Pentra: conversations, agents and workflows
Where your conversations with Pentra are kept, and how it runs your organization's agents and workflows for you.
Your conversations are kept
Every conversation with Pentra is kept in your account, on any device, until you delete it. Only you can see them. A conversation started on a project view is about that project; one started anywhere else is about all the projects you can open. It keeps that scope: if you move to another project, it says so and offers a new conversation there.
New chat starts another conversation and keeps the current one. AI, Assistant lists them all, newest first, with search; each one can be renamed, archived (out of the list, still searchable) or deleted after a confirmation. The floating window shows your recent ones under History.
Running an agent or a workflow
Ask in your own words: "run the risk agent on this project", "run the weekly report workflow". Pentra finds the one you mean -- and asks if more than one could match -- then shows a preview of what it will do and what it may change. Nothing runs until you press Run, and Depentra checks again at that moment that you may run it there.
Who may run one is the organization’s choice, under Organization, Overview, Who may run AI automations: owners and admins only; owners, admins and Project Owners, for the agents and workflows of the project they own (the default); or a custom choice that can also let Editors, and guests, run them on a project they have that role on. Contributors and viewers never run them, and anything organization-wide stays with owners and admins. One agent or workflow can be held to owners and admins.
A project agent runs on the project it watches. An agent that watches a workspace or the whole organization looks at each project in it that you can open (up to 25 a run), never more than you could see yourself. A workflow can be run this way only if its trigger is When someone runs it; one that starts on an event runs when that event happens, and one on a schedule runs at its time. If a workflow asks for inputs, such as a week or who to tell, you are asked for them first, and it does not start until the required ones are filled in. You can also run it from AI, Workflows, with the same form.
The run appears in AI, Runs, with who asked and from which conversation, and its result comes back into the conversation: what it found, what it changed, what failed. It updates as it goes. Changes an agent prepares, and workflow steps that wait for approval, change nothing until you approve them. Deleting a conversation never deletes its runs.
Stopping a run
Stop run, on the run in the conversation or in AI, Runs, stops it. A run that is waiting, for an approval or for a date, stops at once, and a change it proposed is withdrawn. A run that is working stops at its next safe point: before its next step, stage or project. Nothing it already did is undone, and the run shows exactly what happened before it stopped. The person who started a run, and anyone allowed to run that agent or workflow, can stop it.
Whose AI credits it uses
In a Team organization, Pentra, agents and workflows used by its paid members draw on the Team’s shared AI pool: each active paid member adds their tier’s monthly credits (a license nobody has taken yet adds nothing). AI → Usage shows the pool, where it went by source, and your own share; owners and admins also see it by person, agent and workflow. Anywhere else, your own workspace, or as a guest, your own monthly allowance is used.