Ultrahost/ docs
All primitives

Data / semantic retrieval

Vector DB

Search for ideas, not just words. A Vector Database compares the meaning behind a request with the meaning captured in your content—not just exact terms.

The idea

Imagine building a recipe application. Someone searches: “I want something comforting to eat on a rainy day.” A traditional keyword search can miss because no recipe contains those exact words. A Vector Database can still find soups and mac and cheese because their meaning is related to the request.

Application blip

Recipe Discovery

5 recipes indexed
Try:

Closest matches

5 recipes ranked by meaning for “Something warm and comforting.”

Vector similarity

#1 · Comforting

Creamy tomato soup

98%

Slow-simmered tomatoes, a little cream, and toasted sourdough.

#2 · Comforting

Chicken noodle soup

92%

A cozy bowl of noodles, vegetables, and savory broth.

#3 · Comforting

Homemade mac and cheese

84%

Golden baked pasta with a creamy, bubbling cheese sauce.

#4 · Fresh

Spicy summer salad

22%

Crisp greens, ripe tomatoes, and a bright chili-lime dressing.

#5 · Refreshing

Fresh fruit bowl

16%

Chilled seasonal fruit with mint and a squeeze of citrus.

Sample vectors and scores are illustrative. A real app generates embeddings and queries a vector index; this demo uses a tiny local vector space to show how meaning changes ranking.

Behind the scenes

  1. Text
  2. Numerical representation of meaning
  3. Similarity comparison
  4. Relevant results

An embedding turns content and queries into vectors. Similarity scores rank the closest items first, even when their wording differs.

What could you build?

  • Product recommendations
  • Semantic document search
  • Content discovery
  • Matching applications

Good for

  • Semantic document search
  • Recommendations
  • RAG retrieval

Implementation notes

  • Chunk source content with stable IDs.
  • Store source metadata with every vector.
  • Filter by tenant and permissions before ranking.

Implementing it

With a coding agent

Ask your agent to add search by meaning

Agent setup (mock link)

Copy this prompt, fill in the bracketed details, and paste it into your coding agent. It describes the goal, the search behavior, and the guardrails.

Bracketed text marks details only you can fill in. 857 / 6,000 characters.

Fill in:APP OR PRODUCTCONTENT TO FINDSOURCE CONTENTMETADATA TO KEEPNUMBERUSER, WORKSPACE, OR TENANTIMPORTANT QUERIES, SIMILAR-MEANING CASES, AND ACCESS CASES

Good foundation · 72/100

Traditional SDK / API

Use it from a function

TypeScript
const matches = await ultrahost.vectors.search("knowledge", {
  query: question,
  topK: 6,
  filter: { workspaceId },
})

Key takeaway

A Vector Database helps your application recognize similarity in meaning, even when the words are different.

Next primitive

Schemas & Ontology Versioning