This article is auto-synced from its in-app version in Tai.
Reports are AI-generated documents produced by running an agent against your inputs — for example, a risk report for a booking or a summary analysis for a trip.
When you'd use this
Use Reports when you need a standalone document rather than a conversation. For example: run the Trip Risk Manager agent against a booking reference to get a formatted risk assessment you can review, share, or export as PDF.
Steps
Browse the reports library
In the Workbench sidebar, click Reports (under the Terminal section).
Reports opens on a Choose an agent grid — the same picker as Chats, listing only agents configured for report output.
Pick an agent to see the reports it has produced — title, who ran it, date, and status. Use Search and the Status filter to narrow the list. Your filter and sort selection (other than search) is remembered, so it's applied again the next time you open the list.
Click a report title to open its detail view. Click Switch agent in the header bar to return to the picker and browse another agent's reports.
Run a report
In the Workbench sidebar, click Reports.
Click Run report in the top-right corner.
Select an agent from the Agent dropdown. Only agents configured for report output appear here.
Fill in the input fields. Each agent defines its own inputs, and each renders by type — a text box, a number, a single date, a date range (pick a start and an end date), or a file upload. Required fields are marked with an asterisk.
Click Run. The report generates live; you can watch the output stream as it is produced.
When the run completes, the page navigates to the finished report automatically.
If a lot of reports are already running, the run is queued instead of starting straight away. You'll see its place in line, and it starts on its own as soon as capacity frees up — there's nothing to retry. Click View this run to follow it, or just carry on; you can come back to it from the reports library at any point.
To stop a run before it finishes, click Cancel run next to the live status. The run stops and is marked Cancelled.
To have the same report run and be emailed automatically on a recurring schedule, click Schedule this report — see Scheduling reports.
View and export a report
Open a report from the library.
The detail view opens with Run details and the Input parameters the report was generated from — who ran it, when, and with what scope, so you can see the report's inputs without re-opening the agent. Duration measures the run itself, from the moment it started to the moment it finished; for a scheduled report that waited in the queue first, that waiting time is not counted. Below those, Consumption summarises what the run spent — the credits it used and what they went on, and how many connector calls it made. When the agent narrated how it worked, an AI narrative section sits below; expand it for that context.
The report itself renders inline below — HTML reports display in a sandboxed frame; Markdown reports display as formatted prose. Under the title, the agent's name is followed by AI-generated — the same label every report carries in the reports library and the standalone view.
Use the buttons in the report's top-right corner to download a copy: Export HTML works for every report; Export PDF works for every report too, whatever its layout — an interactive report prints with its tabs and panels opened out. Both exports end with a footer, in the language you use TAI in, saying the report was generated by AI; the PDF footer also numbers the pages.
Download the files a report produced
A report can also hand back files of its own — a spreadsheet of the underlying rows, a chart image, a formatted document. These are separate from the Export buttons above: those render the report you're reading, while these are files the agent built during the run.
Open a report from the library.
Any files the run produced appear as download cards below the report — one per file, each showing its name and format.
Click a card to download that file.
Ask for them in the report's requirements when you run it — for example, "include an Excel export of the results" or "also give me the raw data as a CSV". There's no format setting to switch on: the request decides, and one run can produce several files.
Reports can deliver Excel (.xlsx), Word (.docx), CSV, PDF, images, and text formats such as Markdown, JSON, and XML.
Excel and Word exports have a size limit, and one that would go over it is refused before the file is built. Nothing is written, so there is no download card for it and nothing appears in the panel below. The run carries on: the agent is told the limit and can narrow the export itself, usually by aggregating the rows into totals or a top N. If you asked for every row and the export came back summarised, that is why — re-run with a narrower request, such as a shorter date range or fewer columns, when you need the detail.
Separately, a run can build a file it cannot hand over — the finished file came out over the size limit for generated files, it is a format TAI does not serve as a download, or the agent saved it somewhere the run does not collect downloads from. When that happens the report still opens normally, and a Files that could not be attached panel appears under the download cards, naming each file and why. It is worth reading: the report's own summary may well say the file is attached, and this panel is what tells you otherwise. What to do next depends on the reason the panel gives. For a file that came out over the size limit, re-running with a narrower request — a shorter date range, fewer columns, a CSV instead of a spreadsheet — usually brings it back under. For one saved in the wrong place, re-run and ask for it as a download rather than naming a filename yourself.
The same panel appears while a report is running, and on the run's own page if the run later fails — so a file that went missing is named even when the run ended for some unrelated reason. Exported copies (Export PDF / Export HTML) carry the list too, so a report you forward doesn't promise a file the recipient will never find.
