Web Search: Real-Time Knowledge
Web Search: Real-Time Knowledge
Every Agenturo agent has web search built in. No setup, no API keys, no configuration. It's always available, and your agent can use it on any message where real-time information would help.
How It Works
When your agent needs current information, it runs a parallel search cascade:
- Jina Search — searches the web and returns full page content, not just snippets
- Brave Search — runs in parallel as a backup, providing snippets and links
- Deep fetch — if neither produces great results, the system can fetch and read a specific URL for full content
This all happens in under 15 seconds. Your visitors see a "searching..." indicator while it runs.
You Don't Need to Configure It
Web search is a default capability. Your agent will use it when:
- A visitor asks about something current (news, prices, recent events)
- A visitor asks about something not in the soul's knowledge
- The soul explicitly instructs the agent to search (like Jackie Check's pipeline)
You don't need to "enable" it or mention it in your soul — unless you want to control HOW the agent uses it.
The URL-Embedded Knowledge Pattern
This is the most powerful search pattern in Agenturo. Instead of trying to embed all product data in your soul, you instruct your agent to search your own website.
The Padre coffee agent (zoiberg.agenturo.app) does this:
<knowledge>
- For current coffee blends and prices, ALWAYS search padreandsons.com
- For brewing guides, search padreandsons.com/brewing
- For wholesale inquiries, direct to padreandsons.com/wholesale
</knowledge>
When a visitor asks "What coffees do you have?", the agent silently searches padreandsons.com, reads the current product listing, and responds with accurate, up-to-date information. No stale data. No manual updates.
When to use this pattern:
- Product catalogs that change frequently
- Pricing pages
- Event listings or schedules
- Any content that's already on your website
When NOT to use this pattern:
- Core identity facts (your name, role, mission) — these should be in the soul
- Information that's NOT on a public URL
- Answers that need to be instant (web search adds 5-15 seconds)
Tool Discipline: The Silent Search
The biggest quality issue with search-enabled agents is narration. Without explicit rules, your agent will do this:
"Let me search for that information for you. I'll look into the current coffee prices on the Padre website... I found several results. Based on what I found..."
That's 30 words of filler before any useful content. Fix it with tool discipline:
<output_format>
## TOOL DISCIPLINE
- Work silently. Never announce "Let me search for that."
- Deliver only results. Never narrate the search process.
- If search returns nothing useful, say "I couldn't find current information
on that" — don't explain what you tried.
</output_format>
With tool discipline, the same interaction becomes:
"We currently have three single-origin options: Ethiopian Yirgacheffe, Colombian Huila, and Kenyan AA. Blends include our Classic and Dark roast. Check padreandsons.com for current pricing and availability."
Self-Check Rules (For Expert Agents)
If your agent is a fact-checker or research agent, add self-check rules to prevent lazy searching:
<knowledge>
## SELF-CHECK RULES
- If your first search confirms the claim too easily, search for
counter-evidence before delivering a verdict
- If sources disagree, report the disagreement — don't pick a side
- If a claim is about the last 24 hours, flag recency uncertainty
- Cross-reference: at least 2 independent sources for any verdict
These rules from Jackie Check (check.agenturo.app) prevent the agent from accepting the first search result at face value — a critical quality control for any agent making factual claims.
Search Limitations
- Speed: Search adds 5-15 seconds to response time. For questions your agent should answer instantly, put the knowledge in the soul.
- JavaScript-heavy sites: Some sites don't render well for content extraction. If your website is a heavy single-page app, the fallback layers (Readability, Cloudflare, Brave) handle most cases.
- Paywalled content: Search can't access content behind paywalls or login walls.
- Rate limits: The system manages rate limits automatically, but extremely high traffic agents may hit throttling during peak usage.
Making Your Agent Search Smarter
You can influence search behavior through your soul without any technical configuration:
- Name specific domains: "For pricing, search example.com" is better than "search the web for pricing"
- Define when to search: "Only search when the visitor asks about current data" prevents unnecessary searches on simple questions
- Set search scope: "Only reference information from official medical sources" narrows results for expert agents
- Require verification: "Cross-reference at least 2 sources" raises the quality bar