Agents
Agents are custom AI personas you build through conversation in Amazon Quick. Each agent has its own instructions, knowledge sources, and capabilities, and can work with you in chat or run on its own on a schedule. An agent can access only the resources and services that you can access, and you control what each agent is allowed to do.
Creating an agent
To create an agent
-
In the left navigation, choose Customize, and then choose Agents.
-
Choose Create, and then choose From chat (or say "create an agent" in chat). To import an existing agent definition, choose From file instead.
-
Describe your goal in plain language. For example, "I need an agent that monitors competitor pricing and sends me a weekly summary."
-
The builder presents templates you can use as a starting point, or you can describe what you want from scratch.
-
A configuration panel opens alongside the chat. As you converse, the panel fills in with the agent's definition.
-
Review the configuration. Your edits are saved automatically as a draft. Choose Publish to make the agent available to others, or Discard to throw away your draft edits.
Tip
You can upload reference documents (CSV, DOC, DOCX, HTML, MD, PDF, and TXT, up to 10 files) during building. The builder reads them and incorporates relevant content into the agent's instructions.
To add a shared agent
-
In the left navigation, choose Customize, and then choose Agents.
-
Choose Browse all. An Add agent panel opens showing agents shared with you in your organization.
-
Search or browse the list of available agents.
-
Choose Install next to the agent you want to add.
Installed agents appear in your agent list and are ready to use. To remove an installed agent, choose the options menu on the agent and choose Uninstall. This removes the agent from your list.
Agent configuration
The configuration panel has two tabs.
The Instructions tab has these fields.
| Field | Description |
|---|---|
| Name | Agent display name. |
| Description | What the agent does. Also serves as the agent's self-introduction. |
| Instructions | The agent's core prompt: goals, behaviors, workflows, and edge cases. A rich text editor with formatting support (headings, lists, emphasis, code, and quotes), up to 50,000 characters. You can also set a welcome message with starter prompts. |
The Capabilities tab has these fields.
| Capability | Description |
|---|---|
| Web search | Toggle to allow the agent to search the web. |
| Connections | Action connectors the agent can use (Slack, Outlook, Jira, and others). |
| Skills | The skills the agent can use. By default an agent can use all skills; if you add specific skills, the agent is limited to those. Remove them to re-enable all skills. |
| Spaces | Connected data sources the agent reads from. |
| Additional documents | Uploaded files that guide agent behavior, such as templates, workflows, and examples. |
Using an agent
You can use an agent in one of the following ways.
-
Agent switcher: choose the agent name at the top of the chat window to open the agent switcher. This is your list of installed agents, the agents that are ready to use. Browse them by Recent, Favorites, or My agents tabs, or search by name. The agent you are currently using is marked, and your Default agent (the one Quick starts new chats with) carries a Default label. Choose Browse all agents at the bottom to find and install more.
-
@ mention: type @ in the chat input to reference one of your installed agents right inside a message. This is a quick way to ask that agent a question or hand off a task without switching your whole conversation over to it.
-
Chat button: on the agent's landing page, choose Chat to start a conversation.
You can also tailor which agents you see and reach for most.
-
Favorite an agent: choose the star icon next to an agent in the switcher or in your agent list to add it to your Favorites for quick access. Choose the star again to remove it.
-
Change your default agent: in the switcher or your agent list, open an agent's options menu and choose Make default. Quick then uses that agent for new chats until you change it.
Once you select an agent, you can ask questions and assign tasks in real-time dialogue, add files, spaces, or connections at any point during the conversation (the agent incorporates them without delay), and view citations that reference the agent's source material.
Autonomous execution
When the agent has routines configured, it can run on its own. The agent does the following.
-
Executes on its configured routine without user prompting. Routines run in the cloud, so they continue on schedule even when your laptop is closed or offline.
-
Proposes routine changes based on what it finds. The agent communicates what changed and why, and you confirm before the change takes effect.
