# Bring the live web _into your workflows._ Web Search lets a workflow (or an agent) run a real search and act on what comes back. It pages through the web until it has the number of clean results you asked for, throws out the duplicates, and can even read each page back as plain text. **One capability instead of wiring a search API into your stack.** ## One search. _Two ways to use it._ The same live web search shows up in two places. Drop it into an automation as a step you control, or hand it to an AI agent as a tool it can reach for on its own. Same engine underneath, same credits. **As a workflow step** A node you drop on the canvas and configure: the query, how many results, how recent. It runs as a deterministic step in your automation, every time, in the same place. **As an agent tool** The same search, registered as a tool an AI agent can call mid-conversation. When it needs current information, it decides to look it up, and answers from the live web, not stale memory. ## It pages the web until it has _your number_. Most “web search” features hand you whatever the first page returns (about ten links) and call it done. Web Search doesn't. Ask for fifty results and it fetches page after page, dropping every duplicate URL as it goes, and keeps going until it has fifty clean, distinct sources. No padding, no repeats, no “first page and done.” If the web genuinely has nothing left, it stops honestly and tells you. The result even carries a note explaining that it de-duplicated by source. ## Most search gives you links. This one _reads the page_. Ten blue links still leave you with the real work: open each one, hunt past the nav and the ads, and copy out the part that matters. Web Search can do that for you. Turn on page content and it opens every result, strips the page chrome, and hands your workflow the clean article text, ready for an AI step to summarise or pull fields from. You only pay for the results you actually get back. ## When the web has a straight answer, _you get the answer_. Not every search wants a list. Web Search looks at the shape of what comes back. If there's a ranked list of pages, you get the list. If the web has a direct answer or a quick fact, it hands you that instead, and if there's nothing useful, it says so cleanly rather than failing. - “Series A SaaS hiring RevOps” → a ranked list of sources - “What's the GDP of France?” → one answer card - A query with nothing behind it → an honest “no good results” ### What's the GDP of France? *Result · answer* $3.03 trillion (2024) ## You decide _how fresh_ and _how wide_. Two simple controls change everything a search returns. Scope it to the last day, week, month or year when recency matters, and point it at the whole web or a single site when you already know where to look. ## Start with a question already written. You don't have to phrase the perfect query. Pick a ready-made template, drop in your topic, and the search is set up for you, or start from scratch and write your own. **Research a topic** Latest trends and analysis on anything you’re tracking. **Competitor analysis** Products, pricing and market position for a named rival. **Lead generation** Companies and the people who make the decisions inside them. **Content research** Trending angles and what an audience actually cares about. **Start from scratch** An empty query when you already know exactly what to ask. ## What it looks like _in real work_. Web Search isn't a destination. It's a step. Wired into the products you already use, the live web becomes part of how things get done. **Brief the rep before they reply** A lead fills out your TinyForm with their company name. The workflow feeds that answer straight into a competitor search and drops a one-page brief in the rep's inbox, before anyone responds. **A market pulse every morning** A scheduled workflow runs a topic search scoped to the past day, pulls the freshest twenty sources, de-duplicates them and lands a clean digest in a table or an email while you sleep. **Enrich a whole list of accounts** For every row in a TinyTable, search that company's own site for their pricing, read the page back as clean text, and write a tidy summary into a new column. No tabs, no copy-paste. **An agent that cites live sources** Mid-conversation, a research or support agent realises it needs current information, calls Web Search on its own, and answers with up-to-date facts and links instead of stale training data. ## The results go _straight to work_. A search that just shows you results is a dead end. Drop Web Search into a form or a workflow and route what it finds: read the pages, summarise them, and send the result onward without anyone retyping a thing. ## Three ways to drop it in. Web Search lives in the builder as a node and as an agent tool. Add it to a TinyForm or a Workflow, wire a query in, and route the result onward. Metered in the same credits as everything else. **Web Search** 10 credits **Web Search with page content** 2 credits / result **Web Search for agents** metered per call The page-content variant is priced honestly: **two credits per result it actually returns**, so you only pay for what comes back. ## Built so it doesn't fall over. The live web is messy: slow pages, hiccups, dead ends. Web Search expects that. **Retries on a hiccup** A failed call is retried with a short backoff before it gives up, so a single blip won't break your run. **Hard timeouts** Every search has a time limit, so a slow corner of the web can't stall the whole workflow. **Graceful empty results** If a search truly finds nothing, it returns a clean empty result. It never crashes the automation around it. **No duplicate sources** Every result is checked against the ones already collected and dropped if the URL repeats. **Variable-fed queries** The query can be a form answer, a table cell or an earlier step's output, not just a fixed string. **Pay per result** The page-content variant charges only for results it returns: nothing for a search that comes up empty. ## How it compares. The raw search APIs are fast and powerful. The difference is that Web Search is built into the place you already work, and it does the parts they leave to you. | | TinyCommand Web Search | Search APIs | A chatbot's web tool | Manual googling | | --- | --- | --- | --- | --- | | Reads result pages back as clean text | ✓ | No | some | By hand | | Pages until it has the count you asked for | ✓ | you loop it | No | By hand | | De-duplicates sources for you | ✓ | No | No | No | | Works as a workflow step and an agent tool | ✓ | API only | chat only | No | | One bill with the rest of your stack | ✓ | No | maybe | — | If you want a raw search API to wire up yourself, the search vendors do that well. If you want the live web searched, read and acted on inside the tools you already use, that's what we built. ## Good to know. Where do the results come from? + The live web: current, public web results, the same kind you'd get from a search engine. Web Search runs the query server-side, collects the results across as many pages as it needs, and hands them back clean and de-duplicated. What's the difference between the two search nodes? + Plain Web Search returns ranked results: a title, link and snippet for each. Web Search with page content goes a step further: it opens every result, strips the nav and ads, and returns the full page as clean text, ready for an AI step to summarise or pull fields from. Can I limit results to recent pages? + Yes. Scope any search to the last day, week, month or year. It's one setting on the node, handy for a daily market pulse or anything where stale results would only get in the way. Can an agent use it on its own? + Yes. Web Search is registered as a tool your AI agent can call mid-conversation. When it needs current information it decides to look it up, then answers from live results instead of stale training data. Do I need to write any code? + No. Web Search is a drag-in node inside TinyForms and Workflows, and a built-in tool for Agents. Add it, point it at a query, and route the result onward. No keys, no servers, nothing to host. What does it cost? + It's metered in the same credits as the rest of TinyCommand: a plain search runs 10 credits, and the page-content variant is 2 credits for each result it returns, so you only pay for what comes back. On every plan, including Free.