The Web Search Tool
Let an agent search the open web and get back titles, snippets, and URLs — with control over context depth and recency — so its answers reflect what’s true today, not just its training data, governed on infrastructure you control.
A model’s knowledge stops; the world doesn’t
Any model has a knowledge cutoff. For anything recent — a price, a release, a news event, a competitor’s move — an agent without web access either guesses or admits it doesn’t know. Neither is good enough.
Frozen knowledge
Training data ends; questions about today don’t.
Confident hallucination
Without fresh data, an agent invents plausible-but-wrong answers.
No sources
Answers about the world need citable URLs to be trusted.
Ungoverned browsing
Agents reaching the web need to do so under policy and logging.
The open web, on tap
Freshness
Current, cited results
What’s true today.
The tool runs a web search and returns titles, snippets, and URLs, so an agent can ground answers about recent events, prices, and releases in current sources rather than stale training data.
- Live web results
- Titles, snippets, and URLs
- Adjustable context size
- Tunable result count
Cited sources
Control
Recency and depth
Tune the search to the task.
Limit results to the last N days for time-sensitive questions, and dial context size up or down to balance breadth against focus — so the search fits whatever the agent is doing.
Per query
Governance
Governed web access
Under policy and logging.
Agents reach the web through a controlled tool with audit logging, and the platform can run on infrastructure you control — so external access stays governed rather than ad-hoc.
Logged access
Parameters
The web_search tool accepts these inputs when an agent calls it. Required inputs are flagged.
default: medium Optional Amount of context to retrieve. lowmediumhigh
default: 5 Optional Maximum number of results to return (1–10).
Where web search pays back
Current answers
Ground responses in recent events and data.
Competitive research
Pull the latest on a competitor or market.
Fact-checking
Verify a claim against current sources.
Lead research
Find recent context on a prospect before outreach.
Trend monitoring
Track what’s new in a topic over recent days.
Agent grounding
Give any agent a live window onto the world.
Assigned to agents, orchestrated as networks
On VDF AI, an industry’s use cases map to agents, and you assign tools like this one to those agents. Compose multiple agents into a governed, on-premise network.
What changes after you assign it
Questions about the Web Search tool
What does the web search tool do?
It searches the open web and returns titles, snippets, and URLs, so an agent can ground its answers in current sources rather than relying on a training-data cutoff.
Can I restrict results to recent content?
Yes. The recency_days parameter limits results to the last N days, which is ideal for time-sensitive questions.
How much context does it return?
You control breadth with search_context_size (low, medium, high) and the number of results with max_results (up to 10).
Is web access governed?
Yes. Agents reach the web through a controlled tool with audit logging, and the platform can run on infrastructure you control.
How is it used by agents?
Research, sales, and strategy agents use it for fresh context, often paired with the web crawler for depth and the document generator for output.
Assign Web Search to these agents
These VDF AI agents can be assigned this tool. Open an agent to see the full toolkit it can run.
Tools that work well alongside this one
Where this tool delivers value
Give agents a live window on the world
See the web search tool ground an agent’s answers in current sources — governed and logged.