Kiteworks Agent Marketplace
Which AI assistant do you use?
Guide

Governed AI agents for sensitive information

A buyer guide to evaluating marketplace agents, platform boundaries, Claude execution, and a practical conversation before setup.

When a team considers an AI agent for business files, the first question is usually not “How do we install it?” It is “What would we be asking it to see or change, and how would we keep a person in the decision?” The Kiteworks Marketplace is a place to inspect that question. It presents agents with a purpose, current status, examples, and the boundaries stated by their instructions. You can read the resources and talk with the team without having Claude, an MCP connection, or a Kiteworks account ready.

Start with the boundary

The marketplace describes the agent package and the way an agent is intended to work. The Kiteworks connector is the separate path to files and folders. Your assistant session, connector permissions, folder scope, and the agent instructions all affect what a run can observe or do. A listing is therefore a useful starting point for evaluation, not a promise that every file, membership, policy, or business decision will be visible.

The current supported execution surface is Claude. The marketplace lists other assistant brands, but they are not supported for running these agents today and no delivery date or order has been published. This distinction is part of a responsible buying conversation: choose an agent because its stated job fits your process, then confirm that your approved assistant and connector arrangement can run it.

Use a realistic scenario

Suppose a security team wants to understand sharing exposure in a folder. Sharing Auditor can inspect the folder-level sharing signal and offer a report. The signal can cascade from a shared folder to descendants, while the connector does not expose who the members are or whether they are internal or external. Direct file-level sharing is outside that signal. A human still decides whether the observed exposure is appropriate and what should happen next.

For a finance workflow, Invoice Organizer can find invoices and receipts, including photos and scans, and extract vendor, date, amount, and tax information. Renaming and the categorized spreadsheet happen after approval. For legal operations, Contract Radar can find contracts, agreements, and renewal-related documents and save a report. Neither description replaces a review of source documents or an approval policy.

Ask these questions before a pilot

  • What folder scope and connector permissions will the agent receive?
  • Which observations are file-visible signals, and which questions require a person or another system?
  • What action, if any, needs explicit approval before it changes names or writes a report?
  • How will the team record the source material, exceptions, and unresolved questions?
  • Which assistant is approved for execution? If it is not Claude, what is the decision path for a supported run?
  • What would make the pilot useful without claiming a saving, accuracy rate, or compliance result in advance?

Keep the pilot small enough to inspect. Compare the agent’s report with the files and decisions your team would normally review. Record gaps, false assumptions, and work that remains manual. That evidence is more useful than a generic promise about sensitive information.

If you want to discuss a folder scenario, approval boundary, or assistant constraint, email sales@kiteworks.com or start a conversation. You do not need Claude or MCP setup to ask. Browse the agent catalog and assistant support while you prepare the questions.