Build A Weaviate Query Agent - The E-Commerce Assistant
In this recipe, we will be building a simple e-commerce assistant agent with the Weaviate Query Agent. This agent will have access to a number of Weaviate collections, and will be capable of answering complex queries about brands and clothing items, accessing information from each collection.
📚 You can read and learn more about this service in our "Introducing the Weaviate Query Agent" blog.
To get started, we've prepared a few open datasets, available on Hugging Face. The first step will be walking through how to populate your Weaviate Cloud collections.
- E-commerce: A dataset that lists clothing items, prices, brands, reviews etc.
- Brands: A dataset that lists clothing brands and information about them such as their parent brand, child brands, average customer rating etc.
Additionally, we also have access to some other unrelated datasets which you can use to add more capabilities and variety to other agents later on in the recipe:
- Financial Contracts: A dataset of financial contracts between indivicuals and/or companies, as well as information on the type of contract and who has authored them.
- Weather: Daily weather information including temperature, wind speed, percipitation, pressure etc.
1. Setting Up Weaviate & Importing Data
To use the Weaviate Query Agent, first, create a Weaviate Cloud account👇
- Create Serverless Weaviate Cloud account and setup a free Sandbox
- Go to 'Embedding' and enable it, by default, this will make it so that we use
Snowflake/snowflake-arctic-embed-l-v2.0as the embedding model - Take note of the
WEAVIATE_URLandWEAVIATE_API_KEYto connect to your cluster below
Info: We recommend using Weaviate Embeddings so you do not have to provide any extra keys for external embedding providers.
!pip install weaviate-client[agents] datasets
import os
from getpass import getpass
if "WEAVIATE_API_KEY" not in os.environ:
os.environ["WEAVIATE_API_KEY"] = getpass("Weaviate API Key")
if "WEAVIATE_URL" not in os.environ:
os.environ["WEAVIATE_URL"] = getpass("Weaviate URL")
import weaviate
from weaviate.auth import Auth
client = weaviate.connect_to_weaviate_cloud(
cluster_url=os.environ.get("WEAVIATE_URL"),
auth_credentials=Auth.api_key(os.environ.get("WEAVIATE_API_KEY")),
)
Prepare the Collections
In the following code blocks, we are pulling our demo datasets from Hugging Face and writing them to new collections in our Weaviate Serverless cluster.
❗️ The
QueryAgentuses the descriptions of collections and properties to decide which ones to use when solving queries, and to access more information about properties. You can experiment with changing these descriptions, providing more detail, and more. It's good practice to provide property descriptions too. For example, below we make sure that theQueryAgentknows that prices are all in USD, which is information that would otherwise be unavailable.
from weaviate.classes.config import Configure, Property, DataType
# To re-run cell you may have to delete collections
# client.collections.delete("Brands")
client.collections.create(
"Brands",
description="A dataset that lists information about clothing brands, their parent companies, average rating and more.",
vectorizer_config=Configure.Vectorizer.text2vec_weaviate()
)
# client.collections.delete("Ecommerce")
client.collections.create(
"Ecommerce",
description="A dataset that lists clothing items, their brands, prices, and more.",
vectorizer_config=Configure.Vectorizer.text2vec_weaviate(),
properties=[
Property(name="collection", data_type=DataType.TEXT),
Property(name="category", data_type=DataType.TEXT),
Property(name="tags", data_type=DataType.TEXT_ARRAY),
Property(name="subcategory", data_type=DataType.TEXT),
Property(name="name", data_type=DataType.TEXT),
Property(name="description", data_type=DataType.TEXT),
Property(name="brand", data_type=DataType.TEXT),
Property(name="product_id", data_type=DataType.UUID),
Property(name="colors", data_type=DataType.TEXT_ARRAY),
Property(name="reviews", data_type=DataType.TEXT_ARRAY),
Property(name="image_url", data_type=DataType.TEXT),
Property(name="price", data_type=DataType.NUMBER, description="price of item in USD"),
]
)
# client.collections.delete("Weather")
client.collections.create(
"Weather",
description="Daily weather information including temperature, wind speed, percipitation, pressure etc.",
vectorizer_config=Configure.Vectorizer.text2vec_weaviate(),
properties=[
Property(name="date", data_type=DataType.DATE),
Property(name="humidity", data_type=DataType.NUMBER),
Property(name="precipitation", data_type=DataType.NUMBER),
Property(name="wind_speed", data_type=DataType.NUMBER),
Property(name="visibility", data_type=DataType.NUMBER),
Property(name="pressure", data_type=DataType.NUMBER),
Property(name="temperature", data_type=DataType.NUMBER, description="temperature value in Celsius")
]
)
# client.collections.delete("Financial_contracts")
client.collections.create(
"Financial_contracts",
