Build with e-bot

e-botis a chatbot backend that plugs an e-commerce store's social media pages into an AI assistant. When a customer messages your Facebook page (Instagram support is on the way), e-bot answers product, pricing, policy, and delivery questions automatically — and can take an order end-to-end inside Messenger.

It runs on FastAPI, PostgreSQL with pgvector, Redis, Celery, and OpenAI's tool-use API. Visual search is powered by DINOv2 — when a customer sends a product photo, the bot finds the matching item by image similarity, not by keyword guesswork.

Every customer message follows the same five-step pipeline:

1. 1

   Message arrives

   A customer sends text, an image, or both to your Facebook page. Facebook forwards the event to e-bot's webhook.

   0
2. 2

   Background queue

   The webhook hands the event to a Celery worker so the request returns instantly — no Facebook timeouts.

   1
3. 3

   Search or reason

   If the customer sent an image, DINOv2 finds the matching product via pgvector similarity search. If it's text, the LLM decides which tool to call (search_products, list_products, get_delivery_zones, place_order, …).

   2
4. 4

   Compose the reply

   The LLM receives the tool result plus the conversation history and writes a reply in the customer's own language.

   3
5. 5

   Send back

   e-bot sends the reply through the Messenger Send API. The whole loop usually takes under three seconds.

   4

Onboarding takes about 15 minutes if your product catalog is ready. Each step below corresponds to a screen in the e-bot dashboard.

• Image search — send a photo, get the matching product with price, sizes, and stock.
• Text search— ask “black winter t-shirt er price koto?” and get a hit from the catalog.
• Browse — ask to see all products or filter by size or price range.
• Store info — opening hours, branch locations, return policy.
• Delivery — ask about zones and charges before ordering.
• Place an order — a four-step conversational flow inside Messenger: confirm intent → address → delivery zone → final confirmation.
• Follow-up— “XL ache?” or “return kora jabe?” after a product is identified are answered from a 5-minute Redis cache, no repeat database hit.

e-bot detects whether the customer wrote in English, Bangla (বাংলা), or Banglish(Bengali meaning written in English letters — “ei ta koto?”).

Replies are always in the customer's language. The bot never switches unless the customer does. This is enforced in the system prompt — you don't need to configure anything.

The HTTP API is a thin layer over the same database the bot reads. Everything is JSON. Authentication is environment-specific — in self-hosted deployments the API is reached on a private network or behind your own auth gateway.

Base URL

http://localhost:8000/api/v1

Response envelope

Every endpoint returns this shape:

{
  "success": true,
  "data": { /* endpoint-specific payload */ },
  "error": null
}

On failure, success is false, data is null, and error contains a message string.

Create a store

{
  "name": "My Fashion Store",
  "owner_id": "user_123"
}

{
  "data": {
    "id": 1,
    "name": "My Fashion Store",
    "owner_id": "user_123",
    "description": null,
    "return_policy": null,
    "created_at": "2026-04-06T10:00:00Z",
    "updated_at": null
  }
}

Get a store

Update a store

All fields are optional.

{
  "name": "Updated Store Name",
  "description": "We sell premium fashion for all seasons.",
  "return_policy": "Returns accepted within 7 days with original tag."
}

Connect a page

{
  "page_id": "123456789",
  "page_name": "My Fashion Store",
  "page_access_token": "EAAxxxxxx..."
}

The page_access_token comes from the Facebook Graph API. The token is encrypted before persisting.

List connected pages

Add a zone

{
  "zone_name": "Inside Dhaka",
  "charge": 60.00,
  "description": "All areas within Dhaka city"
}

List, update, delete

Add a location

{
  "location_name": "Dhanmondi Branch",
  "address": "House 10, Road 4, Dhanmondi, Dhaka",
  "phone": "01700000000",
  "opening_hours": "Sat-Thu: 10am-9pm"
}

List, update, delete

Add one product

{
  "title": "Black Winter T-Shirt",
  "descriptions": "Premium quality cotton winter t-shirt.",
  "price": 850.00,
  "currency": "BDT",
  "stock": 50,
  "sizes": ["S", "M", "L", "XL"],
  "images": [
    "https://cdn.example.com/black-tshirt-front.jpg",
    "https://cdn.example.com/black-tshirt-back.jpg"
  ],
  "discount": 10.00,
  "offer": "10% off this week",
  "store_id": 1
}

After creation, visual embedding generation runs in the background. Image URLs must be publicly accessible.

Bulk import (max 100 per call)

{
  "products": [
    {
      "title": "Red Hoodie",
      "price": 1200.00,
      "currency": "BDT",
      "stock": 30,
      "sizes": ["M", "L", "XL"],
      "images": ["https://cdn.example.com/red-hoodie.jpg"],
      "store_id": 1
    }
  ]
}

Get a product

List orders

Filter with ?status= where status is one of pending, confirmed, cancelled, delivered.

{
  "data": [
    {
      "id": 1,
      "store_id": 1,
      "messenger_user_id": 5,
      "conversation_id": 2,
      "delivery_address": "House 10, Road 4, Dhanmondi",
      "delivery_zone_name": "Inside Dhaka",
      "delivery_charge": "60.00",
      "subtotal": "850.00",
      "total": "910.00",
      "currency": "BDT",
      "status": "pending",
      "created_at": "2026-04-06T10:00:00Z",
      "items": [
        {
          "id": 1,
          "product_id": 42,
          "product_name": "Black Winter T-Shirt",
          "size": "L",
          "quantity": 1,
          "unit_price": "850.00",
          "currency": "BDT",
          "line_total": "850.00"
        }
      ]
    }
  ]
}

Get one order

Send a message to a customer outside of an in-progress conversation — e.g. an order confirmation or a delivery update.

{
  "page_id": "123456789",
  "recipient_id": "PSID_of_the_customer",
  "message": "Your order has been confirmed! We'll deliver within 2-3 days.",
  "message_tag": "POST_PURCHASE_UPDATE"
}

message_tag is required if the customer last interacted more than 24 hours ago. Valid values: POST_PURCHASE_UPDATE, CONFIRMED_EVENT_UPDATE, HUMAN_AGENT. Omit it if the customer messaged within the last 24 hours.

Facebook calls these endpoints — you generally don't hit them directly. The verification endpoint is used once during Page subscription; the POST endpoint receives every Messenger event going forward.

Returns 200 with { "status": "ok" } when the API is reachable. Use this as your container health probe.