Integration
Web search in Polaris
Nothing to authorize, nothing to store, and every query a worker runs appears in the activity feed while it works.
What connecting buys
Web search is the one connection in the Polaris catalog that needs no credential. A worker running a task on a cloud machine performs live searches during the session, each one recorded in the activity feed as it happens. Searches are capped per session and counted in the human-hours estimate at roughly twelve minutes each, so research effort shows up on the bill line by line.
- Credential
- None
- Runs
- Inside the delivery session
- Counted at
- About 12 minutes per search
- Per session
- Capped at a handful of searches
Why this one is different
Thirteen connections in the catalog exist to give a worker access to something your company owns. Web search gives it access to what everybody owns, which is why there is no token, no consent screen and no owner-only authorization step. Every worker has it.
It is also the connection that changes the character of the output most. A brief written without search is a language model recalling; a brief written with it is a worker reading current pages and citing them. The second kind is the only kind worth paying for.
What search does inside a session
The mechanics, because they explain both the quality and the cost.
It runs while the machine works
Searches happen inside the tool loop on the cloud machine, not as a preparation step. A worker searches, reads, revises what it thought, and searches again.
Every query is posted
Each search is emitted to the activity feed with the query text, so you can watch what a worker looked for and judge whether it looked in sensible places.
The session is capped
A single run allows a handful of searches rather than an unbounded crawl. A question needing forty sources is several tasks, not one heroic session.
It feeds the bill honestly
The hours formula counts each search at about twelve minutes of human-equivalent effort, which is roughly what the same lookup and read costs a person.
What five searches cost
Worked from the published hours formula, not measured on a customer. Hours are human-equivalent and every line is written to the worker's work log.
| Line | Human-equivalent minutes |
|---|---|
| Picking up the task | 15 |
| 5 live web searches at 12 minutes each | 60 |
| The finished prose, at 90 characters per minute | varies with length |
| Search subtotal alone, at $2 per hour | $2.00 |
Workers and work this connection feeds
Hire an AI research analyst
Assign the question on Tuesday afternoon and read the brief before your Wednesday call, with every claim carrying the URL it came from.
Hire an AI competitive analyst
Assign it every Monday and you get a written record of what your market did last week, with a URL behind every line of it.
Hire an AI SEO specialist
It looks at what is actually ranking today, then writes the brief that says what your page has to do differently to deserve the spot.
Research memos waiting for you in the morning
The questions worth researching are the ones you never have a free afternoon for.
Competitor monitoring that actually happens every week
Everyone agrees competitor tracking matters and nobody has done it since the last time a deal was lost over it.
Account briefs waiting for you before the first call
Ten minutes of research changes a first call. Nobody has ten minutes before a first call.
A competitor changed their pricing and you found out in a sales call
A weekly diff of public competitor pages, reporting only what changed since last time.
Questions people ask
+Do I have to enable web search for a worker?
It is in the catalog as a connection with no credential and no authorization step, so there is nothing to enable and nothing to store. Adding it to a worker's list is a statement about the role rather than a permission grant.
+How do I know a worker actually searched?
Every query is posted to the activity feed as it runs, with the text of the search, and the count appears on the work log line that priced the job. A confident brief with no searches logged is a brief worth challenging, and the record to challenge it with is right there.
+Does web search make a task more expensive?
Yes, by design and transparently. The hours formula counts each search at about twelve minutes of human-equivalent effort, so five searches add an hour, which is two dollars. That is what makes research work priceable rather than a mystery line on a bill.
+Can a worker read a specific page I give it?
Put the URL in the task and say what you want from it. Task context is the first thing a worker reads, and pointing it at a page beats hoping a search surfaces the one you had in mind.
+Why is the number of searches limited?
Because a session that searches indefinitely produces a worse answer and a bigger bill. The cap keeps a run bounded to minutes and keeps the hours estimate honest. Deep questions are better split into several tasks, each with its own acceptance criteria.
Related
Every tool a Polaris worker can be given
One catalog, one credential per tool per organisation, authorized by an owner and used by every worker who carries it.
Hire an AI research analyst
Assign the question on Tuesday afternoon and read the brief before your Wednesday call, with every claim carrying the URL it came from.
Hire an AI competitive analyst
Assign it every Monday and you get a written record of what your market did last week, with a URL behind every line of it.
Research memos waiting for you in the morning
The questions worth researching are the ones you never have a free afternoon for.
Work log
The record that makes an unwatched run reviewable afterwards.
Human-equivalent hours
A billing unit denominated in the work replaced, not the compute consumed.
What an AI worker costs, and how the hours are counted
The whole formula is on this page, including the parts that make it an estimate rather than a measurement.