A text embedding is a list of numbers that represents meaning. Cloudflare Vectorize gives those vectors a searchable home, so an application can retrieve related documents even when the user's words do not exactly match the stored text. A useful search system also needs stable document IDs, compatible dimensions, asynchronous update handling and strict customer boundaries.
This course builds a small support-knowledge service one capability at a time. You will create and maintain an index, perform semantic retrieval, keep searches inside server-controlled customer and category scope, and use bounded evidence to ground an answer. Every new database or retrieval concept is explained next to the operation that needs it.
The course contains 5 guided labs and 1 independent customer-isolation challenge. Guided Labs introduce each capability before the independent challenge.
What You Will Learn
- Create a Document Vector Index: configure a compatible index, metadata indexes and identified document vectors.
- Keep Indexed Documents Current: replace changed records, retire one ID and observe asynchronous mutations safely.
- Retrieve Similar Help Articles: embed a question, rank nearby articles and return an explicit empty result.
- Scope Search by Customer and Category: combine a server-selected namespace with an indexed metadata filter.
- Build a Grounded Answer Endpoint: retrieve bounded source passages before generation and cite resolvable sources.
- Cross-Customer Search Results: independently repair an endpoint that trusts an unsafe scope override.
Who This Course Is For
Workers AI learners who want to maintain a vector index and build scoped, source-backed search.
Prerequisites: Complete Cloudflare Workers AI for Beginners and Connect LabEx to Your Cloudflare Account. You should already recognize a JavaScript Worker, a Workers AI binding and a 384-dimensional BGE Small embedding.
Learning environment: Use your own Cloudflare learning account. Every unit starts in a fresh VM, creates uniquely named disposable resources, proves real cloud behavior, deletes only those resources and logs out. The tiny indexes and bounded queries fit the documented Workers Free allowance; the grounded-answer lab also uses the shared daily Workers AI Free allocation and does not require Workers Paid.
Frequently Asked Questions
What is a text embedding?
It is a list of numbers representing meaning. Vectorize stores vectors so an application can retrieve related documents even when question wording differs from the stored text.
Will an index change appear immediately?
Index changes are asynchronous. The labs use stable document IDs and bounded observation to replace changed records, retire an ID and confirm the application’s retrieval result.
How are customer boundaries enforced?
The application selects a namespace on the server and applies an indexed metadata filter. The final challenge repairs an endpoint that trusts an unsafe client scope override.
Does grounded generation always have enough evidence?
No. Retrieve bounded passages, cite resolvable sources and return an explicit empty result when the intended evidence is unavailable. The grounded-answer lab uses the shared Workers AI daily allocation.
