Enterpret Agent Sessions: controls, sharing, and answer verification
Use this reference while working in an Enterpret Agent Session. For the first-session walkthrough, see Getting Started with Enterpret Agent.
Session controls
| Control | How to use it |
|---|---|
| Model picker | Select an available model when the control is enabled. Available choices and editability are shown by the Session. |
| Plus menu | Add Attachments, choose Skills or Connectors, or open Manage rules when those options are available. |
| Slash menu | Type / and choose a Skill from the available list. |
| Approval mode | Check the current Ask or Auto setting and review any action requests. |
| Share | Check the current access setting before copying or distributing the Session link. |
Some controls depend on your permissions, the Session's state, and the features available in your organization.
Changing models
You can change a model between completed turns when the picker allows it. An active turn or pending action can prevent a change. Wait for the current work to finish or resolve the requested action, then follow the message shown by the control. If your interface asks you to start a new Session, use New session.
Speed, thinking, and context
Use the model picker to choose among the models and thinking levels offered to your Session. Options vary by model. When Fast mode is offered, its lightning control uses priority processing for faster responses at a higher cost; it is separate from choosing a lighter model or thinking level. Follow the control's message if your selected model does not support Fast mode.
A message can include up to 500 attached items in total, including supported links and mentions. Attachments also have the smaller limit of 10 files per message. Remove items if the composer reports that the message is over its limit.
Approving connected-tool actions
The default interactive Agent uses Ask mode. Some tools pause for approval, including reads as well as writes. Check the tool, connected account, destination, and proposed action before choosing an available option:
Allow Once: allow this call.
Allow Always: when offered, allow that tool for future use with your connected account.
Deny: decline the call.
Auto, where available, lets eligible tools run without repeated approval prompts. Connected-account permissions, organization restrictions, saved blocks, and tool-specific approval requirements still apply. Approval behavior depends on the tool and Session settings.
Use the Connectors guide to set up accounts and check available tools. Automation reports use their configured delivery destinations; those settings are separate from the accounts selected for Session connector actions.
Attaching and opening files
Use Attachments in the plus menu to add supported data files, documents, PDFs, or images. The composer accepts up to 10 attachments per message, with each file smaller than 10 MiB. Follow any additional limits shown by the uploader; the per-message limit does not describe the total number of files a Session can retain.
Open a generated file to preview it where supported, or download it. Ask for a PDF report, slide deck, HTML dashboard, chart, or CSV and specify the intended audience and the source links you need. Review the file's numbers, quotes, citations, and formatting before distributing it.
Reuse a chat upload in a new automation: ask Agent to create the automation and name the file it should use. Review the saved instructions and attached file list. For an existing automation, add, replace, or remove reference files in its editor and Save. See Attach reference files to an automation for steps, formats, and limits.
Reading the work trail
The Session shows progress, tool activity, and results as the Agent works. For some tasks, it may delegate research to sub-agents. Use the visible activity to understand the work performed, then inspect the evidence linked from the answer.
Checking answers and evidence
The Agent uses citation checks and source links to support answers. Generated responses and files can still be inaccurate or incomplete. Check important claims before acting on them:
Check scope. Confirm the sources, date range, product or topic, customer segment, and other filters match your question.
Check counts. Open linked counts or analyses and confirm what is being counted, such as feedback records, users, or accounts. Check whether the answer describes a count, proportion, trend, or estimate.
Check quotes. Open the source link to review the original text and available attribution. A quote may be an excerpt; distinguish source language from the Agent's interpretation.
Check coverage. A zero-result answer applies to the searched sources, filters, and time window. Review data or query limitations before concluding that an issue does not exist.
Check exports. Review source links and caveats in the generated file. If evidence is missing or unclear, ask the Agent to provide it or revise the result.
Saving, sharing, and continuing
Sessions are saved automatically. Reopen a Session from Sessions to review earlier messages and continue when the available controls allow it. Restate important constraints if you change the scope or return to a long conversation.
Check the current Share setting:
Private: the Share control describes access as “Only you have access.” Use this option when available to restrict access.
Shared: people in your organization with the link can view the Session. Check the current state before distributing its link.
Use Duplicate in the Session header to create a separate Session, or use Fork session on a completed response when available. Teammates can use these controls, when offered, to continue the analysis in their own Session. Sharing a view and creating a separate Session are distinct actions.
Find Sessions and work on mobile
In the Sessions section, choose Search sessions, enter a search term, and open a result to return to the conversation. Clear the search to return to the normal list. Sessions save automatically; search is a way to find saved work.
To investigate one feedback record, open the record and choose Ask Agent when offered. Review the context in the prompt before sending.
On mobile, open the navigation menu to reach Sessions. A file preview can open over the chat; close it to return to the conversation. Available actions follow the Session's access and the controls shown in your layout.
If a control or result is missing
Agent is unavailable: contact your Enterpret team to confirm access.
The model picker is disabled: finish the active turn or resolve the pending action, then follow the control's message.
A Skill or Connector is missing: check the options offered to your workspace and the connected account's status and permissions.
A file will not upload: check its supported type, size, and the number of attachments in the message.
A shared Session is read-only: use Duplicate or Fork session when offered to work in a separate Session.
The evidence does not support the answer: ask the Agent to show the sources, clarify its scope, or correct the result.
For further help, contact your Enterpret team with the Session link, your request, and the result or behavior you expected. Return to Getting Started with Enterpret Agent for the onboarding walkthrough.
Sessions and scheduled reporting
Use an Enterpret Agent Session to investigate feedback and refine a reporting brief. Move recurring reporting, daily summaries, and weekly digests to scheduled Agent automations. Configure the schedule, timezone, destination, connectors, and saved tool permissions on the automation itself, then test it. Session settings and approvals do not automatically apply to unattended work.