Glossary
These definitions cover the concepts that come up most often in Zahen. Longer explanations are linked where they exist.
Access level A label on every document that controls who can read it: Public, Employee, Department, or Restricted. Access levels are set by administrators and enforced before any search runs. See Why you can’t see some documents.
Agent task A multi-step piece of work that Zahen carries out on your behalf — for example, drafting a response or looking up a policy and preparing a summary. Unlike a quick question, an agent task progresses through a series of steps, can pause for human approval, and its progress is saved at each checkpoint. See Agent tasks.
Approval A decision made by an authorised person before a high-risk step in an agent task can continue. An approval can be granted, rejected, sent back for revision, delegated, or escalated. Tasks wait in a paused state until an approval decision is recorded. See Approvals.
Assistant (also called a plugin) A configured agent that handles a particular type of work — for example, answering HR policy questions or drafting customer emails. The platform may have more than one assistant; which ones you can use depends on your role.
Audit log An append-only record of meaningful activity on the platform: questions asked, tasks run, approvals decided, documents uploaded, and more. Because it is append-only, entries cannot be edited or deleted after they are written.
Citation A reference, shown inline in an answer or task output, pointing to the specific document and passage the assistant used. Citations let you verify any answer against its source.
Department An organisational unit — for example, Human Resources or Finance. Documents and users belong to departments. Your department membership determines which department-level documents you can read, and which tasks you can run. See Why you can’t see some documents.
Employee mode A simplified view of Zahen designed for everyday staff who mainly ask questions and check answers. It surfaces the most common actions without the more advanced controls visible to admins and developers.
Grounding The practice of building every answer from your organisation’s approved documents rather than from the model’s general training or the open internet. A grounded answer always cites the documents it drew from. See Why answers are trustworthy.
Notification An in-app alert (shown by the bell icon) that tells you something needs your attention — for example, a task awaiting your approval or a task that has completed. Notifications are best-effort; if one doesn’t arrive, check My requests or the approvals queue directly.
Retrieval (RAG) Short for Retrieval-Augmented Generation. The technical process behind grounding: Zahen searches your approved documents for the passages most relevant to your question, and the assistant uses those passages to compose its answer. Retrieval is always filtered by your role and department before the search runs.
Risk level A rating on a tool or action that determines how it is handled: low-risk actions proceed automatically; medium-risk actions may log a warning; high-risk actions pause and wait for a human approval before continuing. See Risk levels & approvals.
Role (RBAC) A named set of permissions assigned to you by an administrator. Your role controls what you can read, which tasks you can run, and which administrative actions you can take. You can hold more than one role; if you do, you get the combined capabilities of all of them. See What your role lets you do and Roles & permissions.
Step (node) One stage in an agent task’s execution — for example, “retrieve relevant policies” or “draft a response”. Each step is visible in the task timeline, and progress is saved after every step so a task can resume cleanly if interrupted.
Tool A specific capability the assistant can call during an agent task — for example, looking up an HR policy or sending a draft email. Every tool is explicitly approved by an administrator, schema-validated, and recorded in the audit log when used.