Search Engine Configuration

7 min read

This page documents every setting under Search Engine in the Spaces configuration UI. Each setting is shown by its UI label, with the underlying configuration field name in italics.

The list of available engines is decided at the platform level (ACTIVE_SEARCH_ENGINES and credentials). Only engines on that list — and whose credentials are provisioned — appear in the selector. See Activation Reference under Platform / Infrastructure.

There are two kinds of search engine. They behave differently and expose different settings:

Kind	Engines	What happens
Standard search	Google, Brave, Perplexity, Customized API	Your query is sent to a search API. You get a list of results (URL, title, snippet; some providers include page content). A Web Page Reader may run afterward if the engine does not return full page content.
Grounding (agent search)	Grounding with Bing, Grounding with VertexAI	An AI agent or grounded model searches and reads on your behalf, then returns a structured answer that the tool turns into search results. Settings look different (agent / model / instructions).

Brave and Perplexity require Search Proxy on the platform (infrastructure). They are not available when assistants-core uses the legacy direct-provider path.

For runtime data flows, see Search Engines under Platform / Architecture. For credentials (Search Proxy and legacy assistants-core), see Search Engine Setup under Platform / Infrastructure.

Parameter exposure (Standard search only)

Many optional settings on standard engines appear as a pair of controls:

Control	Meaning
Value	The default the platform applies when this setting is active
Expose	Whether the AI assistant is allowed to set or override this parameter for a given search

How to decide

Value empty / deactivated — The parameter is off. It is not sent to the search provider. Turning Expose on without a value does not invent a value by itself.
Expose off + Value set — Every search uses that fixed value. The AI cannot change it (admin lock). Use this for policy (for example always search as Switzerland, or always apply a site filter).
Expose on — The AI may choose the value when it searches. If Value is set, it is the default when the AI does not override.
Grounding engines do not use Expose / Value — Bing and VertexAI settings are fixed deployment / agent configuration.
Web Page Readers do not use Expose / Value — Reader settings are fixed for the Space; the AI does not get per-call knobs for crawling.

Example (Google — Geolocation)

Lock results to Switzerland: set Geolocation (gl) Value to ch, keep Expose off.
Let the assistant pick a country per question: turn Expose on (optionally leave a default Value).

Common setting

Every engine shares:

UI Label	Field	Type	Default
Number of search results to fetch	`fetch_size`	integer	`5`

Higher values bring more URLs into the pipeline at the cost of more downstream work.

Standard search

Google Search

Display name in the UI: Google

What it does

Sends your query to Google's Custom Search JSON API and returns URLs, titles, and snippets. Google does not return full page content, so the configured Web Page Reader usually runs afterward.

Settings

UI Label	Field	Type	Default / notes
Number of search results to fetch	`fetch_size`	integer	`5`
Search engine ID (cx)	`search_engine_id`	string	Optional; often set by the platform
Safe search	`safe`	`active` / `off`	`active` — applied to every search (not an Expose pair)
Geolocation (gl)	`gl`	Expose + Value	Country code (e.g. `ch`, `us`)
Interface language (hl)	`hl`	Expose + Value	UI / snippet language
Language restrict (lr)	`lr`	Expose + Value	e.g. `lang_en`
Date restrict	`dateRestrict`	Expose + Value	e.g. `d7`, `m1`
Exact terms	`exactTerms`	Expose + Value	Phrase every hit must contain
Exclude terms	`excludeTerms`	Expose + Value	Phrase that must not appear
File type	`fileType`	Expose + Value	e.g. `pdf`
Site search	`siteSearch`	Expose + Value	Domain restriction
Site search filter	`siteSearchFilter`	Expose + Value	`i` = include only that site, `e` = exclude
Sort	`sort`	Expose + Value	e.g. `date`

When to use it

Default for many Spaces. Best when you want a familiar index and full control via CSE configuration. Plan for Google Custom Search JSON API lifecycle limits announced by Google; consider Brave or Perplexity (Search Proxy) as alternatives.

Brave Search

Display name in the UI: Brave

Requires Search Proxy. New standard provider; not available without Search Proxy.

What it does

Sends your query to Brave's Search API. Returns URLs, titles, and rich snippets. Snippets are often descriptive enough that a full page crawl is not always needed.

Settings (high level)

UI Label	Field	Notes
Number of search results to fetch	`fetch_size`	Max practical page size is limited by Brave (typically up to 20 per request)
Extra snippets	`extra_snippets`	Fixed platform/default setting (not Expose)
Spellcheck	`spellcheck`	Fixed
Text decorations	`text_decorations`	Fixed
Search operators	`operators`	Fixed
UI language	`ui_lang`	Fixed default language for the response UI
Country / search language / result filters	various	Often available as Expose + Value pairs — see the Spaces form for the full list

When to use it

Privacy-respecting alternative to Google. Requires Search Proxy on the platform; good when you want an independent index.

Perplexity

Display name in the UI: Perplexity

Requires Search Proxy. New standard provider; not available without Search Proxy.

What it does

Sends your query to Perplexity's Search API. Returns search results with configurable how much page content is extracted.

