Specialized Search Providers
Connect agents to external web and research search services including Perplexity, Tavily, Exa, Brave, You.com, Harvey, and custom providers.
Overview
Specialized Search Providers give agents access to live web search and AI-synthesized research beyond the knowledge base. Unlike Azure AI Search (which queries your internal documents), specialized search providers call external APIs to retrieve up-to-date information from the web.
Once configured, a provider appears as a tool agents can call during task execution - for example, searching for current news, looking up competitor pricing, or synthesizing research from multiple sources.
Supported Providers
| Provider | Type | Best For |
|---|---|---|
| Perplexity | AI-synthesized answers | Research questions needing cited summaries |
| Tavily | Web search | Factual lookups with source links |
| Exa | Neural search | Semantic similarity-based web retrieval |
| Brave | Web/news search | Privacy-focused web search |
| You.com Research | AI research | Cited answers across 5 effort tiers |
| You.com Search | Web/news search | Raw results with optional live page crawl |
| Harvey | Legal research | Specialized legal document retrieval |
| Custom | HTTP endpoint | Any search API behind a custom endpoint |
Adding a Specialized Search Provider
Via the Setup Wizard (Recommended)
The setup wizard walks you through selecting a provider, entering credentials, and creating the search tool in one flow.
- Navigate to Configuration > Tools & Integrations > Specialized Search
- Click Add Provider to open the setup wizard
- Select your provider from the card grid
- Enter your API key and provider-specific settings
- Click Test Connection to verify credentials
- Click Save - the search tool is created automatically
Via the Form Page
For full access to all configuration options, use the form page directly:
- Navigate to Configuration > Tools & Integrations > Specialized Search
- Click New Provider (form view)
- Fill in the name, display name, provider type, and credentials
- Configure provider-specific fields (see sections below)
- Click Test Connection, then Save
You.com Research
You.com Research returns AI-synthesized answers with citations, drawn from live web sources. It supports five effort tiers that trade cost and latency for answer depth.
Configuration Fields
| Field | Required | Description |
|---|---|---|
| API Key | Yes | Your You.com API key |
| Effort Tier | Yes | lite, standard, deep, exhaustive, or frontier |
| Include Domains | No | Comma-separated domains to restrict results (e.g., fda.gov, nih.gov) |
| Exclude Domains | No | Comma-separated domains to filter out |
| Freshness | No | Recency filter: any, past day, past week, past month, past year |
| Country | No | ISO 3166-1 alpha-2 country code (e.g., US) |
| Output Schema | No | JSON Schema for structured outputs (ignored on lite tier) |
Effort Tiers
| Tier | Speed | Cost | Use When |
|---|---|---|---|
lite | Fast | Lowest | Quick fact checks, simple questions |
standard | Moderate | Low | General research tasks |
deep | Slower | Medium | Thorough research with multiple sources |
exhaustive | Slow | High | Comprehensive reports requiring broad coverage |
frontier | Slowest | Highest | Maximum depth, complex multi-step reasoning |
Structured output via Output Schema is not available on the
litetier.
You.com Search
You.com Search returns raw web and news search results. An optional livecrawl feature fetches the full page content at query time.
Configuration Fields
| Field | Required | Description |
|---|---|---|
| API Key | Yes | Your You.com API key |
| Result Count | No | Number of results to return (1-100, default 10) |
| Freshness | No | Recency filter: any, past day, past week, past month, past year |
| Country | No | ISO 3166-1 alpha-2 country code (e.g., US) |
| Language | No | BCP 47 language tag (e.g., en-US) |
| Safesearch | No | off, moderate, or strict |
| Livecrawl | No | off, always, or fallback |
| Livecrawl Formats | No | Comma-separated subset of html, markdown (active when livecrawl is on) |
| Crawl Timeout | No | Seconds before livecrawl times out per page |
| Include Domains | No | Comma-separated domains to restrict results |
| Exclude Domains | No | Comma-separated domains to filter out |
Livecrawl Costs
Enabling livecrawl causes ExecAssist to fetch the full contents of each result page at search time. This adds approximately $0.001 per page returned and increases response latency. Use fallback mode to crawl only when cached content is unavailable, or always to force fresh crawls on every search.
The Livecrawl Formats field is disabled when livecrawl is set to
off.
Exa
Exa is a neural search engine over a curated web index. It is optimized for semantic, similarity-based retrieval rather than keyword ranking, making it a strong choice when agents need to find pages that are conceptually related to a query rather than pages that contain exact words.
