At a glance
WHAT IT’S REALLY ABOUT
How OpenAI uses ChatGPT Sites for secure, shareable personal software
- Kath Korevec (Product Lead for Sites) explains how OpenAI teams use ChatGPT Sites as a place to deploy Codex-built software quickly, ranging from internal tools to playful consumer projects.
- A new Enterprise feature, “Plugins in Sites,” lets each viewer bring their own authenticated connectors (Slack/Notion/Calendar, etc.) so the same site displays role-appropriate data without sharing API keys.
- Kath demos an incident-command overview site that aggregates incident timelines, owners, runbooks, and tooling to keep a PM informed without disrupting on-call workflows.
- The conversation highlights Sites as infrastructure—hosting plugins (including MCP), data, and evals—alongside built-in components like D1/R2 and collaborative editing.
- They explore creative use cases such as automated Spotify playlist curation, community-shared game rooms via a “skill,” and experiments to embed inference directly inside sites for real-time generative experiences.
IDEAS WORTH REMEMBERING
5 ideasSites is positioned as the deployment layer for Codex-built software, not just a page builder.
Sites lets builders publish software created with Codex without wrestling with the “last mile” of deployment and hosting. This turns what would otherwise be local prototypes into shareable URLs that can become real internal tools or public demos.
One shared site can securely personalize data per viewer through inherited connector auth.
With “Plugins in Sites” (Enterprise), visitors authorize their own connectors (e.g., Slack/Notion/Calendar) so the same site renders different, permission-scoped data per user/team. This reduces the need to manage API keys and avoids leaking data across roles.
Sites can act as an incident “cockpit” that summarizes operations without adding process overhead.
Kath’s incident-command site pulls timelines, owners, runbooks, and guides from Slack/Notion and provides a high-level, minute-by-minute view so she can stay informed without interrupting engineers. It’s an example of a lightweight executive/PM interface built atop existing systems of record.
Sites doubles as infrastructure—hosting plugins/evals/data—not only as an app surface.
Beyond websites, Sites is used to host plugins, data, and evals; it includes built-in primitives like D1 data and an R2 bucket and supports co-editing. OpenAI also announced hosting MCP plugins through Sites, creating a loop where Sites can host the tooling used by other Sites.
Sites enables throwaway, personalized micro-apps that reshape everyday workflows (even for consumers).
Kath built a weekly “heavy rotation” music site that uses automated discovery (e.g., Reddit top playlists via computer-use) and generates a fresh playlist every Monday morning. The emphasis is on “personal software” that matches how an individual thinks/works and can be disposable (built for a trip/week, then deleted).
WORDS WORTH SAVING
5 quotesWe just launched something today which is Plugins in Sites. It will allow anyone to come to your site and use their connected plugins, and so you see only your data on that site.
— Kath Korevec
They'll be looking at the exact same site, but they're gonna see content based on what their team is.
— Kath Korevec
My brain works in different ways than other people, and I might just want like a different view of Google Calendar for the day or for the week. I might be like going on a, on a business trip, and I just want like things organized differently, and so I'll build a site for that week, and then I'll throw it away at the end of the week, and it's wonderful.
— Kath Korevec
There's something about just, like, the organic community. I don't need it to be... I don't need the community to come together in an AGI way. I'd rather just, like, meet the people. Maybe I'm old school like that.
— Kath Korevec
Because that's... I wanna protect that. That's my voice. I don't want, if it's, you know, like signing my name to something it wrote and then sending that to somebody, that's crossing the line. So I say, "Do not do that."
— Kath Korevec
High quality AI-generated summary created from speaker-labeled transcript.
