AQUIPLEX Platform v2

AQUIPLEX = the platform. AQUA = the only AI.

Architecture

aquiplex-platform/
├── index.js                Platform server (CommonJS, Express 5, EJS)
│                           auth · tools directory · bundles · blogs · billing
├── aqua/                   AQUA engine (ESM subpackage) — mounted at /api/aqua
│   ├── router.js           Single mount point (chat, stream, memory,
│   │                       conversations, project intelligence, upload)
│   └── src/                core · intelligence · memory · orchestrator ·
│                           project · providers · prompts · routes · upload
├── aqua-frontend/          AQUA app (React 19 + TS + Vite) — served at /aqua
├── models/                 User, Tool, Bundle, Payment, Transaction,
│                           BillingLog, WebhookLog
├── routes/billing/         Razorpay REST routes (webhook handled pre-json)
├── services/               ai.client (internal LLM util), execution,
│                           billing, credits
├── middleware/usage/       Credit metering (usageGuard / checkUsage)
├── utils/                  startup checks, logger, credits, validators
├── views/ + public/        Platform pages (EJS)

Request flow

• GET /aqua → requireLogin → React SPA (aqua-frontend/dist, base /aqua/)
• POST /api/aqua/chat, /chat/stream → requireLogin → usageGuard (chat_message) → AQUA engine (classify → orchestrate → plan → intelligence → memory → project retrieve → web search (multi-provider, cached, orchestrator-gated) → prompt → provider router w/ fallback → verify → persist)
• POST /api/aqua/upload → usageGuard (chat_with_file) → universal upload (ZIP / TAR / TAR.GZ / documents / media) → workspace intelligence dashboard
• Conversations + memory scoped per platform user via req.aquaUserId (session identity injected at the mount).
• Legacy AI URLs 301 → /aqua: /chatbot, /aqua-ai, /aqua-project-engine, /workspace, /workspace/*.

Setup

npm install                 # also installs aqua/ (postinstall)
npm run build:aqua          # builds the AQUA SPA into aqua-frontend/dist
npm start

Env (.env): MONGO_URI, SESSION_SECRET, RAZORPAY_KEY_ID, RAZORPAY_KEY_SECRET, RAZORPAY_WEBHOOK_SECRET, GOOGLE_CLIENT_ID, GOOGLE_CLIENT_SECRET, plus AQUA provider keys: GROQ_API_KEY, GEMINI_API_KEY, OPENROUTER_API_KEY.

Web search keys (multi-key, any subset): SERPER_API_KEY_1..N, TAVILY_API_KEY_1..N (bare SERPER_API_KEY/TAVILY_API_KEY also honored). Zero keys = search dormant, chat unchanged. Optional tuning: SEARCH_TIMEOUT, SEARCH_MAX_RESULTS, SEARCH_PROVIDER_PRIORITY, SEARCH_CACHE_TTL, SEARCH_ENABLE_CACHE, SEARCH_RETRY_LIMIT — see aqua/AQUA_SEARCH_ARCHITECTURE.md.

Tests

npm run test:aqua                          # edit + upload + identity + search
cd aqua && node --test src/**/tests/*.test.js

Removed in the v2 restructure

Legacy chatbot (chatbot.ejs, /chat, aqua.orchestrator, ai chat modes, History model, suggest-prompts, multi-generate), legacy memory system (memory.service, Memory model, /memory pages), Workspace Builder (workspace views/routes/service, generation engine: 50+ engine/* files, brain, assembly, composer, compiler, ingestion, manifest, editing, learning, hybrid-engine, templates, ui-packs), builder-engine/, unused Replit client/ + server/ (React scaffold), file.session/file.parser legacy uploads, socket.io, dead deps (react/vite/radix/drizzle/pg/cashfree/socket.io/ws/pdf-parse/mammoth at platform level), stub command.service, unrouted views (lab, every, tool, components), orphan CSS/JS.

Kept as internal utility (not user-facing AI): services/ai.client.js — powers tool insights, bundle generation/execution.