Amazon Web Services pushed a desktop version of its Amazon Quick AI assistant onto macOS and Windows on Tuesday, putting an always-on agent on user laptops that reads local files, watches calendars and inboxes, and builds what the company calls a personal knowledge graph. The app, announced at AWS's What's Next event in San Francisco alongside an expanded OpenAI partnership and a rebranded Amazon Connect family, is in preview, currently available only in US East (N. Virginia). Mac support is Apple Silicon only. Intel Macs need not apply.
This is the second act for a product that launched six months ago as a browser-based productivity layer. The desktop app is the swing: instead of asking you to switch tabs, Quick now sits behind your other windows, scheduling agents in the background, surfacing meeting prep without being asked, and connecting to a growing list of SaaS tools through native integrations.
The knowledge graph, on your laptop
The headline feature is what AWS describes in its knowledge graph docs as a force-directed graph of entities and their relationships, automatically extracted from connected sources like Slack, email, and local folders. Ask Quick what's happening on a given project and it pulls related people, recent meetings, and associated documents into the answer. The graph gets denser the longer you use it (or, if you'd rather not have an entity-relationship dossier of your work life building up in the background, you can toggle ingestion off per source).
The interesting wrinkle: the graph is stored locally. AWS says conversation history, memory, the knowledge graph, and file indexes stay on the device and never get uploaded to the cloud or used to train models. That's a meaningful claim if it holds up, and one that quietly distinguishes Quick from most enterprise AI tools, which treat your context as something to ship to a vector database somewhere.
How well the graph actually works at scale is another question. AWS isn't showing benchmarks. The customer testimonials in the official announcement lean on phrases like "minutes instead of hours" and five-plus hours saved weekly, which sound great until you ask: five hours compared to what baseline? Mondelez CTO Chris Hesse said Quick reduced task completion times for employees doing data analysis. That's the kind of carefully framed claim that's hard to argue with and impossible to verify.
Late to a crowded room
Amazon is not first here. Anthropic's Claude Cowork shipped as a research preview in January 2026 and went generally available across paid tiers earlier this month, with Computer Use letting the agent operate a Mac or Windows desktop directly. Microsoft Copilot has been embedded in 365 for over a year. OpenAI's ChatGPT desktop app has been around for roughly two years.
What AWS is selling, then, isn't novelty. It's enterprise plumbing. The pitch to large buyers, per the product page: same security and governance posture as your AWS account, native connectors for Slack, Teams, Outlook, Gmail, Salesforce, ServiceNow, Asana, Jira, Microsoft 365, Google Workspace, Zoom, Airtable, and Dropbox, plus integration with developer tools including Kiro CLI and Claude Code. The desktop app supports local Model Context Protocol connections, which is becoming table stakes for agentic tools.
And the pricing is aggressive. New Free and Plus tiers, announced the same day, let individuals and small teams sign up without an AWS account or a credit card. Professional and Enterprise plans go through the AWS Console at $3 per agent hour for additional usage, on top of a $250-per-account monthly infrastructure fee, per the Quick pricing page. The free tier is the obvious land grab.
Proactive mode and the trust problem
The proactive mode is where this gets interesting, and a little nerve-wracking. Quick monitors connected apps in the background and surfaces things that need your attention, which sounds useful until you remember that the agent has access to your email, your Slack DMs, your calendar, and your local file system simultaneously. The privacy story rests on AWS's word that nothing leaves the device for training purposes. For enterprises that have spent the last two years writing policies about what can and can't go into ChatGPT, that's either the entire point or a brand-new procurement headache. Probably both.
There's also a custom apps feature in preview that lets users build dashboards and web pages with live data connections from natural-language descriptions. AWS pitches this for sales pipeline trackers and finance dashboards. Whether non-technical users actually build production-quality internal tools this way, or just generate slightly more sophisticated mockups, is the kind of question that takes about a year of customer reference checks to answer.
What's missing
The Microsoft 365 extensions, where Quick would actually live inside Outlook, Word, PowerPoint, and Excel, are still in preview. So is the desktop app, per the preview release. So are custom apps. So are the Microsoft 365 extensions. The list of generally available features today is noticeably shorter than the list of preview features, which is a familiar pattern for big AWS launches.
Apple Silicon-only on Mac. US East only on regions.
Whether Amazon, with its uneven track record on consumer-facing software, can build the kind of always-on agent that enterprise users will actually trust to read their inbox in the background, is the question that won't be answered until the preview stickers come off. AWS hasn't said when that happens.