description="A dataset of financial contracts between indivicuals and/or companies, as well as information on the type of contract and who has authored them.",
vectorizer_config=Configure.Vectorizer.text2vec_weaviate(),
)
from datasets import load_dataset
brands_dataset = load_dataset("weaviate/agents", "query-agent-brands", split="train", streaming=True)
ecommerce_dataset = load_dataset("weaviate/agents", "query-agent-ecommerce", split="train", streaming=True)
weather_dataset = load_dataset("weaviate/agents", "query-agent-weather", split="train", streaming=True)
financial_dataset = load_dataset("weaviate/agents", "query-agent-financial-contracts", split="train", streaming=True)
brands_collection = client.collections.get("Brands")
ecommerce_collection = client.collections.get("Ecommerce")
weather_collection = client.collections.get("Weather")
financial_collection = client.collections.get("Financial_contracts")
with brands_collection.batch.dynamic() as batch:
for item in brands_dataset:
batch.add_object(properties=item["properties"])
with ecommerce_collection.batch.dynamic() as batch:
for item in ecommerce_dataset:
batch.add_object(properties=item["properties"])
with weather_collection.batch.dynamic() as batch:
for item in weather_dataset:
batch.add_object(properties=item["properties"])
with financial_collection.batch.dynamic() as batch:
for item in financial_dataset:
batch.add_object(properties=item["properties"])
2. Set Up the Query Agent
When setting up the query agent, we have to provide it a few things:
- The
client - The
collectionwhich we want the agent to have access to. - (Optionally) A
system_promptthat describes how our agent should behave - (Optionally) Timeout - which for now defaults to 60s.
Let's start with a simple agent. Here, we're creating an agent that has access to our Brands & Ecommerce datasets.
from weaviate.agents.query import QueryAgent
agent = QueryAgent(
client=client, collections=["Ecommerce", "Brands"],
)
3. Run the Query Agent
When we run the agent, it will first make a few decisions, depending on the query:
- The agent will decide which collection or collections to look up an answer in.
- The agent will also decide whether to perform a regular search query, what filters to use, whether to do an aggregation query, or all of them together!
- It will then provide a reponse within
QueryAgentResponse. We will use theprint_query_agent_responsefunction for a nice display of various information provided in the response object.
Ask a Question
Let's start with a simple question: "I like the vintage clothes, can you list me some options that are less than $200?"
We can then also inspect how the agent responded, what kind of searches it performed on which collections, whether it has identified if the final answer is missing information or not, as well as the final answer 👇
from weaviate.agents.utils import print_query_agent_response
response = agent.run("I like the vintage clothes, can you list me some options that are less than $200?")
print_query_agent_response(response)
╭─────────────────────────────────────────────── 🔍 Original Query ───────────────────────────────────────────────╮ │ │ │ I like the vintage clothes, can you list me some options that are less than $200? │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭──────────────────────────────────────────────── 📝 Final Answer ────────────────────────────────────────────────╮ │ │ │ If you are looking for vintage clothing options under $200, here are some great choices: │ │ │ │ 1. Vintage Philosopher Midi Dress - Priced at $125, this dress from Echo & Stitch embraces a classic │ │ scholarly look with its deep green velvet fabric and antique gold detailing. It's tailored for elegance and is │ │ ideal for sophisticated occasions. │ │ │ │ 2. Vintage Gale Pleated Dress - This $120 dress from Solemn Chic features deep burgundy pleats and │ │ vintage-inspired sleeve details, perfect for a timeless scholarly appearance. │ │ │ │ 3. Retro Groove Flared Pants - For $59, these electric blue flared pants from Vivid Verse bring back the │ │ playful spirit of the early 2000s with a modern touch. │ │ │ │ 4. Vintage Scholar Tote - At $90, this tote from Echo & Stitch combines functionality and elegance, ideal │ │ for everyday use, especially if you enjoy a scholarly aesthetic. │ │ │ │ 5. Electric Velvet Trousers - Priced at $60, these neon green velvet trousers from Vivid Verse offer a fun, │ │ throwback vibe to early Y2K fashion. │ │ │ │ 6. Victorian Velvet Jumpsuit - For $120, this jumpsuit from Solemn Chic offers an elegant blend of romance │ │ and scholarly charm, suited for library visits or cultured gatherings. │ │ │ │ 7. Vintage Scholar Turtleneck - This $55 turtleneck from Echo & Stitch suits the Dark Academia vibe, │ │ perfect for layering or wearing alone. │ │ │ │ 8. Vintage Ivy Loafers - These $120 loafers from Solemn Chic offer timeless sophistication, with a deep │ │ burgundy finish that complements any vintage wardrobe. │ │ │ │ These options cater to various preferences, from dresses and jumpsuits to pants and accessories, all capturing │ │ the vintage