Leish · ليش؟
Is training really the answer? Leish is an AI performance-diagnosis tool for training consultancies and L&D teams. A manager asks for a course; Leish reads the company's tickets, surveys, interviews and policies, finds the real cause of the performance gap (Gilbert's Behavior Engineering Model), and a person approves the diagnosis before any fix is designed.
This repository is the web app. The AI agents run in n8n (LS 01 to LS 07 workflows, Gemini, Qdrant). The app never calls the AI directly.
Browser ──> Next.js (this app) ──X-Leish-Key──> n8n webhooks ──> LS 01 Intake → LS 02 Investigation → LS 05 Diagnosis + Critic → LS 06 Approval → LS 07 Report
▲ │
└──────────────────────────── reads the "Leish Cases" sheet via LS API · Cases ─────────────────────────┘
Pages
/ what Leish does/new submit a diagnosis request (goes to LS 01)/cases all cases with live status/cases/[id] request, diagnosis, evidence, decision and action plan
Run locally
npm install
cp .env.example .env.local # fill in the n8n webhook URLs and the shared key
npm run dev
Without the n8n variables the app runs in demo mode with sample cases.
Environment variables
| Name | What |
|---|
N8N_INTAKE_URL | Production URL of the web-intake webhook (POST) that starts LS 01 |
N8N_CASES_URL | Production URL of the LS API · Cases webhook (GET) that returns the rows of Leish Cases |
LEISH_API_KEY | Shared secret; n8n checks it with Header Auth, header name X-Leish-Key |
Deploy
Import the repository in Vercel, add the three variables, deploy.