Configuration Fields
| Field | Required | Description |
|---|---|---|
| API Key | Yes | Your Exa API key |
| Result Count | Yes | Number of results to return (1-10, default 10) |
| Search Type | Yes | auto (Exa picks), neural (embedding-based), or keyword (traditional matching) |
| Use Autoprompt | No | Let Exa rewrite the query for better neural recall (recommended for neural type) |
| Category | No | Restrict results to a content category (see table below) |
| Include Domains | No | Comma-separated domains to restrict results (e.g., nytimes.com, bbc.com) |
| Exclude Domains | No | Comma-separated domains to filter out |
| Start Crawl Date | No | Only return pages first crawled on or after this date (YYYY-MM-DD) |
| End Crawl Date | No | Only return pages first crawled on or before this date (YYYY-MM-DD) |
| Start Published Date | No | Only return pages published on or after this date (YYYY-MM-DD) |
| End Published Date | No | Only return pages published on or before this date (YYYY-MM-DD) |
| Include Page Text | No | Fetch the full page text with each result |
| Max Characters | No | Maximum characters of page text to return per result (1-100,000, default 1,000) |
| Include HTML Tags | No | Keep HTML tags in the returned page text |
Search Types
| Type | Description | Use When |
|---|---|---|
auto | Exa selects neural or keyword based on the query | General use - good default |
neural | Embedding-based semantic retrieval | Finding pages semantically similar to a concept |
keyword | Traditional keyword matching | Exact phrase lookups, proper nouns |
Content Categories
Restrict results to a specific content type by setting Category:
| Category | Description |
|---|---|
company | Company home pages and profiles |
research paper | Academic and scientific papers |
news | News articles |
linkedin profile | LinkedIn user profiles |
github | GitHub repositories and code |
tweet | Twitter/X posts |
movie | Movie pages |
song | Song and music pages |
personal site | Personal websites and blogs |
pdf | PDF documents |
financial report | Annual reports, 10-Ks, investor documents |
Leave Category blank to search across all content types.
Page Text and Cost
Enabling Include Page Text causes ExecAssist to fetch the full text of each result page. This increases response latency and may affect Exa billing. Use Max Characters to cap the text size per result when full pages are not needed. Set Include HTML Tags only if downstream processing requires structured HTML.
Getting Your Exa API Key
- Sign up at exa.ai and go to your developer dashboard
- Create an API key under API Keys
- Copy the key and paste it into the API Key field in ExecAssist
Getting Your You.com API Key
- Sign up or log in at you.com and navigate to your developer dashboard
- Create an API key under API Access
- Copy the key and paste it into the API Key field in ExecAssist
Both You.com Research and You.com Search use the same API key.
Tavily
Tavily is a web search API built for AI agents. It returns factual results with source links and optional AI-extracted answers, without the synthesized narrative of Perplexity or You.com Research. Pick Tavily when agents need fast, reliable factual lookups with transparent source attribution and a straightforward per-call pricing model. For LLM-synthesized cited prose, prefer Perplexity or You.com Research instead; for semantic similarity retrieval, prefer Exa.
Getting Your Tavily API Key
- Sign up at tavily.com and verify your email
- Open the Tavily dashboard and navigate to API Keys
- Copy your key (it begins with
tvly-) - Paste the key into the API Key field in ExecAssist
Configuration Fields
| Field | Required | Description |
|---|---|---|
| API Key | Yes | Your Tavily API key (tvly-...), sent as api_key in the JSON request body |
| Search Depth | No | basic (faster, lower cost) or advanced (deeper crawl, higher cost). Default basic |
| Topic | No | general or news - scopes the index queried. Default general |
| Result Count | No | Number of results to return (1-20, default 5) |
| Include Answer | No | When enabled, Tavily appends an AI-generated direct answer above the source list |
| Include Raw Content | No | When enabled, each result includes the full page text |
| Include Images | No | When enabled, results include related image URLs |
| Include Domains | No | Comma-separated domains to restrict results (e.g., reuters.com, apnews.com) |
| Exclude Domains | No | Comma-separated domains to filter out |
| Days | No | For news topic only - limits results to pages published within the last N days |
Pricing
Tavily bills on API credits. One credit equals one search call.
| Plan | Credits Included | Approx. Cost per Call | Notes |
|---|---|---|---|
| Free | 1,000 credits/month | $0 | Non-commercial evaluation only |
| Starter | 10,000 credits/month | ~$0.004 | Light production workloads |
| Growth | 50,000 credits/month | ~$0.002 | Volume discount |
| Enterprise | Custom | Custom | Dedicated support, SLAs |
advanced search depth consumes more credits per call than basic. Enabling Include Raw Content also increases cost because Tavily fetches and trims full page text server-side. Confirm current rates on the Tavily pricing page before committing to high-volume usage.