-
Stores deliverables in its Agent Files.
-
Surfaces results on its Agent briefing page (when briefing is turned on) and on its activity page in Mission Control.
Monitoring agents
You can monitor agent activity in two places.
-
Agent landing page: the agent's overall page, with Overview, Activities, Agent Files, and Settings tabs. View a specific agent's runs, approvals, routines, and deliverables.
-
Mission Control: a centralized command center that gives you visibility across all your agents in one place. Monitor all agent activity, unblock agents that need input, view running agents, manage routines, and track KPIs, without switching between individual agent pages. For more information, see Mission Control. Access Mission Control from the icon in the top-right navigation bar.
Agent results can also appear in your main activity feed alongside other notifications.
Agent landing page
Each agent has a landing page with four tabs.
| Tab | Description |
|---|---|
| Overview | Summary of the agent's purpose, items needing your input, and an AI-generated agent briefing. |
| Activities | Activity log with filter tabs: All, Needs input, Running, and Routines. |
| Agent Files | Persistent file storage: artifacts and deliverables the agent has produced across runs. |
| Settings | Agent preferences, such as automatic briefing updates. Instructions and capabilities are edited from Configuration, not here. |
The Overview tab includes a Needs input section, where an approval card appears when the agent pauses on an action that requires your decision (each card shows a description of the proposed action with Approve and Deny buttons), and an Agent briefing section, an AI-generated summary of the agent's work (for example, top findings, trend analysis, or status dashboards) that updates after each run and can be regenerated on demand.
The Activities tab shows past and current agent work, filtered by sub-tabs: All (all activities in reverse chronological order, each with a title, summary, agent name, and timestamp), Needs input (items where the agent paused and is waiting for your decision), Running (tasks in progress), and Routines (the agent's configured routines). Time-based routines (schedules), for example "Every Monday at 9 AM" or "Every weekday at 8:30 AM," are supported on the desktop. Each routine has a play button to trigger a run and a toggle to turn it on or off. Choose Create to add a new routine. Activities can include file attachments, such as drafted documents or reports, that you can open and review.
Run status
Each run in the Activities tab shows one of the following statuses.
| Status | Description |
|---|---|
| Running | Execution in progress. |
| Needs input | Run finished or paused, but one or more actions are awaiting your decision. |
| Completed | Run finished with no issues. |
| Interrupted | Run was stopped before completing. |
| Failed | Run ended with an error. |
| No runs yet | The agent has not run yet. |
Managing agents
You manage your agents from the left navigation, under Customize > Agents.
-
Search and filter: find agents by name or description, or filter by Favorites, Owned by me, or Shared with me.
-
Chat: open the agent in a conversation.
-
Configuration: open the agent's configuration to view its instructions and capabilities, switch between the published and draft versions, or choose Edit to make changes.
-
Run now: trigger an immediate run from a routine's play button on the Activities > Routines tab.
-
Share: share the agent with individuals or groups in your organization, each as a Viewer or Owner.
-
Manage versions: switch between the published and draft versions of the agent.
-
Manage routines: view and edit the routines that run the agent on its own.
-
Duplicate: create a copy of the agent that you own.
-
Export: export the agent's configuration to a JSON file.
-
Uninstall or delete: use the options menu to uninstall a shared agent from your list, or delete an agent you own.
For others to use your agent's connected sources, you must share those sources with the intended recipients. Otherwise, they receive responses only from the sources they already have access to. You do not need to share all underlying resources.
Draft and publish
When you edit an agent, your changes are saved automatically as a draft until you choose to publish. The following options are available from the editing header.
-
Draft version selector: a Draft dropdown in the header lets you switch between the draft and published versions.
-
Release notes: optional notes shown to your team when you publish.
-
Publish: push changes to everyone who uses the agent.
-
Discard: throw away draft edits.
Routine runs always use the published version. If no published version exists, routines do not run.
A published agent is also available from Quick on the web.
