Skip to main content

Web Search

DevoxxGenie can ground answers in live web results, so the LLM can use information that is newer than its training data.

Overview​

Web search:

  1. Sends your search query to the web search provider you selected
  2. Retrieves the most relevant results (title, URL and snippet)
  3. Hands those results to the LLM, which answers from them
  4. Helps with anything that changed after the model's knowledge cutoff

It is particularly useful for:

  • Technical questions about recent framework versions
  • Documentation lookups for APIs and libraries
  • Current best practices and patterns
  • Specific error messages

There are two ways to use it:

WhereHow
/search commandType /search <question> in the prompt. DevoxxGenie searches, then answers from the results.
Agent Mode web_search toolThe agent decides by itself when to search, and gets the raw results (title, URL, snippet) to reason over.

Both use the same provider and the same credentials, chosen once in Settings → DevoxxGenie → Web search.

Choosing a Provider​

Web search uses one provider at a time. Pick it in the Provider dropdown:

ProviderWhat you needNotes
TavilyA Tavily API keySearch API built for LLMs.
KeenableNothing (an API key is optional)Works without signing up; free searches are rate-limited per network. Add a key for higher limits. Available after v1.18.0.
ParallelA Parallel API keyWeb search API built for AI agents; billed to your Parallel account. Available after v1.18.0.
Google Custom Search (deprecated)A Google API key and a Programmable Search Engine IDClosed to new customers; stops working on January 1, 2027. See Google Custom Search is deprecated.
None—Web search is off for /search and the agent tool.

New to web search? Keenable needs no account; Tavily and Parallel need an API key.

Only the selected provider's fields are shown. Keys you entered for other providers are kept, so you can switch back without typing them again. Keys are stored in IntelliJ's credential store (PasswordSafe), not in the settings XML.

Upgrading from an earlier version

Earlier versions had a separate checkbox per provider. When you upgrade, DevoxxGenie keeps your choice: if Tavily was enabled it is selected, otherwise Google if it was enabled, otherwise None. Your keys are not touched.

Search services with an MCP server​

Any search service that offers an MCP server can also be used in Agent Mode by adding it in Settings → DevoxxGenie → MCP. It then works as an agent tool, but not for /search, Test search or the provider choice above.

Google Custom Search is deprecated​

warning

Google has closed its Custom Search JSON API to new customers and discontinues it on January 1, 2027. In Google's words: "Existing Custom Search JSON API customers have until January 1, 2027 to transition to an alternative solution." See Google's Custom Search JSON API overview.

What this means in DevoxxGenie:

  • New setups: choose Keenable (no account needed) or Tavily. Google is listed last, marked (deprecated), and is never selected by default.
  • Already using Google: it keeps working until January 1, 2027. While it is selected, Settings → Web search shows a warning with the date. Switch to Keenable or Tavily before then.
  • When Google fails: /search, the agent's web_search tool and Test search explain that the API is being (or has been) discontinued and suggest another provider, instead of showing a raw error.
  1. Open Settings → DevoxxGenie → Web search
  2. Check Enable feature
  3. Choose a Provider
  4. Enter that provider's credentials (see below)
  5. Optionally set Max search results (1–10, default 5)
  6. Type a query under Test web search and press Enter or click Test search to check the setup. The test uses what is in the form, so you can try a key before you click Apply. A status line shows which provider answered, how many results came back and how long it took, or a clear error, such as a rejected or missing key. Each result is listed with its title and site. Select a result to see its cleaned-up snippet and full URL, and double-click it or click the URL to open the page in your browser.
  7. Click Apply

Web Search Settings

Tavily Setup​

  1. Sign up at app.tavily.com and copy your API key
  2. Select Tavily as the provider
  3. Paste the key into Tavily API key

Keenable Setup​

Keenable works with no setup at all:

  1. Select Keenable as the provider
  2. Click Test search to check it, then Apply

Without a key, DevoxxGenie uses Keenable's free public search. It is rate-limited per network, so on a shared office or VPN connection you may hit the limit sooner. When that happens you'll see:

Keenable's free search limit for your network was reached. Try again later, or add a Keenable API key in Settings → DevoxxGenie → Web search for a higher limit.

