Web search on VM0
Current results from the public web with the source behind every one, so an agent can find the right page before it reads it.
Web data · No setup · Managed credits per successful call
Web search is how an agent finds things. It asks a question and gets back ranked results from the public web with the link behind each one, narrowed to the last day, week or month, or kept inside a handful of sources you trust.
It runs on Perplexity, which VM0 pays for, so there is no search account, no key and no per-seat plan on your side. What comes back is material to work from rather than a finished answer: search, pick the results that matter, then read those.
What Web search is
An agent that cannot search the live web only knows what its model learned before it was trained. That is fine for reasoning and useless for anything about this week.
A search sends the question out to Perplexity through VM0 and returns ranked results with titles, links and short extracts. The agent can ask for more or fewer results, keep them to what was published recently, and hold them inside sources you name.
Two things follow from this being a call out to the open internet. The question itself leaves VM0, so it must never carry secrets or private context. And what comes back is public text written by strangers, read as information and never as instructions.
People arrive at this from several directions: looking for a web search API, comparing Perplexity API pricing, or hunting for something to replace a search-results API in an agent. All three end up in the same place, which is a search your agent can run without an account of its own.
What Web search can do
What an agent can do with it, and what comes back when it does.
Coverage and limits
What it does not do, said up front, so nobody plans a workflow around something that is out of scope.
The public web only
It searches what anyone can find. Your own documents, wikis and tickets are reached through the connectors for those systems instead.
Sources, not answers
You get ranked results back. When the work needs what a page actually says rather than a two-line extract, the next step is reading the results worth reading.
The question leaves VM0
The wording of the search goes to an outside provider, so it must never contain passwords, customer data or internal context you would not publish.
Results are not trusted
Titles, links and extracts are information to read, never instructions to follow, which is what keeps a planted result from steering a run.
What Web search costs
Web services are billed in credits per call, not per token. There is no separate vendor bill to reconcile.
zero web-searchEach successful search costs credits, with no plan or seat licence underneath it. The number of results is the lever worth using: one search that brings back five good results is cheaper than three that bring back thirty mediocre ones.
Setup and access
Nothing to set up
Nothing to connect. No search account to create, no key to store, no plan to choose. An agent can search on its first run.
Who can use it
Searching the web is a permission, granted per person and per agent and taken back on its own. A team that wants research agents but no access to the live web simply does not grant it.
What teams use Web search for
Finding out what happened this week
Keeping results to the last few days is what makes a daily or weekly brief possible: take the handful that matter, read the two that count, and write the summary from the page rather than the extract.
Research held to sources you trust
Naming the sites keeps a legal, medical or financial question inside official sources, which is usually the difference between a usable answer and a plausible one.
Checking a claim before it goes out
Before a report or an outbound email states something, a search can confirm it and attach the link, so whoever reviews it can see where the number came from.
When not to use Web search
Skip it for anything inside your own systems, where the right tool is the connector for that system. Skip it when you already know the link, where reading the page is one step instead of two. And do not treat it as a fact checker: it finds sources, and judging whether a source is any good is still yours to do.
What this is called elsewhere
The same thing goes by several names in the market. If you have shopped for one of these, this is how it maps to what you get here.
Web search API
The general name for search you call rather than type into a browser. This is that, with the account, the key and the plan on VM0's side.
The Perplexity API without a Perplexity key
Perplexity answers the search. You never see its pricing page, generate a key or pick a tier; the search is charged to your VM0 credits per call.
A search-results API
What people reach for when they want ranked links rather than a written answer. That is what comes back here: titles, links and short extracts you can then read properly.
Search for an agent or a tool call
The same capability people wire into agents as a search tool. Here it is already there, so an agent can search on its first run with nothing configured.
Live or real-time web search
Search restricted to what was published recently, which is the whole point when the question is about this week rather than about the world in general.
Web search compared
Web search vs paying for search yourself
Your own search API means an account, a key, a plan and usage someone reconciles per team. This moves all of that onto VM0 and turns access into a permission you can grant for a week and take back.
Web search vs what the model already knows
A model answers from its training data with no way to tell you how stale it is. A search returns dated, linkable sources, which is what lets a reviewer check the work instead of trusting it.
Web search vs scraping a search engine
Reading results pages yourself breaks constantly, gets blocked, and is a licensing problem. This is a proper search with filters and a stable shape.
The short version
The front door for anything that has to be current. Search to find the sources, read them to get the substance, and keep private context out of the question, because the question leaves VM0.
Frequently asked questions
Do I need my own search account or key?
No. VM0 runs the search on your behalf and charges it in credits. There is nothing to sign up for on your side.
Can I keep results to specific sites?
Yes. A search can be pinned to a handful of sources you name, which is how a regulatory or medical question stays inside official ones.
Can I get only recent results?
Yes. Results can be held to a recent window, which is what makes daily and weekly briefs work.
Does it search my company's own content?
No. It searches the public web. Internal content is reached through the connector for the system that holds it.
Can I put internal context in the question?
No. The question goes to an outside provider, so it must not contain passwords, customer data or private internal context.
How many results come back?
Five by default, and the agent can ask for more or fewer. Asking for fewer good ones is usually cheaper and better than asking for many.
Is this the Perplexity API?
It is Perplexity's search, called by VM0 and charged to your credits. You do not have a Perplexity account, a key, or a Perplexity bill.
What does it cost compared with buying search directly?
Search APIs generally sell plans or credit bundles you commit to. Here each search costs credits as you use it, with nothing to commit to and nothing to reconcile.
Can it replace a search-results API in my own setup?
For work that happens inside a VM0 agent, yes. It is not a general-purpose API you can call from your own application; it is a capability your agents have.
Can an agent use it as a tool without me configuring anything?
Yes. It is built in, so an agent reaches for it on its own when a request needs current information. There is no tool to install or key to supply.
How you ask for Web search
You describe what you want in plain language. The agent works out which service it needs and makes the calls.
“Search for what shipped in AI agent tooling this week and summarize the five most substantial items with sources.”
“Find the current EU rules on AI transparency, using only official EU sites, and quote the relevant lines.”
“Confirm the number in this draft before it goes out, and attach the source you used.”