MyBenAI's optional web search lets your local AI find fresh answers on the internet, cached for 30 days. Use your own Serper API key or fall back to DuckDuckGo—all results stay on your phone with citations to original sources, giving you the best of both worlds: offline AI with current information.
Why Local AI Needs Web Search
A language model on your phone has knowledge baked in from training data, but that knowledge has a cutoff date. Qwen 3.5, Llama 3.2, Gemma 3—all trained months or years ago. If you ask about current events, new research, product prices, or real-time information, the model has no answer. It's not stupid; it's just offline.
That's where web search comes in. MyBenAI can optionally search the internet for current information, grab the results, and synthesize them into an answer—all while keeping everything local and under your control. You're not sending queries to OpenAI or Google. You're using your own API key (if you choose to) or falling back to public search, then storing results on your device.
How It Works: Serper to DuckDuckGo Fallback
MyBenAI uses a two-tier search strategy. If you provide a Serper API key, the app uses Serper's search results—fast, comprehensive, and accurate. Serper is a search API aggregator; you control it, not MyBenAI. Your searches go to Serper's servers, not ours.
If you don't have a Serper key, MyBenAI falls back to DuckDuckGo. DuckDuckGo is public, doesn't track users, and provides search results anonymously. The fallback means you always have a way to search, even without an API key. No paywalls, no forced signup.
Either way, once search results arrive on your phone, they stay there. They're cached locally in your device's private storage.
The 30-Day Cache: Why It Matters
Every search result is stored for 30 days. Here's why that's powerful: most questions you ask repeat. You search "best TypeScript linters" once, then search again in two weeks because you forgot the recommendation. With caching, that second search hits your local database—no network call, no API cost, instant answer.
For Serper users, this dramatically reduces API bills. Serper charges per query. If 40% of your searches are repeats within 30 days (a conservative estimate for typical use), you're cutting your API costs in half. For DuckDuckGo users, caching means you can re-summarize cached results offline without needing the internet—useful if you're on a plane or in an area with spotty signal.
After 30 days, results expire and are removed from your phone automatically. This keeps your cache lean and refreshes information periodically.
Citations: Know Where Information Comes From
When MyBenAI generates an answer from web search, it includes citations. Not vague citations—concrete ones. Each cited fact links to the URL where it appeared and shows the snippet from the original page.
For example, if you ask "What's the current price of Ethereum?", MyBenAI searches, retrieves results, and responds: "Ethereum is trading at $3,847 (as of today)." Then it cites the source: "CoinGecko: $3,847, retrieved from coinmarketcap.com." You can tap the citation and see the exact snippet where the data came from.
This is especially useful for research. Combine web search with MyBenAI's multi-agent research for cited answers, and your assistant becomes a fact-checking machine. No hallucinations—every claim is traced to a source on your phone.
Privacy: You Control Your API Key
If you use Serper, you own the API key. You set the monthly budget, monitor usage in Serper's dashboard, and can revoke it anytime. MyBenAI never stores your key on its servers; it stays in your app's encrypted storage (MMKV). Your Serper account is yours; your search history is yours.
If you use DuckDuckGo fallback, searches go to DuckDuckGo, which doesn't log searches by user (it's their privacy promise). You're not creating an account or logging in to MyBenAI to search.
Message content never leaves your device. Your questions are sent to Serper or DuckDuckGo only if you explicitly trigger a search. Then the search query itself is sent (necessary to get results), but the LLM's processing—your follow-up questions, your reasoning, your chat history—stays local.
Re-Summarizing Offline: The Cache Advantage
Here's a subtle but powerful feature: after you've searched something and cached the results, you can ask follow-up questions about that search entirely offline.
For example: You're on a flight with no internet. Earlier, you cached search results for "best Python frameworks 2025." Now, offline, you ask your AI, "Which of these is fastest for real-time APIs?" The model can re-read the cached results and answer. No network call needed. The cache acts like a personal knowledge base that you built with web searches.
This is different from cloud AI. With ChatGPT or Claude, once you're offline, you can't do web searches. With MyBenAI, if the data was recent enough and cached, you work fully offline.
Search Controls and Privacy Settings
You can disable web search entirely if you prefer fully offline AI. Search results are visible in your network log, so you can audit every search your AI made: the query, the URL it hit, the timestamp. You stay in control—transparency is built in.
Trade-Off: Cache Staleness and Internet Dependency
The 30-day cache is a double-edged sword. If you're looking for real-time data (stock prices, live sports scores), cached results can be stale. A cached result from day 20 might be outdated by day 25. You're trading freshness for savings and offline capability.
Additionally, without internet, you can't search new topics—you can only re-query cached results. If you're traveling and need information that wasn't cached before, you'll need to find a coffee shop with WiFi to add new searches to your cache.
For most research and reference queries, 30 days is plenty. For real-time data, you'll want to refresh occasionally.
Fitting Into Your Workflow
Web search integrates seamlessly with MyBenAI's other features. Use it with multi-agent research to synthesize multiple search results. Combine it with function calling to fetch live data (crypto prices, weather) alongside web search results. Citations work automatically with sub-agents, so your fact-checker agent can verify sources.
Getting Started
To enable web search, open MyBenAI settings and toggle web search on. If you have a Serper key, paste it in. Otherwise, MyBenAI uses DuckDuckGo by default. Then, in any chat, if you ask a question that would benefit from current information, MyBenAI will search and cache results. You'll see a citation next to each piece of sourced information.
Web search is entirely optional—you can use MyBenAI as a fully offline AI if you prefer. But if you want fresh information cached on your phone for 30 days, with citations and offline re-summarization, web search turns your local AI into a powerful research assistant. Ready to try it? Get MyBenAI and enable web search today. Learn more about how research assistants combine agents and search for comprehensive answers.