Ask Torki

Send Torki to a page

Name a site and one thing you want from it; Shadow loads the page itself and answers in the thread, or tells you the page did not carry enough to answer from.

See which plan includes itStart free anyway

Shadow is not included on Free. It starts on Lite with five runs every five hours; Arc gives 30 and Zenith 150, each counted over a rolling five-hour window — so the button below goes to pricing rather than sign-up.

Shadow is Torki's web agent. You name a site by domain and say what to bring back, and Torki loads that page on its own side — the page does not open in your browser. Everything happens in one continuous thread, and each new run lands at the bottom of that same scroll, under the one before it. The flow below is the whole of it, including what Shadow says when a page does not have enough text on it.

How it works, step by step

1

Open Shadow

In the left sidebar of the web app, click Shadow — the row carries a purple BETA pill. Shadow opens on a welcome screen headed What can I do for you? over Browse the web, research topics, extract data, and more — in one continuous thread.

2

Name the site and the one thing you want

Type one request that names the site by domain and says what to bring back — Go to techcrunch.com and tell me the top stories, or Go to github.com/trending and list the top repos — then press Enter. Keep it to a single ask: adding a second verb such as look up or search sends the request to a planner instead of a straight page fetch. The four suggestion chips only fill the composer, so you still press Enter — and each of them asks for two things at once, so none of them is the single-ask shape described here. Type your own.

3

Open the workspace panel to follow the run

The narration line under Torki's mark reads Starting task..., then analyzing, then your own sentence echoed back, then Working... while the page is fetched — nothing narrates the load itself. Open the right-hand panel with the top-bar button whose tooltip reads Toggle workspace: Progress lists the run as Step 1, Step 2 — a live run streams counts, not names. A Browser section carrying the address joins the panel once Shadow has the page, which is as the step closes rather than while it loads.

4

Read the answer

The answer replaces the dots, rendered as markdown — a heading, a list or a table, kept short on purpose. If the page did not carry enough text, Shadow says so and links the page rather than guessing: I loaded the page but it didn't contain enough text to answer your question reliably. You can check the source directly: followed by the URL. If nothing was browsed at all, the answer is topped with a quoted line: ⚠ This answer was generated without browsing. Results may be inaccurate.

5

Check what it cited

Links inside the answer render as brand pills with an external-link icon — Wikipedia, GitHub, TechCrunch. When Shadow has a page to cite, the answer closes with a horizontal rule and a bold Sources row of up to five of those pills. Do not count on the row being there: a clean single-page extraction often carries none, while it turns up reliably in the weaker outcomes — when Shadow could only open the page, or when the page was too thin to answer from.

6

Copy it, or spend another run

The run closes with a finish bar: a check icon, Done, then Copy and Again. Copy puts the answer on your clipboard. Again re-sends the same instruction and spends another run from your window.

Before you start

The things people hit on their first try, so you do not have to find them the hard way.

  • When you have no runs left, nothing happens

    This is the first thing most people meet, because a Free account has no Shadow runs at all and Lite has five in a rolling five-hour window. Sending a task with none left does not raise an error in the thread — the request simply does not start. Check your plan before you use this one.

  • A run is spent when you send it

    Your allowance is charged as the request goes out, not when the answer lands, and nothing gives it back if the answer is not useful. A run that returns only a link still costs one, and Again costs another.

  • One ask per request

    Shadow is at its best opening one site and pulling one thing out of it. Pairing a navigation verb with a second one — go to X and search for Y — routes the request to a planner instead of a straight page fetch, which is a different and less reliable path. The four suggestion chips on the welcome screen are all written that way, so they are not the shape to copy.

  • The box is the chat composer

    Shadow types into the same composer as the chat, so the extras around it look available here. Only the sentence you type is sent: a file you attach to a Shadow request is dropped without a message.

  • There is no stop control

    Once a run starts, the composer is disabled and no stop button appears. If you sent the wrong thing, reloading the page is the way out, and the run is still spent.

  • One continuous thread per account

    Everything you ask Shadow stays in one scroll, so yesterday's questions and answers are still there tomorrow. Recents in the left panel stays hidden behind Click to view recent activity and is a jump-list of your past instructions, not a viewer for a past answer. A long answer comes back shortened after a reload.

Next

Other walkthroughs in Ask Torki.

Ready when you are.

This one needs a paid plan. An account is still free to make, and everything else in Torki works on it.

See plansCreate a free account