essence at an affordable price. │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭─────────────────────────────────────────── 🔭 Searches Executed 1/1 ────────────────────────────────────────────╮ │ │ │ QueryResultWithCollection( │ │ queries=['vintage clothes'], │ │ filters=[ │ │ [ │ │ IntegerPropertyFilter( │ │ property_name='price', │ │ operator=<ComparisonOperator.LESS_THAN: '<'>, │ │ value=200.0 │ │ ) │ │ ] │ │ ], │ │ filter_operators='AND', │ │ collection='Ecommerce' │ │ ) │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮ │ │ │ 📊 No Aggregations Run │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭────────────────────────────────────────────────── 📚 Sources ───────────────────────────────────────────────────╮ │ │ │ - object_id='5e9c5298-5b3a-4d80-b226-64b2ff6689b7' collection='Ecommerce' │ │ - object_id='48896222-d098-42e6-80df-ad4b03723c19' collection='Ecommerce' │ │ - object_id='00b383ca-262f-4251-b513-dafd4862c021' collection='Ecommerce' │ │ - object_id='cbe8f8be-304b-409d-a2a1-bafa0bbf249c' collection='Ecommerce' │ │ - object_id='c18d3c5b-8fbe-4816-bc60-174f336a982f' collection='Ecommerce' │ │ - object_id='1811da1b-6930-4bd1-832e-f8fa2119d4df' collection='Ecommerce' │ │ - object_id='2edd1bc5-777e-4376-95cd-42a141ffb71e' collection='Ecommerce' │ │ - object_id='9819907c-1015-4b4c-ac75-3b3848e7c247' collection='Ecommerce' │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
📊 Usage Statistics
┌────────────────┬──────┐ │ LLM Requests: │ 3 │ │ Input Tokens: │ 7774 │ │ Output Tokens: │ 512 │ │ Total Tokens: │ 8286 │ └────────────────┴──────┘
Total Time Taken: 16.93s
Ask a follow up question
The agent can also be provided with additional context. For example, we can provide the previous response as context and get a new_response
new_response = agent.run("What about some nice shoes, same budget as before?", context=response)
print_query_agent_response(new_response)
╭─────────────────────────────────────────────── 🔍 Original Query ───────────────────────────────────────────────╮ │ │ │ What about some nice shoes, same budget as before? │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭──────────────────────────────────────────────── 📝 Final Answer ────────────────────────────────────────────────╮ │ │ │ Here are some great shoe options under $200 that you might like: │ │ │ │ 1. Vintage Noir Loafers - Priced at $125, these loafers are part of the Dark Academia collection by Solemn │ │ Chic. They come in black and grey, featuring a classic design with a modern twist. Reviews highlight their │ │ comfort and stylish appearance, making them suitable for both casual and formal settings. │ │ │ │ 2. Parchment Boots - At $145, these boots from Nova Nest's Light Academia collection are noted for their │ │ elegant ivory leather and classical detail stitching. They are praised for their comfort and versatile style. │ │ │ │ 3. Bramble Berry Loafers - These loafers, priced at $75, come in pink and green and are marked by their │ │ eco-friendly material and countryside aesthetic. Produced by Eko & Stitch, they are loved for their comfort and │ │ sustainability. │ │ │ │ 4. Glide Platforms - Available for $90 from the Y2K collection by Vivid Verse, these platform sneakers are │ │ both comfortable and stylish with a high-shine pink finish. │ │ │ │ 5. Sky Shimmer Sneaks - Costing $69, these sneakers are from the Y2K collection by Nova Nest and offer a │ │ comfortable fit with a touch of sparkle for style. │ │ │ │ These selections offer a mix of formal and casual styles, ensuring you can find a perfect pair under your │ │ budget of $200. │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭─────────────────────────────────────────── 🔭 Searches Executed 1/1 ────────────────────────────────────────────╮ │ │ │ QueryResultWithCollection( │ │ queries=['nice shoes'], │ │ filters=[ │ │ [ │ │ IntegerPropertyFilter( │ │ property_name='price', │ │ operator=<ComparisonOperator.LESS_THAN: '<'>, │ │ value=200.0 │ │ ) │ │ ] │ │ ], │ │ filter_operators='AND', │ │ collection='Ecommerce' │ │ ) │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮ │ │ │ 📊 No Aggregations Run │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭────────────────────────────────────────────────── 📚 Sources ───────────────────────────────────────────────────╮ │ │ │ - object_id='96b30047-8ce1-4096-9bcf-009733cf8613' collection='Ecommerce' │ │ - object_id='61e4fcd7-d2bc-4861-beb6-4c16948d9921' collection='Ecommerce' │ │ - object_id='6e533f7d-eba1-4e74-953c-9d43008278e7' collection='Ecommerce' │ │ - object_id='f873ac48-1311-462a-86b2-a28b15fdda7a' collection='Ecommerce' │ │ - object_id='93b8b13e-a417-4be2-9cce-fda8c767f35e' collection='Ecommerce' │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
📊 Usage Statistics
┌────────────────┬───────┐ │ LLM Requests: │ 4 │ │ Input Tokens: │ 9783 │ │ Output Tokens: │ 574 │ │ Total Tokens: │ 10357 │ └────────────────┴───────┘
Total Time Taken: 18.02s
Now let's try a question that sholud require an aggregation. Let's see which brand lists the most shoes.