Gotchas
- The Days field is only respected when Topic is set to
news. Setting it on ageneralsearch has no effect. - Include Answer and Include Raw Content add latency as well as credit cost. Leave them off unless the agent needs to reason over full page content or wants a direct answer string.
- Tavily does not enforce domain-level authentication. Any valid API key can query any public URL, so rotate keys if they are exposed.
- The free tier is for non-commercial evaluation. Move to a paid plan before pointing production agents at Tavily.
- Include Domains and Exclude Domains are mutually exclusive in practice - combining both on the same provider narrows results unpredictably. Use one or the other.
- Tavily's index skews toward English-language content. For non-English research tasks, evaluate result quality before committing.
Exa
Exa is a neural search engine that retrieves results by semantic similarity rather than keyword match. Pick this provider when agents need to discover content based on meaning - "find pages like this one", research-grade discovery across curated web content, or category-scoped lookups (research papers, GitHub repos, company sites). For LLM-synthesized cited answers, prefer Perplexity or You.com Research instead; for raw keyword web search, prefer Brave or Tavily.
Configuration Fields
| Field | Required | Description |
|---|---|---|
| API Key | Yes | Your Exa API key, sent as x-api-key on every request to https://api.exa.ai/search |
| Result Count | No | Number of results to return (1-10, default 10) |
| Search Type | No | neural (semantic), keyword (lexical), or auto (Exa picks per query). Default auto |
| Use Autoprompt | No | When true, Exa rewrites the query for better neural retrieval. Default true. Only meaningful for neural or auto |
| Category | No | Restrict results to a content category: company, research paper, news, linkedin profile, github, tweet, movie, song, personal site, pdf, or financial report |
| Include Domains | No | Comma-separated domains to restrict results (e.g., arxiv.org, nature.com) |
| Exclude Domains | No | Comma-separated domains to filter out |
| Start Crawl Date | No | ISO 8601 lower bound for when Exa first crawled the page |
| End Crawl Date | No | ISO 8601 upper bound for crawl date |
| Start Published Date | No | ISO 8601 lower bound for when the page was published |
| End Published Date | No | ISO 8601 upper bound for publish date |
| Return Page Text | No | When enabled, each result includes the page text. Adds per-result cost (see below) |
| Max Characters | No | Cap on returned text length per result (active when Return Page Text is on) |
Pricing
Exa charges per search and adds an extra per-result fee when page text is requested. Approximate rates:
| Component | Approx. Cost | When Charged |
|---|---|---|
| Search call | ~$0.005 per query | Every /search request |
| Result with page text | ~$0.001 per result | Only when Return Page Text is on |
A typical call with Return Page Text off costs about half a cent. With page text on and Result Count at 10, expect roughly 1.5 cents per call. Confirm current rates on the Exa pricing page before committing to high-volume usage. Exa offers a free monthly allotment for evaluation; paid plans bill on usage above that.
Gotchas
- Exa is stateless. ExecAssist's multi-turn search flow re-issues a fresh
/searchcall on each turn rather than continuing a session, so per-call cost multiplies withMaxTurnsPerSession. - Use Autoprompt has no effect on the
keywordsearch type. Disable it explicitly if you want the agent's exact query passed through verbatim on neural runs. - Return Page Text adds latency as well as cost - Exa fetches and trims each page server-side before responding. Leave it off for fast title/snippet lookups; turn it on only when the agent needs to reason over full content.
- Start/End Crawl Date and Start/End Published Date filter different signals. Crawl date is when Exa indexed the page; published date is when the source claims the page was authored. They can diverge by months for older content.
- Exa does not return an LLM-synthesized answer. The tool wraps results into a Markdown digest client-side, but agents that need cited prose should prefer Perplexity or You.com Research.
- The Exa
/searchendpoint covers thetextcontent option used here. Other Exa endpoints (/findSimilar,/contents) are out of scope for this provider type.
