> ## Documentation Index
> Fetch the complete documentation index at: https://docs.exec.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Limits
> How much comes back at once, what gets trimmed, and how to ask so it does not matter.
None of these will break anything, and most of them only start to matter on a workspace with hundreds of people or thousands of sessions. They shape how you ask, and knowing them saves you assuming something is missing when it is only paginated.
***
## How much comes back at once
| | |
| :------------------ | :------------------------------------ |
| Results per request | 50 by default, 100 at most |
| More than that | Your assistant asks for the next page |
On a big workspace, "show me every session this quarter" gets the first page rather than all of it. Two better habits:
**Narrow the question.** A team, a scenario, or a date range gets you a complete answer instead of the first hundred rows of an incomplete one.
**Ask for the summary, not the rows.** `What was the average score per rep this month?` is both more useful and cheaper than asking for every session and adding them up yourself.
***
## Transcripts and detailed feedback are large
A full transcript with per-criterion grades is a lot of text, and asking for several at once will get them trimmed.
The pattern that works:
```text theme={null}
Summarize the coach feedback across my team's last ten roleplay sessions
and tell me the common theme.
```
Coach feedback for many sessions comes back in one request, so this is fast.
```text theme={null}
Now show me the full transcript and per-criterion feedback for that session.
```
Asking for ten transcripts at once is the most common way to get an incomplete answer.
***
## Charts need an assistant that can display them
Charts, dashboards, and interactive tables render as a panel inside the conversation, which requires an assistant that supports interactive panels. If yours does not, you get the same numbers as text and nothing fails.
**The first chart in a session is slow.** It takes a few seconds to start up, and everything after that is quick. If you are about to demo this, ask for one throwaway chart first.
Charting a very large amount of data is slower and less useful than charting a summary. Prefer `chart the average score per rep this month` over charting every individual session.
***
## Building a roleplay
| | |
| :------------------------------- | :----------------------------------------------------------------------- |
| How long it takes | Usually under a minute, sometimes a few |
| How long you can come back to it | About an hour before your assistant loses track of that particular build |
| Publishing the same draft twice | Not possible. It is live after the first time. |
| Workspace scenario limit | Enforced when publishing. If you are at your limit, it says so. |
Two things your assistant cannot pick up:
* **A scenario you started building inside Exec.** Conversations begun in Scenario Studio in the product cannot be continued through an assistant. It works the other way round though: a scenario your assistant starts can be opened and finished in Exec.
* **Somebody else's build.** Only the person who started it can check on it.
***
## Rate limits
There is a generous cap on how many requests one person can make per minute. Normal conversation, even a fast back-and-forth, will not reach it. You would have to be running something automated to notice.
If you do hit it, waiting a minute clears it.
***
## What is not available at all
* **Knowledge Hub content** cannot be searched or read. It can be used as source material when building a scenario.
* **Billing and subscription data.**
* **Any other workspace.** A connection is bound to the one you chose at sign-in.
* **Changing anything except creating a roleplay.** No deleting, editing, assigning, or inviting.
***
## Next
When something is not working
When it works but the answer looks wrong