INTENT SEARCH

Your Visitors Don't Know Your Vocabulary

Zunoy's semantic search matches what people mean rather than what they typed, so content is findable in the words your audience actually uses.

Start Building Free

Meaning, Not String Matching

Keyword search

1

Requires the searcher to guess your terminology

2

Returns nothing when wording differs

3

Synonyms need manual configuration

4

No signal about answer quality

1

Matches the concept regardless of phrasing

2

Returns the relevant entry anyway

3

Synonymy is inherent to the embedding

4

Every result carries a relevance score

FALLBACK

Falls Back to Keyword Search, and Says So

When semantic retrieval is unavailable — intelligence not configured, or no usable embedding — Zunoy falls back to keyword search rather than returning an error, and reports which mode produced the results so you can reason about what you're seeing.

Automatic fallback to keyword matching, never a hard failure

The response states whether results came from semantic or keyword

Configurable result limit and content-type filtering

One Search System, Used Everywhere in the Product

The same embedding index serves internal search for your team, public search on your website, conversational answers through the chatbot, and direct tool access for AI agents.

Admin search

Matches drafts by keyword too

Public site search

Visitors search in their own words

Conversational answers

Feeds the on-site AI chatbot

Agent tool access

intent_search via the MCP server

Only Published Content Is Ever Embedded

Start Building Free

Index built from published entries only

Drafts never surface through semantic search

Admin keyword search can still find drafts

Public search never sees drafts, either path

TUNING

Pick a Faster Model or a More Accurate One

Zunoy supports two embedding models — a smaller, faster default and a larger, higher-fidelity option — so the trade-off between retrieval quality and cost is a decision you make, and for most libraries the default is enough, while large or highly technical ones are where the larger model earns its cost.

Two embedding models, switchable per workspace

Reindex on demand when you change models or content

Result limits and type filters set per query

RELEVANCE SCORING

Every Result Shows How Confident the Match Is

Every result carries a numeric relevance score rather than a binary match or no-match — a high score means the retrieved passage genuinely addresses the query, a low one means it's the closest thing found rather than a good answer, and Zunoy surfaces that score instead of hiding it.

A numeric score per result, not just a rank order

Low scores are visible, not smoothed over

The same score powers the chatbot's "I'm not sure" behavior

How It Works

Indexing happens as a consequence of publishing, so search quality improves as your library grows without a separate maintenance workflow.

1

Enable Intelligence

Turn it on for the workspace and select an embedding model.

2

Content is chunked and embedded

Published entries are broken into passages and indexed automatically.

3

A query arrives

From the admin, your public site, the chatbot, or an AI agent.

4

Results return with scores

Ranked by relevance, with the retrieval mode reported.

1

Enable Intelligence

Turn it on for the workspace and select an embedding model.

2

Content is chunked and embedded

Published entries are broken into passages and indexed automatically.

3

A query arrives

From the admin, your public site, the chatbot, or an AI agent.

4

Results return with scores

Ranked by relevance, with the retrieval mode reported.

1

Enable Intelligence

Turn it on for the workspace and select an embedding model.

2

Content is chunked and embedded

Published entries are broken into passages and indexed automatically.

3

A query arrives

From the admin, your public site, the chatbot, or an AI agent.

4

Results return with scores

Ranked by relevance, with the retrieval mode reported.

1

Enable Intelligence

Turn it on for the workspace and select an embedding model.

2

Content is chunked and embedded

Published entries are broken into passages and indexed automatically.

3

A query arrives

From the admin, your public site, the chatbot, or an AI agent.

4

Results return with scores

Ranked by relevance, with the retrieval mode reported.

1

Enable Intelligence

Turn it on for the workspace and select an embedding model.

2

Content is chunked and embedded

Published entries are broken into passages and indexed automatically.

3

A query arrives

From the admin, your public site, the chatbot, or an AI agent.

4

Results return with scores

Ranked by relevance, with the retrieval mode reported.

Who This Is For

Content teams

Find the entry you half-remember without knowing its exact title, and see which visitor searches your library fails to answer.

Keyword search reaches drafts; semantic search covers what's published

Content gaps surfaced from real query logs

Site visitors

Ask in their own words and find the right page, rather than guessing the internal terminology your team happens to use.

No need to match your exact phrasing

Developers

A retrieval primitive with transparent modes and scoring, usable from the API and callable directly as an MCP tool.

Mode reported per response for debuggability

Configurable limits and type filtering

AI engineers

Semantic retrieval over structured content, plus a pre-chunked context endpoint for agents you build yourself.

intent_search and public_ask as agent tools

Connected to the Rest of the Platform

Retrieval is one layer serving every AI surface in the product.

AI Chatbot

Conversational answers built on this

Learn more

AI Studio

Same index powers 20 editing actions

Learn more

MCP Server

intent_search as an agent tool

Learn more

Insights

Query logs surfacing content gaps

Learn more

Common Questions

What is intent-based semantic search?

It retrieves content by conceptual similarity rather than shared keywords, embedding both the query and your content as vectors. A query and a document can match strongly while sharing no words at all.

Does public search expose unpublished content?

No. Only published entries are ever embedded for semantic matching. In the admin, keyword search can additionally surface drafts on request, but public search never can, through either method.

Which embedding model does Zunoy use?

Two options are supported — a smaller, faster default and a larger, higher-fidelity alternative — so you choose the balance between retrieval quality and cost for your workspace.

What happens if semantic search is unavailable?

Zunoy falls back to keyword matching rather than returning an error, and the response reports which mode produced the results so the behaviour is transparent rather than a silent degradation.

Can I control how many results come back?

Yes. Result limits are configurable per query, defaulting to ten, and results can be filtered by content type to narrow retrieval to a specific part of your library.

Do I need to reindex when content changes?

Published content is indexed as part of publishing. A manual reindex is available for when you switch embedding models or need to rebuild the index deliberately.

Make Your Content Findable in Your Visitors' Words

They already know what they want. The search box should not require them to know what you called it.

Start Building Free
See Pricing

Ask a question about Zunoy's products, pricing, or docs.

⌘K