Agents are automated coworkers: they are created as another member of the team, assigned work just like a person, and respond in the task discussion. The feature is enabled separately for each organization.
What is an Agent?
An Agent is an automated member of your team. It has its own user account, Staff profile, roles, and permissions, and appears in search results like any other colleague.
The difference is how it works: instead of opening Sytex and clicking through screens, it reads its mission, reviews the object assigned to it, and takes action. It then reports what it did in the discussion.
Most importantly, an agent can never see or do more than its roles allow. It follows exactly the same permission layer as people.
How to give it work
There are two ways, both of which work just like assigning work to your teammates:
Assign an object to it: set the agent as the assignee of a task, workflow, or form (or as a form reviewer). From that moment on, it reacts to changes in that object.
Mention it in the discussion: type @ followed by its name in a task or workflow chat, or reply to one of its messages. It will answer just like a person would.
Assigning an agent to a task does not make it respond to every comment. If two people are talking in the discussion of a task assigned to the agent, it will not interrupt. It only joins when you mention it or reply to something it said.
What activates it
When the agent is assigned, it runs automatically when the object it is responsible for changes: the task, workflow, form, or form answers.
There are two limits to keep in mind:
A task without an availability date has not been released yet, so it does not activate anyone.
A closed workflow does not activate an agent either.
However, if you mention the agent in the discussion, it will always reply—even if the task is not available yet or the workflow is closed. An explicit request from a person always gets a response.
Create an agent
Create agents from the Agents menu. You can find it in the settings menu or by searching for “agents” or “artificial intelligence.” You need permission to view and manage agents.
The form has three sections.
1. Mission
Name: how everyone will see and mention the agent. A name that describes its purpose works best, such as “North Region Form Reviewer” or “SLA Monitor.”
Mission and instructions: the core of the agent. Explain what it should do, how it should approach the work, what it should consider, and what it should avoid. The more specific you are, the more predictable and useful the agent will be.
2. Access and knowledge
Roles: the permissions the agent works with. They define exactly what it can see and change.
You can only give it roles that you already have in the organization. You cannot delegate permissions you do not have.
They must be general-context roles. A role limited to a project or operational unit cannot be assigned to an agent.
Pinned skills: pages from the knowledge base that the agent always reads as its work manual. Add your company procedures here: how to complete a form, which approval criteria to use, or which naming convention to follow.
You can only pin pages that you are allowed to view.
3. Execution limits
These are the safeguards. It is best to start conservatively:
Reasoning effort: how much the agent thinks before acting. More effort means better judgment on complex tasks, but it is slower and more expensive.
Timeout (seconds): how long an execution can run before it is stopped, between 30 and 600 seconds.
Runs per day: the daily budget. The default is 20. Once it is used up, the agent stops until the following day.
Cooldown (seconds): how long it waits between executions on the same object. This prevents a burst of changes from triggering many executions in a row.
Who can use it?
Each agent has its own access setting:
Everyone: anyone in the organization can assign work to it or mention it.
Restricted: only its creator and the people added to the access list can use it.
If someone without access tries to assign a task to the agent, Sytex rejects the assignment.
Track what an agent does
The agent window
Open an agent from search to see its full execution history, from newest to oldest. Each execution shows:
The object it worked on
The status: pending, running, success, error, timeout, needs a person, or skipped
Its own summary of what it did
The changes: every object it actually modified
For a conversation, the message that triggered it
You can also see its daily usage, including the execution budget already consumed, and its current configuration in the side panel.
The report in the discussion
Whenever an execution finishes, the agent posts the result in the task or workflow discussion, signed with its name. The person who assigned the work also receives a notification.
Forms do not have their own discussion. If the form belongs to a task, the report is posted in that task's discussion. If it is a standalone form, only a notification is sent.
Limits and security
Three safeguards are always active:
1. Role permissions. An agent cannot view or modify anything its roles do not allow, just like a person.
2. Scope of the assigned object. An execution is triggered by one specific object and can only modify what belongs to that object: the task and its forms, answers, and messages. Nothing above it or alongside it. If the agent attempts to go beyond that scope, the attempt is recorded in the execution.
3. Execution limits. Timeout, daily cap, and cooldown limit how much it can work.
In addition, if the organization runs out of AI credits, executions are skipped for that reason until credits are replenished.
Common use cases
Review forms: the agent acts as a reviewer, checks the form against a procedure pinned as a skill, and returns its observations in the discussion.
Monitor progress: when assigned to a workflow, it reviews dates and statuses after each change and reports deviations.
Complete repetitive data: fill in or standardize task fields according to fixed rules.
Answer questions about a task: mention it in the discussion and it responds using the object's information.
The best first agent is small and boring: one specific review, a written procedure, a low daily cap, and access restricted to your team. Once you trust its work, you can make it available more broadly.
FAQs
I do not see the Agents menu. The feature is enabled separately for each organization. If you do not see it, contact us and we will enable it.
Can I give an agent more permissions than I have? No. You can only delegate roles that you have in that organization, and only general-context roles.
Can an agent break something without me knowing? Everything it changes is listed in the execution, and the result is posted in the object's discussion under the agent's name. The history is complete and cannot be edited.
Why did my agent not react to a change? The most common causes are: the task has no availability date, the workflow is closed, the daily execution budget has been used up, the agent is in its cooldown period, or the change is outside its role permissions.
Can one agent reply to another agent? Yes. Agents can mention each other like any team member, but use this carefully: daily caps and cooldowns are the only controls that stop a back-and-forth conversation.