Skip to content

Sheets11 min read

Stop copying app data by hand: have an agent build the sheet

Prices, listings, messages, schedules and stats often live only inside a phone app, with no export button and no API. A Lusor agent can open the app on a phone, read what is on screen and return it as rows that save to a Sheet. Here is how that works, one worked example, and the limits.

TL;DR

  • Useful data often lives only inside phone apps, with no export and no API, so it gets copied by hand. A Lusor agent can open the app on a phone, read what is on screen and return it as rows.
  • A run's results card stays in the chat. From it, the rows can be saved to a Sheet, the Sheet can be opened, and the rows can be downloaded or copied.
  • Sheets are part of the Lusor app, not Google Sheets. Agents can list Sheets, read them and add rows to them, and typing @ in the composer points an agent at one.
  • The Replay page exports a run's results as CSV, TSV, JSON or Markdown. Through the API, structured output returns a session's result as JSON that follows a schema.
  • Routines rerun a saved prompt every few minutes, hours or days, and the number of routines depends on the plan. The agent can be wrong, so check the rows before relying on them.

A lot of useful data has one home: a screen in a phone app. The price of the same basket in a grocery app. Listings in a marketplace. The class timetable in a gym's booking app. The stats page of a fitness tracker. Messages in a group chat. The app shows it, but it often has no export button and no API, so the usual way to get it into a spreadsheet is to read it off the phone and type it in.

A Lusor agent does the reading instead. It opens the app on a phone, works through the screens, and returns what it found as rows that can be saved to a Sheet. What follows is how that works, one worked example, and the limits.

Why does app data get copied by hand?

Because the app is often the only place the data can be read, and it gives no way to take the data out. Three things tend to be true at once.

  • There is no export. The screen is built for looking, not for saving. At best there is a share button for one item at a time.
  • There is no API. Nothing a script could call returns the same list, or the one that exists is not open to the app's own users.
  • The data depends on the account. Prices, availability and messages differ by sign-in and location, so the version that matters is the one on that phone.

Copying by hand works once. It is slow, it invites typing mistakes, and the table is out of date the day after, so the second and third copies tend not to happen. The data that would be most useful as a series ends up as a single snapshot.

How does an agent get data out of an app?

It uses the app the way a person does: it opens it on a phone, reads what is on screen, and moves through the screens until it has what the task asked for. In the Lusor app that starts with a chat. You give an agent a phone, describe a task in plain words, and it works through the phone's apps until the task is done. The phone can be your own iPhone, your own Android phone, or a Lusor Phone, a cloud phone Lusor runs for you.

Lists are the part that matters for data. Agents read a long list in an app to the end, then open each item in turn, on Android and on iPhone, and the run view shows this as one step with a row for each item. That is the loop a person runs with a spreadsheet open beside the phone: scroll, read, open the item, note the detail, go back. List-reading tools gather what is on screen across a long list, page by page, which is also part of what makes agents fast.

When the agent finishes, its reply and results appear in the chat. What happens to the results next is where Sheets come in.

What are Sheets in Lusor?

Sheets are part of the Lusor app, alongside Phones, Skills, Vault, Inbox, Internet and Settings, and a Sheet is where a run's rows can be saved. They are Lusor's own, not Google Sheets, and they do not sync to Google Sheets.

Google apps come in by a different route. When a workspace admin connects a Google account under Settings → Connectors and switches it on for the agent, the agent can also work with that Google app.

Three things tie Sheets to an agent's work.

  • The results card saves to a Sheet. A run's results card stays in the chat. From it you can save the rows to a Sheet, open the Sheet, or download or copy the rows.
  • Agents can work with Sheets. Agents can list your Sheets, read them, and add rows to them.
  • A mention points the agent at one. Type @ in the composer to point the agent at one of your Sheets.

Read together, a Sheet is more than the place results land. An agent can read it and add to it, so the rows from one task can be what the next task starts from.

What does a run look like from start to finish?

It is one message, one run and one results card. The walkthrough below is an illustration, not a recording of a real run: the app is a generic one and the rows are made up.

The task: someone is shopping for a used road bike and wants every local listing in one table, instead of scrolling a marketplace app each evening. It pays to be specific about the app and the result you want, so the prompt names the app, the search and the columns.

The prompt, sent to an agent that has a phone attached: “Open the marketplace app and search for road bikes within 10 km. For every listing in the results, give me the title, the asking price, the distance and how long ago it was posted, as one row per listing.”

What the agent does:

  • Opens the marketplace app on the phone and runs the search.
  • Reads the results list to the end, not only the first screen of it.
  • Opens a listing when a detail the prompt asked for is not shown in the list, then goes back to the list.
  • Replies in the chat with its results when it is done.

While it works, the phone appears beside the chat and a list of the agent's steps fills in, so the run can be watched, or stopped with Stop.

What the rows look like: one listing per row, one column for each thing the prompt asked for.

Made-up rows for the worked example: one marketplace listing per row
Asking priceDistancePosted
Road bike, 54 cm$4503 km2 days ago
Touring bike, 56 cm$3207 km5 days ago
Carbon bike, 52 cm$1,1004 kmToday
Gravel bike, 58 cm$7809 km1 week ago

From the results card, the rows can be saved to a Sheet and the Sheet opened. Four rows are not worth an agent. Forty are, and so is the second week, when the same table is wanted again with new listings in it.