response = agent.run("What is the the name of the brand that lists the most shoes?")
print_query_agent_response(response)
╭─────────────────────────────────────────────── 🔍 Original Query ───────────────────────────────────────────────╮ │ │ │ What is the the name of the brand that lists the most shoes? │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭──────────────────────────────────────────────── 📝 Final Answer ────────────────────────────────────────────────╮ │ │ │ The brand that lists the most shoes is Loom & Aura with a total of 118 shoe listings. │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮ │ │ │ 🔭 No Searches Run │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭──────────────────────────────────────────── 📊 Aggregations Run 1/1 ────────────────────────────────────────────╮ │ │ │ AggregationResultWithCollection( │ │ search_query=None, │ │ groupby_property='brand', │ │ aggregations=[ │ │ IntegerPropertyAggregation(property_name='collection', metrics=<NumericMetrics.COUNT: 'COUNT'>) │ │ ], │ │ filters=[], │ │ collection='Ecommerce' │ │ ) │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
📊 Usage Statistics
┌────────────────┬──────┐ │ LLM Requests: │ 3 │ │ Input Tokens: │ 3976 │ │ Output Tokens: │ 159 │ │ Total Tokens: │ 4135 │ └────────────────┴──────┘
Total Time Taken: 5.33s
Search over multiple collections
In some cases, we need to combine the results of searches across multiple collections. From the result above, we can see that "Loom & Aura" lists the most shoes.
Let's imagine a scenario where the user would now want to find out more about this company, as well as the items that they sell.
response = agent.run("Does the brand 'Loom & Aura' have a parent brand or child brands and what countries do they operate from? "
"Also, what's the average price of a item from 'Loom & Aura'?")
print_query_agent_response(response)
╭─────────────────────────────────────────────── 🔍 Original Query ───────────────────────────────────────────────╮ │ │ │ Does the brand 'Loom & Aura' have a parent brand or child brands and what countries do they operate from? Also, │ │ what's the average price of a item from 'Loom & Aura'? │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭──────────────────────────────────────────────── 📝 Final Answer ────────────────────────────────────────────────╮ │ │ │ Loom & Aura is itself a well-established brand based in Italy and operates as the parent brand to several child │ │ brands. These child brands include 'Loom & Aura Active', 'Loom & Aura Kids', 'Nova Nest', 'Vivid Verse', 'Loom │ │ Luxe', 'Saffron Sage', 'Stellar Stitch', 'Nova Nectar', 'Canvas Core', and 'Loom Lure'. The countries │ │ associated with the operations or origins of these child brands include Italy, USA, UK, Spain, South Korea, │ │ Japan, and some extend beyond Italy as suggested by the presence of these brands in different countries. │ │ │ │ The average price of an item from Loom & Aura is approximately $87.11. This reflects the brand's positioning as │ │ offering items of timeless elegance and quality craftsmanship. │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭─────────────────────────────────────────── 🔭 Searches Executed 1/2 ────────────────────────────────────────────╮ │ │ │ QueryResultWithCollection( │ │ queries=['parent brand of Loom & Aura', 'child brands of Loom & Aura'], │ │ filters=[[], []], │ │ filter_operators='AND', │ │ collection='Brands' │ │ ) │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭─────────────────────────────────────────── 🔭 Searches Executed 2/2 ────────────────────────────────────────────╮ │ │ │ QueryResultWithCollection( │ │ queries=['Loom & Aura'], │ │ filters=[ │ │ [ │ │ TextPropertyFilter( │ │ property_name='name', │ │ operator=<ComparisonOperator.LIKE: 'LIKE'>, │ │ value='Loom & Aura' │ │ ) │ │ ] │ │ ], │ │ filter_operators='AND', │ │ collection='Brands' │ │ ) │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭──────────────────────────────────────────── 📊 Aggregations Run 1/1 ────────────────────────────────────────────╮ │ │ │ AggregationResultWithCollection( │ │ search_query=None, │ │ groupby_property=None, │ │ aggregations=[IntegerPropertyAggregation(property_name='price', metrics=<NumericMetrics.MEAN: 'MEAN'>)], │ │ filters=[ │ │ TextPropertyFilter( │ │ property_name='brand', │ │ operator=<ComparisonOperator.EQUALS: '='>, │ │ value='Loom & Aura' │ │ ) │ │ ], │ │ collection='Ecommerce' │ │ ) │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