Stop a report that is still running
You don't have to be on the page that started the run — a long run can be stopped from any tab, or after you've navigated away and come back.
In the reports library, click the row of a report whose status is Running. Its detail view opens and refreshes on its own. A scheduled run that hasn't started yet shows Queued, and can be stopped the same way from the schedule's run history.
Click Cancel run. The run stops and its status becomes Cancelled.
Whatever the run managed to produce before stopping stays on the page, and Re-run / Edit & re-run are offered so you can start again — usually with a narrower request.
Re-run a report that did not finish
In the reports library, a report that failed, was cut off (Incomplete), or was cancelled shows a status badge. Click its row to open the run's detail view. (A finished report opens the document instead.)
The detail view explains why the run stopped, alongside the Input parameters it ran with.
Click Re-run to launch the report again with the same parameters in one click, or Edit & re-run to adjust the parameters first.
See what is queued and running (Company Admin)
In the Workbench sidebar, open Reports and click Queue.
The list shows every report run in your company that is queued or already running, in the order they were started, with who launched each one and its position in the line.
Click a row to open that run's detail view, or use the cancel button on the row to stop a run that is holding up the queue. Cancelling stops the run for its owner, not just for you.
The list refreshes on its own, so you can leave it open while a backlog clears.
Tips and limits
All users can view and run reports; deleting a report requires the Company Admin role. The Queue view is Company Admin only.
There is a limit on how many reports can run at the same time, both per person and across your whole company. It keeps a burst of launches from slowing every run down. Going over it doesn't lose anything — the extra runs queue and start automatically.
A queued run behaves like any other unfinished run: it appears in the library with a Queued badge, and you can open or cancel it while it waits.
A report that needs a file upload can't wait in the queue, because the uploaded file isn't kept. If the limit is reached, that run is turned away and you'll be asked to try again in a few minutes.
File inputs accept text, CSV, Markdown, HTML, JSON, XML, PDF, and image files (up to the per-file size limit). The agent reads the uploaded file as part of the run.
The file type is taken from the filename extension. A file saved without one —
exportrather thanexport.csv— is turned away even when its contents are fine, so rename it and upload again.A date range expects both a start and an end date; the run won't start until a required range has both.
A JSON input expects valid JSON — an object like
{"region": "EMEA"}, an array, or a quoted value like"EMEA"(a bare word is not valid JSON). The field flags Invalid JSON as you submit, and an empty{},[], or""counts as no input for a required field.Only agents that have the Report channel enabled appear in the agent picker. To make an agent available here, enable the Report channel in its Channels tab.
A cancelled run stays in the library with a Cancelled badge and keeps what it produced, so you can review it or re-run it. It is not a failure, and it produces no report to open or export.
An Incomplete run is one that ran to the end but could not produce a usable report — most often because the answer grew too long and was cut off. The run's detail view says what stopped it and keeps whatever it produced, but there is no report to open or export. Narrow the request — a shorter date range, fewer sections — and re-run.
A report marked Stopped still opens and exports normally — it is not a failure. It means the agent chose to stop before finishing its full request, under its own rules, rather than deliver something incomplete or wrong. A banner on the report explains this; open it to see what the agent did produce. The Status filter lists these under Stopped; Ready leaves them out.
Cancelling stops the run itself, not just your view of it — a scheduled run that hasn't started yet can be cancelled the same way, and won't produce or email a report.
A report is linked to the agent that produced it. If that agent is deleted, the report remains in the library — but a failed run can only be re-run while its agent is still available to you.
A report whose agent requires a file upload can't be re-run in one click, because a past run's uploaded file isn't kept. Use Edit & re-run and choose the file again.
PDF export renders the report at the time you click the button — it reflects the current content.
Every report offers both Export PDF and Export HTML, whatever its layout. An interactive report prints with its tabs and panels opened out, so the PDF carries the whole report rather than the one view you were looking at.
The Consumption section on a report shows what the run consumed in credits — see Credits and billing.
Reports cannot be edited after they are generated. Run the agent again to produce a new version.
Keep the AI labels when you share a report. TAI adds the AI-generated label and the export footer; they can't be switched off. If you publish a report as-is — no person reviewing and taking responsibility for the text — to inform the public on a matter of public interest, EU rules ask you to disclose that it is AI-generated; leaving the footer in place does that. Reports also carry hidden markers so software can tell they are AI-generated; don't strip them. See EU AI Act.
Related