Web Search Setup

What is Web Search?

Web Search gives AIDA the ability to look things up on the internet — news, prices, sports scores, weather, research, and anything else that changes over time.

Without web search, AIDA relies only on its training data, which has a knowledge cutoff date. With web search, AIDA always has access to up-to-date information, and can also read a web page you link.

Two ways AIDA can search

Open Settings → Web search. The card has two parts, top to bottom.

1. Built-in provider search (on by default)

Some model providers can search the web themselves, inside the model's turn. AIDA supports this for Anthropic, Google (Gemini 3 and newer models), OpenAI, OpenRouter, xAI and Ollama Cloud. When the toggle is on and your current model comes from one of those providers, AIDA uses that provider's own search — no extra account, no extra key.

• With Anthropic, OpenAI, OpenRouter and xAI the search runs on the provider's side and is billed by that provider, per search, on the key you already use for the model. Check the provider's pricing page for the current rate. Answers end with a Sources list (with OpenAI and xAI the sources appear as links inside the answer text itself, so the footer may be empty — that is normal).
• With Google, Gemini 3 and newer models search through Google Search inside their turn, billed by Google per search on your Gemini key (Google includes a free monthly allowance — see Google's pricing). Answers end with a Sources list, and Google's Search Suggestions box appears under the answer: a row of chips with the searches Google ran. The box is required by Google and cannot be hidden; tap a chip to open that search on Google. This works in the web app only — a Gemini chat through Telegram, Discord or WhatsApp uses the search engine below instead, because those apps cannot show the box. Gemini 2.5 and older models cannot combine Google Search with AIDA's tools, so they also use the search engine below.
• With Ollama Cloud the search is free: AIDA uses Ollama's own web search with the same key you chat with. Results carry a (provider: Ollama Cloud) header. Nothing to set up.
• The card tells you whether built-in search is available on your current provider and model. If it says it is not, AIDA uses the search engine below instead.

2. Search engine (pick one)

The search engine is used by providers and models that have no built-in search — Gemini 2.5 and older, and local models — by Gemini chats through Telegram, Discord or WhatsApp, and by everyone when the built-in toggle is off. You choose one:

Only the selected engine runs. If it fails, AIDA tells you plainly that the search failed and why; it does not quietly switch to another engine. Each engine row has its own key or URL field, so you can keep both saved and switch between them later. If you only ever use providers with built-in search, select No search engine (the first row) — your saved keys stay saved.

Recommendation: leave Built-in provider search on. If you chat through Anthropic, OpenAI, OpenRouter, xAI, Ollama Cloud or a Gemini 3/4 model in the web app, you are done. Pick an engine if you use Gemini 2.5, a local model, or Gemini from a messaging app.

Option A — Tavily

1. Go to app.tavily.com and sign up — free, no credit card required.
2. Copy the API key from the dashboard — it starts with tvly-.
3. Open AIDA → Settings → Web search, select Tavily and paste the key.
4. Click Save, then Test search.

The free plan is 1,000 searches a month and resets monthly.

Option B — SearXNG (self-hosted)

SearXNG is a free, open-source meta-search engine you run yourself. No API key, just a URL. Note that many of the engines SearXNG queries treat server traffic as a bot; in practice a plain install answers mostly from Bing and Wikipedia. It works, but it is the advanced option.

Quick deploy with Docker:

docker run -d --name searxng -p 8080:8080 searxng/searxng

Your instance is then at http://your-server-ip:8080. The JSON output format must be enabled in settings.yml (the Test search panel tells you if it is not).

1. Open AIDA → Settings → Web search, select SearXNG and paste the URL.
2. Click Save, then Test search.

Reading web pages

When you give AIDA a link — "Read https://example.com/article and summarise it" — it fetches the page and works from its text. This uses Ollama Cloud's page reader and needs an Ollama Cloud key: the one you use for Ollama Cloud models, if you have it as a provider, or a key from ollama.com/settings/keys (free with an ollama.com account) saved under Settings → Providers → Ollama Cloud. It works whatever engine you selected and whether or not built-in search is on. If no key is saved, AIDA tells you so.

Test search

Test search runs one real query through your selected engine and shows the result, then checks every engine you have a key or URL for: reachable or not, how long it took, and — for SearXNG — how many results came back and which of its underlying engines are currently failing. Individual SearXNG engines failing is normal; the search still succeeds as long as some answer. The first line of the panel names the selected engine.

Using Web Search

Once configured, AIDA searches automatically when a question needs current information. You can also ask explicitly:

• "Search the web for the latest news on AI"
• "Look up the current price of Bitcoin"
• "What were the Oilers' recent results?"
• "Read this page and give me the key points: <link>"

Upgrading from an earlier version

• xAI (Grok) models now search by themselves when Built-in provider search is on. xAI runs the search inside the model's turn, billed by xAI per search on your xAI key (see x.ai for the rate); sources appear as links inside the answer. Nothing to set up. Since 0.6.83.
• Gemini 3 and newer models now search by themselves in the web app when Built-in provider search is on. Nothing to set up; Google's Search Suggestions box appears under grounded answers. Gemini 2.5 keeps using your engine.
• No search engine is a row on the card again, so you can switch back to built-in-only after trying an engine. Google's Search Suggestions row scrolls sideways with the mouse wheel on desktop.
• Gemini 3 and newer models: tools work again. Earlier versions could show "The model returned an empty response" whenever a Gemini 3 model used a tool (weather, reminders, email, search). Fixed in 0.6.82.
• Built-in provider search is on by default. If you had switched it off before, it stays off.
• Ollama Cloud is no longer an engine you pick — it is built in. If you chat through Ollama Cloud models, searches now run through Ollama's web search by themselves, free. If you had Ollama Cloud selected as your engine, AIDA now uses SearXNG if you have a URL saved, otherwise Tavily if you have a key, otherwise no engine — until you pick one yourself. Your saved keys are kept.
• The old Both option is gone. There is no automatic fallback between engines. One engine, picked by you, answers or reports its failure.

Troubleshooting

"Web search not configured" — Either turn on Built-in provider search (if your provider supports it) or select an engine and save its key or URL. A selected engine with no key shows needs a key on the card.

"Web search failed (Tavily): invalid key" — Re-copy the key from app.tavily.com.

"Web search failed (Ollama Cloud): invalid key" — This is the Ollama Cloud key under Settings → Providers. Keys live at ollama.com/settings/keys.

"Ollama Cloud rate limit reached" — The free tier has a usage limit; wait a little and try again.

"Rate limit" on Tavily — You have used this month's 1,000 free searches. Wait for the monthly reset, upgrade your plan, or switch the engine.

Built-in search is on but AIDA used the engine — Your current model has no built-in search (Gemini 2.5 and older, local models), or you are chatting with a Gemini model through Telegram, Discord or WhatsApp. The card says which providers support it.

No Sources footer with OpenAI or xAI — Expected: both place their citations as links inside the answer.

"LLM streaming error (xai): 403 … used all available credits or reached its monthly spending limit" — Your xAI team has no credits left, or hit its monthly limit. Add credits or raise the limit in the xAI console; chats and built-in searches resume at once.

"LLM streaming error (google): 402 … prepayment credits are depleted" — Your Gemini key's prepaid balance is empty. Top up in Google AI Studio under the project's billing.

No Google Search Suggestions box under a Gemini answer — The answer did not use Google Search (the model decides per question; ask for something that needs today's information), or you are on Gemini 2.5, or the chat came through a messaging app.

SearXNG not responding — Check the container: docker ps | grep searxng; restart with docker restart searxng. A 403 in Test search means the JSON format is not enabled in settings.yml.