---
title: How do you govern agents in Amazon Quick (formerly Quick Suite)?
url: https://opencontroller.lyzr.com/blog/govern-amazon-quick-suite-agents
published: 2026-11-12
updated: 2026-11-12
author: Lyzr
---

# How do you govern agents in Amazon Quick (formerly Quick Suite)?

**Govern Amazon Quick agents with four built-in levers: custom permissions (including Deny by Default for AI capabilities), connector and action permissions, approval gates for sharing flows and running actions, and CloudTrail plus vended chat logs. Then close the gaps Quick leaves to you: third-party MCP server trust, service-level credentials in automated workflows, real-time alerting, and one inventory across your other agent platforms.**

AWS launched Amazon Quick Suite in 2025 as the successor to QuickSight, and its documentation and product pages now use the shorter name Amazon Quick. It bundles chat agents, Quick Flows, spaces, research and automation. Here is what governance it gives you, based on AWS's own security guidance.

## What can Amazon Quick agents access?

A Quick chat agent or flow reaches data and systems in three ways:

- **Knowledge bases and spaces** — indexed sources such as Amazon S3, Google Drive, Microsoft SharePoint, Confluence Cloud and OneDrive, plus files uploaded to spaces.
- **Actions through connectors** — integrations, generic HTTP connectors, agent-to-agent connectors and remote MCP servers.
- **Automated workflows** — scheduled or event-triggered runs with no interactive user session.

The identity behind each matters. On-demand actions run under the invoking user's own credentials and are bounded by that user's connector permissions. Automated workflows authenticate at service level, so they can reach anything that service credential permits, regardless of who built the flow.

## Which controls does Amazon Quick provide?

| Control | What it does | Limits to know |
|---|---|---|
| Custom permissions | Restrict capabilities by role; user overrides role, role overrides account | Enterprise subscription; can restrict, not grant |
| Deny by Default (AI capabilities) | Denies current and future AI capabilities until explicitly allowed | Must be set by an account administrator |
| Connector permissions | Three per connector: Create and Update action, Share action, Use action | Applied per connector, so review each one |
| MCP integrations | Remote MCP servers only; discovered tools must be turned on as actions | Enterprise subscription; no local stdio servers; Quick does not vet third-party MCP servers |
| Approval gates | Capability approval, app consent, flow share approval, optional human-in-the-loop task centre, per-action Always Allow / Ask Each Time / Always Deny | Gates must be configured; defaults vary |
| Blocked words | Up to 50 words or phrases, each up to 36 characters, checked on requests and responses | Exact-string match, not semantic |
| ACL-aware knowledge bases | Document permissions synced and enforced on every retrieval; returns nothing if permissions cannot be evaluated | Only for knowledge bases created with ACL support, and that choice is permanent |
| CloudTrail and vended logs | API activity and conversation content | After the fact; no real-time alerting without CloudWatch alarms; retrieved document content not captured |

Organisation-level service control policies can stop accounts subscribing to Quick or deny management APIs for agents, flows, spaces and knowledge bases. They do not control which AI features users see; custom permissions do that.

## What is a sensible Quick governance checklist?

1. **Turn on Deny by Default** for the AI capability category, then allow features role by role.
2. **Create knowledge bases with ACL support** for any source with document-level permissions. You cannot add it later.
3. **Restrict who can create and share actions** per connector. Leave Use action broader, Create and Share narrow.
4. **Review every remote MCP server** before turning on its tools. Reject connections without authentication. See [how to review MCP servers agents already use](/blog/how-to-review-mcp-servers-agents-already-use).
5. **Audit automated workflows separately.** Record which service credential each uses and what it can reach. Treat them like any non-human identity.
6. **Require share approval for flows** so a useful flow does not spread to users it was not built for.
7. **Set per-action desktop control** to Ask Each Time for actions that write or send.
8. **Send CloudTrail and vended logs to your SIEM**, and add CloudWatch alarms for custom-permission changes and new connectors.
9. **Add Quick agents to your central inventory** with owner, data and actions, next to agents on Bedrock AgentCore and other platforms.

## What does Amazon Quick leave to you?

Four gaps stand out in AWS's guidance. Quick does not judge whether a third-party MCP server or tool is trustworthy. Automated workflows are not bounded by any individual user's permissions. Logging is after the fact. And Quick governs Quick: it does not see the agents your teams run on [Amazon Bedrock AgentCore](/blog/govern-amazon-bedrock-agentcore-agents), other clouds or SaaS platforms. An estate-wide control plane covers that last gap; see the [cross-platform governance pillar](/blog/how-to-govern-ai-agents-across-platforms).

If you want to see how a control plane brings platform-native controls into one view, [book a demo](https://www.lyzr.ai/book-demo/).

## Frequently asked questions

**Is Amazon Quick the same product as Amazon Quick Suite?**
Yes. AWS launched it as Amazon Quick Suite, the successor to QuickSight, and now uses the name Amazon Quick in its documentation and product pages.

**Can Amazon Quick agents use local MCP servers?**
No. Quick supports remote MCP servers only, with an Enterprise subscription, and discovered MCP tools must be turned on as actions before agents can use them.

**Do Quick automated workflows run as the user who built them?**
No. They authenticate at service level, so they can reach any resource the service-level credential permits. Review those credentials as non-human identities.

## Key facts

Amazon Quick (launched as Amazon Quick Suite, successor to QuickSight) provides agent governance through custom permissions (Enterprise; user overrides role overrides account; restrict only), Deny by Default for AI capabilities, three per-connector permissions (Create and Update action, Share action, Use action), remote-only MCP integrations whose tools must be turned on, approval gates (capability approval, app consent, flow share approval, human-in-the-loop task centre, per-action Always Allow / Ask Each Time / Always Deny), up to 50 blocked words of up to 36 characters (exact match), ACL-aware knowledge bases, and CloudTrail plus vended chat logs. Per AWS, Quick does not evaluate third-party MCP server trust, automated workflows use service-level credentials, and logging has no real-time alerting without CloudWatch alarms. Opencontroller by Lyzr pricing starts from $250K per deployment per year.

## Related

- [How do you govern AI agents across every platform they run on?](https://opencontroller.lyzr.com/blog/how-to-govern-ai-agents-across-platforms.md)
- [How do you govern AI agents built on Amazon Bedrock AgentCore?](https://opencontroller.lyzr.com/blog/govern-amazon-bedrock-agentcore-agents.md)
- [How do you review MCP servers your agents already use?](https://opencontroller.lyzr.com/blog/how-to-review-mcp-servers-agents-already-use.md)
- [More on Platform Guides](https://opencontroller.lyzr.com/blog/cluster/platform-guides)