Settings (high level)

UI Label	Field	Notes
Number of search results to fetch	`fetch_size`
Search context size	`searchContextSize`	`low` / `medium` / `high` — how much content to extract
Max tokens / Max tokens per page	`max_tokens` / `maxTokensPerPage`	Optional caps; do not combine carelessly with context size
Country	`country`	Expose + Value
Search language filter	`searchLanguageFilter`	Expose + Value
Search domain filter	`searchDomainFilter`	Expose + Value
Search recency filter	`searchRecencyFilter`	Expose + Value (`hour` / `day` / `week` / `month` / `year`)
Date filters	`lastUpdated` / `searchAfter` / `searchBefore*`	Expose + Value where shown

When to use it

When you want flexible content extraction and filters from Perplexity, without running a separate grounding agent. (Requires Search Proxy on the platform.)

Grounding (agent search)

Grounding engines are not classic search APIs. An upstream AI agent or grounded model searches and reads sources, then returns an answer that Unique turns into structured search results. They do not use Parameter exposure (Expose / Value).

Grounding with Bing

Display name in the UI: Grounding with Bing

What it does

Delegates search to an Azure AI Foundry Agent with Bing grounding. The agent searches Bing, reads sources, and produces a structured response. By default no separate Web Page Reader pass is needed.

Settings

UI Label	Field	Type	Default
Number of search results to fetch	`fetch_size`	integer	`5`
Requires Scraping	`requires_scraping`	boolean	`false`
Agent ID	`agent_id`	string	empty (auto-provision)
Endpoint	`endpoint`	string	often set by the platform; with Search Proxy, Foundry endpoint is on Search Proxy as `BING_AGENT_ENDPOINT` (see Search Engine Setup)
Generation Instructions	`generation_instructions`	textarea	Built-in prompt
Timeout	`timeout`	integer (seconds)	`120`
Language Model	`language_model`	language model	Toolkit default (fallback JSON parser if the agent answer is not valid JSON)

Latency is typically higher than direct standard-search APIs.

When to use it

When you want Bing grounding via Azure AI Foundry without managing a classic search-API integration yourself.

Grounding with VertexAI

Display name in the UI: Grounding with VertexAI

What it does

Calls a Gemini model on Vertex AI with Google's grounding tool. Gemini searches, reads, and returns a grounded answer with citations. Redirect URLs can be resolved to canonical sites when enabled. By default no separate Web Page Reader pass is needed.

Settings

UI Label	Field	Type	Default
VertexAI Model Name	`vertexai_model_name`	string	`gemini-3-flash-preview`
Generation Instructions	`generation_instructions`	textarea	Built-in prompt
Timeout	`timeout`	integer (seconds)	`120`
Language Model	`language_model`	language model	Toolkit default (fallback JSON parser if the grounding response is not valid JSON)
Requires Scraping	`requires_scraping`	boolean	`false`
Enable Enterprise Search	`enable_enterprise_search`	boolean	`false`
Enable Redirect Resolution	`enable_redirect_resolution`	boolean	`true`

VertexAI has no fetch_size. Result count is decided by the grounded Gemini response, not a classic search-API page size.

When to use it

When you want Google-quality grounded answers and your platform already uses Google Cloud identity.

Customized API Search

Customized Search API allows you to connect your custom implementation of the search engine, provided that it adheres to the search response schema accepted by the the web search tool.

Display name in the UI: Customized API

What it does

Sends your query to a REST endpoint you define and expects a response matching the Web Search results schema. This path runs from assistants-core (it is not routed through Search Proxy).

For GET, the query is added as a query parameter. For POST, it is included in the JSON body as {"query": "..."}. The response must be JSON with a results array; each item needs url, title, snippet, and optionally content.

Settings

UI Label	Field	Type	Default
Number of search results to fetch	`fetch_size`	integer	`5`
API Endpoint	`api_endpoint`	string	example placeholder
API Request Method	`api_request_method`	`GET` / `POST`	`GET`
API Headers	`api_headers`	string (JSON)	`{"Content-Type": "application/json"}`
API Additional Query Params	`api_additional_query_params`	string (JSON)	`{}`
API Additional Body Params	`api_additional_body_params`	string (JSON)	`{}`
Search Engine Mode	`search_engine_mode`	`standard` / `agent`	`standard`
Requires Scraping	`requires_scraping`	boolean	`false`
Timeout	`timeout`	integer (seconds)	`120`

Platform env vars (CUSTOM_WEB_SEARCH_API_*) can lock some fields centrally — see Search Engine Setup.

When to use it

Internal or vendor search indexes that are not one of the built-in providers.

Does this engine return content directly?

Engine	Kind	Returns content directly	Notes
Google Search	Standard	No	Web Page Reader usually runs afterward
Brave Search	Standard	Often snippets only	Reader may still help for full pages
Perplexity	Standard	Depends on context size / token settings	Tunable extraction
Customized API	Standard	Depends on your endpoint	Set Requires Scraping if URLs-only
Grounding with Bing	Grounding	Yes (default)	Configurable via Requires Scraping
Grounding with VertexAI	Grounding	Yes (default)	Configurable via Requires Scraping

Search Mode Configuration
Web Page Reader Configuration
Architecture: Search Engines
Infrastructure: Activation Reference, Search Engine Setup

Search Engine Configuration

Parameter exposure (Standard search only)

Common setting

Standard search

Google Search

What it does

Settings

When to use it

Brave Search

What it does

Settings (high level)

When to use it

Perplexity

What it does

Settings (high level)

When to use it

Grounding (agent search)

Grounding with Bing

What it does

Settings

When to use it

Grounding with VertexAI

What it does

Settings

When to use it

Customized API Search

What it does

Settings

When to use it

Does this engine return content directly?

Related pages