Feature tour: agents overview

Meaning

Documents need automation that scales. Manual review cannot keep up with large piles of invoices, contracts, and scans. An agent is the professional automation unit for Document AI. It is not a vague script. It is a named configuration with a kind — extract, split, trust, or redact — and a clear scope. When agents run, they apply the same rules every time inside the workspace rights that protect the files. Think of agents as specialist workers assigned to document tasks. Operators should understand this page before changing production settings. A clear mental model prevents accidental misrouting and keeps Document AI predictable for every team that relies on the same workspace boundary. In PaperOffice, this page is part of a larger Document AI system: workspaces hold documents, connectors feed them, agents automate them, and document types classify them. Understanding this page means understanding how your team keeps professional control over information.

Why

Agents matter because work must be repeatable. Extraction and checks run the same way every time. Clear kinds keep families apart: IDP extract, PDF split, trust verification, and redaction. Scoped execution means agents run only where documents and rights allow. Teams stop reinventing one-off processes and gain visible automation in one list. That is why the agents overview is the control room for Document AI automation. That operational clarity is what separates professional Document AI from ad-hoc file sharing. When the purpose of the page is explicit, teams spend less time searching and more time completing the real work. Use this page as the daily control surface, not as a rarely opened settings screen. The more consistently teams work here, the fewer exceptions accumulate outside the intended process.

Flow

Agents follow a clear flow. Define name, kind, and configuration. Bind the agent to the right workspace scope. Run so documents trigger automated processing and outputs. Search finds agents by name as fleets grow. Filters by status and kind keep the list readable. Opening an agent by name edits configuration — selection stays for list operations. Predictable interaction keeps automation safe. Keep this flow in mind when onboarding colleagues. If someone only learns buttons without the flow, mistakes appear later in status columns and review queues. Practice the happy path once with a real example from your organization. Real examples beat abstract training because status, labels, and rights become concrete.

Rights

Control and visibility stay explicit. Only permitted users manage configurations. Active and inactive states appear in the list. Kind filters separate automation families. Agents do not bypass workspace membership — they operate inside allowed boundaries. Review output targets before activation. Treat agents as privileged automation, not casual experiments in production. Security is not an afterthought on this page. Every list, filter, and open action should reinforce the same rights model so automation never outruns authorization. If a colleague cannot see an entry, that is usually membership — not a broken filter. Confirm access before assuming a technical failure.

Create

Creating an agent is deliberate. Pick the kind: extract, split, trust, or redact. Name it so teams recognize the purpose. Configure fields, instructions, and outputs. Activate only when the scope is correct. Do not leave incomplete agents enabled. A clear kind and name prevent wrong automation on live documents. Take a moment after create to confirm the first result looks correct. Early verification is cheaper than correcting a fleet of misconfigured entries later. Prefer fewer, well-named entries over many vague ones. Names and scopes that teams recognize reduce support tickets and accidental edits.

Operate

Operate every day with search and filters before scrolling. Open an agent by name to edit. Keep inactive agents visible until decommissioned. Watch status when document volume changes. Pause agents when workspace scope changes. Healthy automation is quiet — surprises in outputs mean configuration needs review. Healthy daily operation is quiet. When indicators stay green and lists stay readable, Document AI remains trustworthy for auditors and for the people who depend on it. Schedule a short weekly review of this list when volume is high. Ten minutes of hygiene prevents hours of firefighting later.

Abschnitt

Open Agents, create one agent with a clear kind, and bind it to the right workspace. Activate only when configuration is complete. PaperOffice keeps automation visible and scoped — so Document AI work stays reliable. Start with one deliberate action today, verify the outcome, and only then scale the pattern to the rest of the team.