A multi-tenant RAG + Chatwoot platform where customers interact through Chatwoot, while a RAG backend retrieves relevant knowledge, queries an external LLM, and posts AI-generated responses back to the conversation. Operators manage tenants, users, knowledge documents, and Chatwoot integrations through a centralized dashboard. Each tenant has an isolated knowledge base, Chatwoot configuration, conversations, and vector store, ensuring complete data separation and secure AI-powered customer support.

A multi-tenant RAG + Chatwoot platform. Customers chat through Chatwoot; rag-backend retrieves knowledge, calls an external LLM, and posts replies back. Operators manage tenants, documents, and Chatwoot integration in the dashboard portal.

This monorepo has three parts. Use this README for the big picture; each folder has its own setup guide.

1. Overview
   • Enterprise (tenant)
2. Architecture
3. Repository
4. Getting started
5. Installation by part
   • Part 1 — Chatwoot
   • Part 2 — RAG stack
   • Part 3 — Widget test page
6. Integration
7. Reference

Tenant isolation is enforced in the API (enterpriseId on every row).

An enterprise is one customer organization (tenant) on the platform — for example a company, brand, or support desk. Everything that belongs to that customer is scoped by a single enterpriseId (UUID). One Priyo deployment can host many enterprises; their data never mixes.

What belongs to an enterprise

Typical lifecycle

1. SUPER_ADMIN creates an enterprise in the dashboard (name, slug, optional description).
2. Upload knowledge for that enterprise until documents are READY.
3. ENTERPRISE_ADMIN opens Settings in the dashboard — configure name, description, status.
4. Configure Chatwoot Setting for the same enterpriseId.
5. Register the webhook in Chatwoot using that UUID.
6. End-customer messages on the Chatwoot inbox trigger RAG using only that enterprise’s knowledge base.

Platform (one rag-backend)
├── Enterprise A  →  knowledge A  →  Chatwoot inbox A  →  webhook …/chatwoot/{id-A}
├── Enterprise B  →  knowledge B  →  Chatwoot inbox B  →  webhook …/chatwoot/{id-B}
└── SUPER_ADMIN   →  manages all enterprises (no tenant scope)

Full diagram (also at the top of this README):

Markdown link: docs/priyo-architecture.png

The stack has four layers — Users, Part 1 (Chatwoot), Part 2 (RAG), and External (OpenAI). Each enterprise gets isolated knowledge, webhook URL, and Chatwoot settings.

1. Customer sends a message via the Chatwoot widget → Chatwoot inbox.
2. Chatwoot fires message_created webhook → AI agent (rag-backend :4000).
3. RAG backend embeds the query → searches PostgreSQL/pgvector → calls embedding (:8088) as needed.
4. RAG backend sends context + query to OpenAI → gets an answer.
5. AI agent posts the reply through the Chatwoot API → customer sees it in the widget.

Human agents reply in Chatwoot directly. Operators use the dashboard for enterprises, Settings, knowledge, users, and Chatwoot integration.

Each environment needs its own RAG database, secrets, and webhook URLs.

<repository-root>/
├── chatwoot-setup/      # Part 1
├── rag-backend/         # Part 2a (+ compose: DB, embedding)
├── embedding/           # Part 2b
├── dashboard-portal/    # Part 2c
├── demo-ui/             # Part 3
├── docs/                # priyo-architecture.png
└── README.md

Share production secrets via a password manager or CI — not git.

cd rag-backend
cp .env.example .env
# Edit: JWT_SECRET, OPENAI_API_KEY
docker compose up -d

Wait 1–3 minutes on first boot (embedding model), then:

curl http://127.0.0.1:8088/health
curl http://127.0.0.1:4000/api/v1/internal/live

cd dashboard-portal
cp .env.local.example .env.local
npm install && npm run dev

Open http://localhost:5000 — dashboard-portal/README.md.

Embedding may need 1–3 minutes on first start — rag-backend troubleshooting.

Each part follows the same structure: overview → tech stack → prerequisites → install → docs.

• pgvector — chunk embeddings for similarity search
• Migrations — on API startup (DATABASE_MIGRATIONS_RUN)
• New env — fresh volume, new JWT_SECRET, no shared prod secrets

demo-ui/README.md

Connect Part 1 (Chatwoot) to Part 2 (RAG). Use Part 3 to test as a customer.

• chatwoot-setup/README.md
• rag-backend — Chatwoot Setup
• dashboard-portal — Connect Chatwoot
• demo-ui — Run demo