How do the rows get out of Lusor?

By download or copy from the results card, or by export from the Replay page, which offers CSV, TSV, JSON and Markdown.

The results card is the short route. It stays in the chat after the run, and the rows can be downloaded or copied from it without leaving the thread.

The Replay page is the route with named formats. To open it, hover over a finished reply from the agent and choose Replay. The Replay page exports a run's results as CSV, TSV, JSON or Markdown. CSV and TSV open in any spreadsheet program, JSON suits code, and Markdown suits notes and documents.

A downloaded or exported copy stays as it was when it left Lusor: Lusor Sheets do not sync to Google Sheets. Separately, when a workspace admin has connected a Google account under Settings → Connectors and switched it on for the agent, the agent can also work with that Google app.

How does the sheet stay up to date?

By sending the agent back to it: an agent can read a Sheet and add rows to it, and a Routine can rerun a saved prompt on a schedule.

For a one-off update, type @ in the composer to point the agent at the Sheet, then ask for the new rows. Agents can list your Sheets, read them, and add rows to them, so new rows can go into the Sheet that already holds the old ones.

For a repeating one, there are Routines. Save a prompt on an agent's profile and the agent runs it on its phone every few minutes, hours or days. The number of routines depends on the plan.

A routine runs with nobody watching. Treat the first scheduled runs as a test: check that their rows arrived and look right before relying on the schedule.

How do developers get the data as JSON?

With structured output: give a session a JSON Schema and the agent returns its result as JSON that follows it, so code can read the result without parsing prose.

Take a price check. The prompt asks for the price of a 12-pack of sparkling water in a grocery app. The schema defines a price as a number and a currency as a string, and marks both as required. When the session is completed, its output is an object with those two fields, for example a price of 6.49 and the currency USD.

Three details matter when planning around it.

  • Structured output is available on sessions without an agent, which are started on a phone directly through the API. Turns sent to an agent do not take a schema.
  • The output is filled only when the session completes. A failed session has an error instead.
  • The schema does not replace instructions. Describe the task in the prompt and the shape of the result in the schema, and mark the fields that are always needed as required.

What are the limits?

The main one is that the agent can be wrong, so rows need checking before anything depends on them. The rest are about sign-ins, privacy, cost and where the rows live.

  • Check the rows. A row can be misread, or an item missed. Lusor keeps a record of every task so you can check what the agent did, with the agent's steps and a screenshot of the phone for every step. Compare a sample of rows with the replay, and all of them when a decision rests on the numbers.
  • Sign-ins come first. An app that needs a login needs one on the phone. Sign in to the app on the phone yourself first, or give the task in the Lusor app, where the agent can ask: a secure field appears, and the value is passed to the phone and is not shown in the chat. A session started with an API key does not pause for a secret. It uses the saved Vault secrets it was allowed, by name, and ends as failed when it needs one it was not given.
  • What is on screen is in scope. The agent reads the screen to do the work, and that includes whatever else is on it. Before pointing an agent at messages or account pages, read up on phone agent safety.
  • Every run uses credits. The AI work behind each reply is paid from credits, and on a Lusor Phone the agent's time also counts toward Lusor Phone minutes. On your own iPhone or Android phone it does not. A routine spends credits each time it runs, so a short interval adds up against the plan's monthly credit allowance.
  • History has a window. On the Free plan, messages, steps and screenshots are kept for 7 days and then no longer show in the app. Download or export rows that matter while the run is still there.
  • Sheets do not sync to Google Sheets. A Lusor Sheet lives in the Lusor app. If a workspace admin connects a Google account under Settings → Connectors and switches it on for the agent, the agent can also work with that Google app.

To try it, pick one table that gets copied by hand today, write the prompt with the app and the columns named, and compare the agent's rows with the app. The Free plan costs $0 a month and includes 1,000 credits and one phone slot, and the pricing page compares the rest.

Frequently asked questions

Are Lusor Sheets the same as Google Sheets?

No. Sheets are part of the Lusor app, alongside Phones, Skills, Vault, Inbox, Internet and Settings, and they do not sync to Google Sheets. Google apps come in by a different route: when a workspace admin connects a Google account under Settings → Connectors and switches it on for the agent, the agent can also work with that Google app.

What can an agent do with a Sheet?

Agents can list your Sheets, read them, and add rows to them. Typing @ in the composer points the agent at one of your Sheets. A run's results card also saves its rows to a Sheet and opens that Sheet.

How do the rows get out of Lusor?

By download or copy from the results card in the chat, or by export from the Replay page, which offers CSV, TSV, JSON and Markdown. A downloaded or exported copy stays as it was when it left Lusor.

Can an agent collect the same data on a schedule?

Yes, with a Routine: save a prompt on an agent's profile and the agent runs it on its phone every few minutes, hours or days. The number of routines depends on the plan. A routine runs with nobody watching, so treat the first scheduled runs as a test and check their rows before relying on the schedule.

How do developers get the data as JSON?

With structured output: give a session a JSON Schema and the agent returns its result as JSON that follows it. It is available on sessions started without an agent. Turns sent to an agent do not take a schema.

Do the rows need checking?

Yes. The agent can be wrong, so rows need checking before anything depends on them. Lusor keeps a record of every task, with the agent's steps and a screenshot of the phone for every step, so a row can be compared with the screen it came from.