To raise the limit, get a free API key from keenable.ai, which gives you 100K searches per month, and paste it into Keenable API key (optional). With a key, searches go to Keenable's authenticated endpoint. If the key is wrong, DevoxxGenie tells you it was rejected; clear the field to go back to free keyless search.

Requests identify themselves as DevoxxGenie. The API key is only ever sent to Keenable, never to another site after a redirect.

Parallel Setup​

  1. Create an API key at platform.parallel.ai
  2. Select Parallel as the provider
  3. Paste the key into Parallel API key, then click Test search and Apply

DevoxxGenie uses Parallel's REST Search API in its fast mode, which suits interactive searches and is the cheaper tier (Parallel pricing). Searches are billed to your Parallel account. If something goes wrong you get a specific message: a rejected key, an account that is out of credit, or the rate limit (try again in a minute).

Google Custom Search Setup​

note

Only for existing Google Custom Search JSON API customers: Google no longer accepts new ones, and the API stops working on January 1, 2027.

  1. Create a Google Cloud project and enable the Custom Search API
  2. Create an API key
  3. Create a Programmable Search Engine and copy its Search Engine ID
  4. Select Google Custom Search as the provider
  5. Enter both the API key and the Search Engine ID

What happens when something is missing​

DevoxxGenie checks the selected provider before it sends anything. If something is missing, no request is made and you get a message that says what to fix, for example:

Tavily is missing its API key. Enter it in Settings → DevoxxGenie → Web search.

Google Custom Search is missing its Search Engine ID. Enter it in Settings → DevoxxGenie → Web search.

No web search provider is selected. Choose one in Settings → DevoxxGenie → Web search.

The same check applies to /search, the agent's web_search tool and Test search.

The /search command​

/search is available when Enable feature is on and a provider is ready:

/search How to implement JWT authentication in Spring Boot 3.2?

DevoxxGenie sends the text after /search to your provider, then asks the LLM to answer from the results.

Sources​

Every /search answer ends with a Sources list: the title and original URL of each result the answer was based on, as clickable links. Links open in the browser configured in your IDE.

Use since-build in plugin.xml to declare the earliest compatible IDE build…

Sources
1. Build number ranges → https://plugins.jetbrains.com/docs/intellij/build-number-ranges.html
2. plugins.jetbrains.com → https://plugins.jetbrains.com/docs/intellij/plugin-configuration-file.html

When a result has no title, the site's host name is shown instead. Only http and https results are listed.

If the provider finds nothing, you get a clear message instead of an answer, and no LLM call is made:

No web results found for "…". Try different keywords, or check the provider in Settings → DevoxxGenie → Web search.

Use Cases​

Framework and Library Documentation​

/search What's the correct way to configure WebClient in Spring Boot 3.2?

Error Resolution​

/search Hibernate HHH90000022 legacy SequenceStyleGenerator warning: how do I fix it?

Best Practices and Patterns​

/search Current best practice for handling JWT authentication in React applications

Version-Specific Questions​

/search New features in Java 21 compared to Java 17

Agent Mode: web_search Tool​

DevoxxGenie also exposes web search as an agent tool when Agent Mode is active. The LLM decides by itself when a live web search is needed.

Enabling the tool​

  1. Open Settings → DevoxxGenie → Agent Mode
  2. Under Built-in Tools, check web_search
  3. Select a provider and enter its credentials in Settings → DevoxxGenie → Web search. The agent tool uses the same provider and keys as /search.

If the selected provider isn't ready, the Agent Mode settings show a warning with a link to the Web search settings. At run time the tool returns a descriptive error string, such as which key is missing, and the agent carries on with other tools. There are no modal dialogs and the agent loop doesn't break.

