Project index
Verified liveWhite-label AIProduction

Atlee Chat 

A tenant-configurable RAG product with multi-source ingestion, grounded chat, embeddable delivery, operational inboxes, and billing.

Open live product Open full system flow
Architecture knowledge graph16 connected nodes
System coreAtlee ChatWhite-label AI
01 / ManageThe tenant console owns the assistant lifecycle
02 / ExtractSource-specific services remove format differences
03 / StoreObjects, vectors, and operational state stay separate
04 / ReasonA graph evaluates context before answering
05 / ObserveConversations return as inspectable product data
06 / DistributeHosted and embedded experiences share one policy layer
Next.js
Express
LangChain
LangGraph
Pinecone
MySQL
AWS S3
Stripe
Docker
Architecture stage Technology dependencyMove pointer to inspect depth

System brief

A white-label knowledge assistant for service teams

Engineering focus
Full-stack AI product delivery
Domain
White-label AI
Delivery window
Production
System state
Verified live
6+supported knowledge-source types
4separate deployable services
1tenant-controlled assistant lifecycle

The engineering move

Complex behavior, made operational.

Constraint

The product problem

Service organizations need more than a chat window. They need controlled knowledge ingestion, grounded answers, conversation review, brand and access settings, lead capture, source refresh, and a safe way to distribute the assistant across web properties.

System

The architecture decision

A Next.js management surface coordinates an Express application API and a separate generic AI service. Source-specific controllers handle documents, websites, sitemaps, YouTube, audio, and text before vectorization. LangGraph utilities analyze questions, choose retrieval, grade context, and generate a response.

Outcome

The operating result

A team can move through the full assistant lifecycle from account and subscription to source training, QA refinement, deployment, inbox review, feedback, and ongoing configuration.

Delivered capability

What the system actually does.

01

Tenant accounts, authentication, roles, and subscriptions

02

Bot creation, update, access, lead, and appearance settings

03

Document, web, sitemap, YouTube, audio, and text ingestion

04

Pinecone retrieval with question and document grading

05

Multi-provider generation through LangChain and LangGraph

06

Conversation history, review, likes, and QA refinement

07

Hosted chat, generic API, and bundled website widget

08

Stripe, S3, MySQL, Docker Compose, and nginx operations

Full system flow

The Mermaid diagram, rendered.

The interactive graph above is the executive view. This is the detailed service, data, control, and delivery path reviewed from source.
Mermaid / system architecture120%

Rendering the reviewed system flow

Drag the horizontal viewport at higher zoom levelsView .mmd source

Production architecture

A concrete path through the system.

Each stage below comes from the reviewed source tree, routes, services, infrastructure, and deployment evidence.
  1. 01
    Manage

    The tenant console owns the assistant lifecycle

    Accounts, plans, bots, sources, prompt documentation, appearance, access, leads, inbox, and analytics share one product surface.

    Next.jsReduxNextAuth
  2. 02
    Extract

    Source-specific services remove format differences

    Files, URLs, sitemaps, YouTube, audio, and text each enter through dedicated routes and extraction controllers.

    CheerioPDFAudio processing
  3. 03
    Store

    Objects, vectors, and operational state stay separate

    S3 or Google Cloud stores source objects, Pinecone stores semantic chunks, and MySQL retains product and conversation state.

    AWS S3PineconeMySQL
  4. 04
    Reason

    A graph evaluates context before answering

    Question analysis, retrieval decisions, document grading, response generation, and no-data handling form an explicit path.

    LangGraphLangChainModel APIs
  5. 05
    Observe

    Conversations return as inspectable product data

    History, reviews, likes, bulk updates, inbox state, and refinement workflows let operators audit real interactions.

    Conversation APIFeedbackWinston
  6. 06
    Distribute

    Hosted and embedded experiences share one policy layer

    The client, main API, generic service, and bundle build deploy behind nginx with a consistent bot configuration.

    Embed bundleDockernginx

Engineering judgment

The decisions behind the delivery.

Source lifecycle matters

Create, refresh, retrieve, and delete paths keep object storage and vector state synchronized.

Operators can inspect failure

No-data paths, conversation history, reviews, and QA refinement expose behavior that would otherwise stay hidden.

Deployment is part of the product

The hosted interface and embeddable widget share tenant policy rather than diverging into separate chatbot implementations.

Technical constellation

01Next.js02Express03LangChain04LangGraph05Pinecone06MySQL07AWS S308Stripe09Docker
Atlee Chat production sign-in screen
Deployment evidence

This is running software.

Live product and source reviewed on 2026-07-16. The public link is provided as evidence, while protected product areas correctly remain behind authentication.

Open Atlee Chat

Continue through the project archive

Next caseDilpasand