╭────────────────────────────────────────────────── 📚 Sources ───────────────────────────────────────────────────╮ │ │ │ - object_id='88433e18-216d-489a-8719-81a29b0ae915' collection='Brands' │ │ - object_id='99f42d07-51e9-4388-9c4b-63eb8f79f5fd' collection='Brands' │ │ - object_id='0852c2a4-0c5a-4c69-9762-1be10bc44f2b' collection='Brands' │ │ - object_id='d172a342-da41-45c3-876e-d08db843b8b3' collection='Brands' │ │ - object_id='a7ad0ed7-812e-4106-a29f-40442c3a106e' collection='Brands' │ │ - object_id='b6abfa02-18e5-44cf-a002-ba140e3623ad' collection='Brands' │ │ │ ╰─────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
📊 Usage Statistics
┌────────────────┬───────┐ │ LLM Requests: │ 5 │ │ Input Tokens: │ 9728 │ │ Output Tokens: │ 479 │ │ Total Tokens: │ 10207 │ └────────────────┴───────┘
Total Time Taken: 11.38s
Changing the System Prompt
In some cases, you may want to define a custom system_prompt for your agent. This can help you provide the agent with some default instructions as to how to behave. For example, let's create an agent that will always answer the query in the users language.
Let's also create a QueryAgent that has access to two more collections, Financial_contracts and Weather. Next, you can try out more queries yourself!
multi_lingual_agent = QueryAgent(
client=client, collections=["Ecommerce", "Brands", "Financial_contracts", "Weather"],
system_prompt="You are a helpful assistant that always generated the final response in the users language."
" You may have to translate the user query to perform searches. But you must always respond to the user in their own language."
)
For example, this time lets ask something that is about weather!
response = multi_lingual_agent.run("Quelles sont les vitesses minimales, maximales et moyennes du vent?")
print(response.final_answer)
Python output:
Les vitesses de vent minimales, maximales et moyennes sont respectivement de 8,40 km/h, 94,88 km/h et 49,37 km/h. Ces données offrent une vue d'ensemble des conditions de vent typiques mesurées dans une période ou un lieu donné.
Try More Questions
For example Let's try to find out more about the brans "Eko & Stitch"
response = multi_lingual_agent.run("Does Eko & Stitch have a branch in the UK? Or if not, does it have parent or child company in the UK?")
print(response.final_answer)
Python output:
Yes, Eko & Stitch has a branch in the UK. The brand is part of the broader company Nova Nest, which serves as Eko & Stitch's parent brand. Eko & Stitch itself operates in the UK and has its child brands, Eko & Stitch Active and Eko & Stitch Kids, also within the UK.
Our multi_lingual_agent also has access to a collection called "Financial_contracts". Let's try to find out some more information about this dataser.
response = multi_lingual_agent.run("What kinds of contracts are listed? What's the most common type of contract?")
print(response.final_answer)
Python output:
The query seeks to identify the types of contracts listed and determine the most common type. Among the types of contracts provided in the results, the following were identified: employment contracts, sales agreements, invoice contracts, service agreements, and lease agreements. The most common type of contract found in the search results is the employment contract. However, when considering data from both search and aggregation results, the aggregation reveals that the invoice contract is the most common, followed by service agreements and lease agreements. While employment contracts appear frequently in the search results, they rank fourth in the aggregation data in terms of overall occurrences.