Getting Your Exa API Key
- Sign up at exa.ai and verify your email
- Open the Exa dashboard and navigate to API Keys
- Create a new key and copy it
- Paste the key into the API Key field in ExecAssist
Brave
Brave Search is a privacy-focused web and news index. Pick this provider when agents need fast factual lookups with broad coverage and a low per-call price, and when the answer should come from raw web results rather than an AI-synthesized summary. For research-style cited answers, prefer Perplexity or You.com Research instead.
Configuration Fields
| Field | Required | Description |
|---|---|---|
| API Key | Yes | Your Brave Search API subscription token, sent as X-Subscription-Token on every request |
| Endpoint | Yes | web or news - selects the Brave Search index queried |
| Result Count | No | Number of results to return (1-20, default 10) |
| Country | No | ISO 3166-1 alpha-2 country code (e.g., US) |
| Search Language | No | Two-letter language code for results (e.g., en) |
| UI Language | No | BCP 47 language tag for localized result strings (e.g., en-US) |
| Safesearch | No | off, moderate, or strict (default moderate) |
| Freshness | No | Recency filter: pd (past day), pw (past week), pm (past month), py (past year), or a YYYY-MM-DDtoYYYY-MM-DD range |
| Extra Snippets | No | Return additional excerpt strings per result (Pro tier and above) |
| Goggles ID | No | A custom Goggle URL or ID that re-ranks results against a community ranking profile |
Pricing Tiers
| Tier | Approx. Cost | Rate Limit | Notes |
|---|---|---|---|
| Free | $0 | 1 query/sec, 2,000 queries/month | Non-commercial use only; good for evaluation |
| Data for Search Base | ~$3 per 1,000 queries ($0.003/call) | 20 queries/sec | Standard web/news results |
| Data for Search Pro | ~$5 per 1,000 queries ($0.005/call) | 50 queries/sec | Adds extra snippets, summarizer access, Goggles |
| Data for AI Pro | ~$9 per 1,000 queries ($0.009/call) | High | Includes data rights for LLM grounding and RAG pipelines |
Pricing changes occasionally; confirm current rates on the Brave Search API pricing page before committing to a tier.
Gotchas
- The free tier prohibits commercial use. Move to a paid tier before pointing production agents at it.
- Extra Snippets and the Brave summarizer require the Pro tier or higher; the field is accepted on Base but ignored.
- The
webandnewsendpoints are mutually exclusive on a single provider entry. Configure two providers if agents need both. - Brave's index is global, but result quality varies by region. Set Country and Search Language when answers should reflect a specific market.
- Each subscription token is bound to one tier. Rotating from Free to Base or Pro requires generating a new token in the dashboard.
Getting Your Brave API Key
- Sign up at brave.com/search/api and verify your email
- Open the API dashboard and pick a subscription tier
- Generate a subscription token under API Keys
- Copy the token and paste it into the API Key field in ExecAssist
Managing Providers
Viewing Providers
The Specialized Search list shows all configured providers with their type, connection status, and the date they were added.
Editing a Provider
- Click on the provider name in the list
- Update settings as needed
- Click Test Connection to verify the updated credentials
- Click Save
Deleting a Provider
- Remove any agent tools that reference this provider first
- Click the delete icon next to the provider
- Confirm deletion
Deleting a provider removes its associated search tool from all agents that use it.
Troubleshooting
Connection Test Fails
| Symptom | Likely Cause | Solution |
|---|---|---|
401 Unauthorized | Invalid API key | Re-copy the key from your provider dashboard |
403 Forbidden | Subscription limit | Verify your plan includes API access |
Timeout | Network or rate limit | Retry; check provider status page |
Search Returns No Results
- Verify the query is supported by the selected provider
- Check that domain filters are not too restrictive
- For You.com Research, try a lower effort tier to confirm basic connectivity
Livecrawl Not Returning Page Content
- Confirm livecrawl is set to
alwaysorfallback(notoff) - Check that the crawl timeout is sufficient (some pages load slowly)
- Some pages block crawlers; this is expected behavior
FAQ
Can I add both You.com Research and You.com Search with the same API key? Yes. Create separate providers for each type - they share the same API key but have different configuration options and produce different tool behaviors.
Which You.com tier should I start with?
Start with standard. It balances cost, speed, and answer quality for most research tasks. Move to deep or exhaustive only when agents need comprehensive coverage.
Does livecrawl affect my You.com API rate limits? Livecrawl is billed separately by You.com per page crawled and does not count against your search query limits. Check your You.com plan for crawl pricing.