge
Developer, or on Linux? Install via npm (Option B below).
Option A -- Windows installer (recommended: no Node, no terminal)
The fastest path if you are not a developer. Five steps, all clicks -- no commands to type.
Click Download for Windows above. You get YujinForge-Setup.exe (~40 MB). It already bundles Node and Forge -- there is nothing else to install.
If Windows shows "Windows protected your PC" (the installer is not code-signed yet), click "More info" then "Run anyway".
It installs into your own user folder (%LOCALAPPDATA%\YujinForge) -- no administrator password. It adds a "Yujin Forge" shortcut to the Start Menu and the Desktop.
Launch it from the Start Menu or Desktop. It starts in the background (no terminal window) and opens the chat panel in your browser at localhost:4847.
On first run the in-panel wizard asks for one thing: your LLM key (Anthropic / OpenAI / Google AI / DeepSeek) or your Claude plan login. Paste it once and start chatting. Already bought? Click "Open Forge" on the Polar success page -- it activates itself, nothing to paste.
Option B -- via npm (developers)
For developers, or on Linux. Needs Node 18.17+ and a terminal. Works on any OS where Node runs.
Node 18.17+ required. Installs the yf command globally.
npm install -g @nac3/forge-cli@latest
yf doctor verifies Node, license, vault, voice providers, brain provider, manuals.
yf doctor
Pick your brain provider (Anthropic / OpenAI / Google AI / DeepSeek) and your voice providers. Keys encrypt locally in the vault and never leave your machine -- never sent to our servers.
yf keys setup
A browser panel on localhost:4847 -- text + voice + project reader.
yf chat
Forge scans the repo and lists exactly what to add to make it NAC3-driveable.
yf migrate audit ./my-react-app
Forge on your phone -- two ways
Two options. The PWA is a web app -- no store, no install: text chat, the live board (graph / workflows / pendings) and remote control of your desktop. It has no microphone. The native app (APK) adds hands-free voice that runs on the phone itself.
PWA -- web app (no store, no voice)
The fastest way to mirror and drive your desktop from your phone. Text + board + remote control. No microphone (browser dictation does not work in an installed phone web app) -- for voice, use the APK below.
Go to this address on your phone (Chrome or Safari):
https://yujin-hito4.yujinapp.workers.dev/
iOS Safari: Share -> "Add to Home Screen". Android Chrome: menu -> "Install app". It then opens like a native app.
On the desktop panel: Settings -> Pair phone (shows a QR + code). On the phone: scan the QR (Android) or paste the code (iPhone). Your phone then mirrors the graph, workflows and pendings -- and can drive the session remotely.
Native app (APK) -- adds on-device voice
Android. Everything the PWA does, plus voice: the mic uses the phone's own speech recognizer -- free, keyless, nothing leaves the device. (iOS native build on request.)
On the phone (signed in to GitHub), open the latest release and download forge-mobile.apk:
github.com/yujinapp/yujin-forge/releases/tag/mobile-latest
Android asks to allow installing from your browser: Settings -> Apps -> Install unknown apps -> allow, then open the downloaded file. (Not code-signed yet, so Android warns the first time.)
Open the app, pair it with the desktop (same QR / code as the PWA), and the mic button works -- press to talk, release to send. Optional: load a premium STT/TTS key on the phone for higher-quality voice; it stays in the phone's secure storage and goes straight to the provider.
What you get inside.
声Voice loop
Hands-free dev: ask Forge to add a route, refactor a component, or run the suite. STT in 18 languages, TTS through Google Cloud or ElevenLabs.
脳Multi-brain
One yf, four providers. Switch from Claude Opus to Gemini Flash mid-session. Cost-aware: see token spend before you spend.
読10-format reader
Drop a PDF spec, DOCX, XLSX, MD, HTML, EPUB, CSV, RTF or TXT and Forge ingests it into the chat context.
移Migration auditor
Point Forge at an existing React or HTML app and get a step-by-step migration plan to NAC3 compliance, with diff previews.
試Auto e2e tests
Forge emits Playwright + Cypress + Vitest + WebdriverIO suites from your NAC3 manifest. 4.7x more cases than hand-written, average.
携Mobile companion
Pair your phone with a six-digit code -- the PWA mirrors and drives your desktop session. The native app adds hands-free voice on the phone itself.
船Ship
Built-in deploy gate (validate + license + tests + build) and autonomous publish to npm + Docker registries.
鎖Lock-in zero
Every app Forge scaffolds is plain React + NAC3 attributes. Apache-2.0 runtime. Eject anytime; nothing to remove.
Receipts, not promises.
NAC3, the contract Forge generates against, has been benchmarked end-to-end. Numbers are reproducible from the published bundle (~ USD 15 / 1h 23m). Read the benchmark · Read the Conduit case study.
FAQ
- What does the $10 / month pay for?
- The tooling: CLI, chat panel, voice loop, document reader, multi-provider router, auto-test generator, mobile pairing, deploy gate, updates. The LLM calls themselves are billed by your provider (Anthropic, OpenAI, Google) on your own key.
- What does BYOK mean in practice?
- You paste your LLM API key once in
yf keys setup. The key is encrypted on disk in the local vault and never leaves your machine -- it is never sent to our servers. Every call carries your key as a header that we forward straight to the upstream provider. We never log it. - Can I cancel during the trial?
- Yes. You will not be charged. After the 7-day trial, billing starts at USD 10 / month and you can cancel anytime from the Polar customer portal -- no email required.
- Do I have to use NAC3?
- Forge scaffolds NAC3-compliant apps by default, but you can use it on any React project. The NAC3 surface is opt-in via five HTML attributes; if you do not annotate anything, you have a normal React app.
- What platforms does it run on?
- The desktop installer is for Windows (no Node, no terminal). The CLI is pure Node.js and runs anywhere Node 18.17+ runs, so on Linux you install it via npm (Option B above).
- Is there a mobile app?
- Yes -- a paired companion, two ways. A PWA (web app, no store): text chat, the live board (graph / workflows / pendings) and remote control of your desktop -- no microphone. And a native Android app (APK) that adds hands-free voice on the phone itself (its own OS recognizer; keys stay on the device). Your desktop keys never travel to the phone. iOS native build on request.
- How do I get help?
- Hit the Discord or write to contact@yujin.app.
Already purchased?
Five commands. The invoice email from Polar is the only thing you need -- use that same email in step 2.
npm install -g @nac3/forge-cli@latest
yf license activate --user-handle <your-email@example.com>
yf doctor
yf keys setup
yf chat
Need help? contact@yujin.app or the Discord.
Try it for a week. Decide then.
Seven days free, your own LLM key, your own machine. If Forge does not save you ten times its cost in the first week, cancel and we part friends.
Start 7-day free trial →