How it works​

When the agent decides a query needs external information, it calls web_search directly:

Found 3 results for "langchain4j ChromaDB 0.6 migration":

1. LangChain4j 0.37 release notes
URL: https://github.com/langchain4j/langchain4j/releases/tag/0.37.0
LangChain4j 0.37 updates the ChromaDB store to API v2 (0.6.x)...

2. ...

The tool returns raw structured results (title, URL and snippet), so the agent can reason over sources without an extra summarisation step. Only the selected provider is ever contacted. Keys stored for other providers are not used.

When a result looks promising, the agent can follow up with the fetch_page tool, which fetches the full content of a page given its URL. That makes web_search → fetch_page a natural two-step pattern for reading external documentation.

Like semantic_search, the web_search tool is classified as read-only, so it is auto-approved when the "Auto-approve read-only tools" setting is on.

The three-tier retrieval stack​

With Agent Mode, web search becomes one layer of a complete retrieval hierarchy:

ToolBest for
semantic_searchConceptual queries over your indexed codebase
search_filesExact-string or regex lookups in project files
web_searchDocumentation, release notes, external references

The LLM picks the right tier based on the query. You can enable or disable each tool independently under Settings → Agent Mode → Built-in Tools.

Web Search and RAG​

Web search complements the RAG feature:

  • RAG (semantic_search tool): retrieves information from your indexed project codebase
  • Web search: retrieves up-to-date information from the internet

To use both:

  1. Enable RAG, web search and Agent Mode for the full tool stack
  2. Ask questions that relate to both your code and external information
  3. DevoxxGenie combines insights from both sources

Best Practices​

Writing Effective Search Queries​

  1. Be specific: include version numbers, framework names and specific technologies
  2. Use technical terminology rather than general descriptions
  3. Ask one question at a time: multi-part queries tend to return less relevant results
  4. Mention recency when it matters (e.g. "latest approach")

Web search is most valuable when:

  1. You need information about recent software versions
  2. You're looking for current best practices
  3. You need help with specific error messages
  4. The information might have changed since the LLM's training cutoff

It is less useful for:

  1. Conceptual computer science questions
  2. Questions about your own codebase (use RAG instead)
  3. General programming patterns that haven't changed recently

Privacy Considerations​

  1. Search queries: the text you search for is sent to the selected provider, and only to that provider
  2. API keys: stored in IntelliJ's credential store (PasswordSafe)
  3. Data usage: check the selected provider's privacy policy for how it handles queries
  4. Code privacy: avoid putting sensitive code in search queries

Troubleshooting​

Common Issues​

/search says web search isn't configured​

  1. Check that Enable feature is on in Settings → Web search
  2. Check that a Provider is selected (not None)
  3. Read the message: it names the missing key or ID
  4. Use Test search to confirm the setup before applying

No Search Results​

If /search answers No web results found:

  1. Verify the key with Test search
  2. Make sure you haven't exceeded the provider's rate limits
  3. Make your query more specific or use different keywords

Poor Search Results​

  1. Make your query more specific
  2. Include version numbers and exact error messages
  3. Try the other provider
  4. Break complex questions into simpler ones

API Key Issues​

  1. Re-enter the key in Settings → Web search and run Test search
  2. Check that the key is active in your provider account (Tavily or Google Cloud)
  3. Ensure billing is set up if your provider requires it
  4. Check for restrictions on the key (IP, usage limits)

Advanced Configuration​

Search Settings​

The settings page exposes Enable feature, the Provider choice, the selected provider's credentials, and Max search results (1–10). It does not expose search-type, locale or Safe Search controls.

Custom Prompting​

You can steer how results are used by being explicit in your prompt:

/search Spring Boot 3.2 security configuration changes versus Spring Boot 2.x, focusing on OAuth2 resource server setup

Future Enhancements​

Planned improvements to web search:

  1. Context-aware searching: automatically incorporating project context into searches
  2. Search result caching: improving performance by caching common searches