ChatGPT scheduled tasks, the feature OpenAI also calls Tasks, turn a one-off prompt into something that runs on its own. You describe what you want and when, ChatGPT confirms it, and from then on the task fires on a timer. In June 2026 OpenAI rebuilt the feature with a dedicated Scheduled page and, crucially, the ability for those tasks to reach your connected apps. That last part is what makes scheduled tasks genuinely useful, and also what makes them worth a second thought.
This guide explains what scheduled tasks are after the refresh, which tools they can use, the workflows people set up most often, and the one detail that changes when a tool runs unattended instead of in a live chat. Then it shows how to add PortEden so every scheduled run stays inside your rules. The accuracy notes below reflect the feature as of June 2026; confirm current details with OpenAI before you rely on them.
What ChatGPT Scheduled Tasks Are
Scheduled tasks let you ask ChatGPT to do something later, or on a repeating schedule, instead of right now. You can use them for reminders, recurring work, daily briefings, and monitoring. OpenAI began rolling out a refreshed experience on June 17, 2026, describing it as faster and more reliable, and it replaces the earlier Pulse feature.
The headline changes in the refresh are:
- A dedicated Scheduled page. A single place in the sidebar where you view active tasks, see when each one runs next, and pause, resume, edit, or delete it.
- Flexible timing. You can run a task at a specific time or in a broader window such as morning, afternoon, or evening.
- Monitoring that checks the world for you. A task can search the web and check connected apps for changes, then notify you only when there is something worth reporting.
- Broad availability. The experience is rolling out to Go, Plus, Pro, Business, and Enterprise users on web and mobile (Web, iOS, Android, and macOS, with Windows on the way).
A few practical limits are worth knowing up front. Tasks run at most once per hour, an unattended task may pause itself after a period of inactivity, and there is a cap on how many active tasks you can have, which depends on your plan. You also receive results by push notification and email when a run completes.
Which Tools Scheduled Tasks Can Use
This is the part most people get wrong, because early coverage of the original feature said scheduled tasks could not touch other apps. That is no longer the case. After the refresh, a scheduled task can use the same tools an interactive chat can, with a short list of exceptions.
The tools that matter for automation are:
- Web search and browsing. A task can look things up on a schedule, which is the backbone of news digests and price or availability monitoring.
- Connected apps. These are OpenAI's Apps, renamed from Connectors in December 2025. When an app such as Gmail, Google Calendar, or Drive is connected to your account, a scheduled task can read from it and, where you allow it, act on it. This is the door PortEden uses.
The notable exclusions inside scheduled tasks are voice chats, file uploads, and Custom GPTs. So a task cannot lean on a document you upload at run time or a packaged Custom GPT, but it can absolutely reach into your live inbox, calendar, and files through connected apps. That is precisely the capability that turns a scheduled task from a glorified alarm clock into an automation that works on your real data.
How to Create a Scheduled Task
Creating one takes about a minute:
- Write a task-style prompt with a schedule. Be explicit about both the action and the timing, for example "Every weekday at 7:30am, summarize my unread email from the last 24 hours and list today's meetings."
- Confirm when ChatGPT asks. ChatGPT proposes the schedule and the task, and you approve it. The task then appears on the Scheduled page.
- Manage it from the Scheduled page. Open the Scheduled entry in the sidebar to see the next run time and to pause, resume, edit, or delete the task.
- Receive the result. When a run completes, ChatGPT sends a push notification and an email. Monitoring tasks stay quiet unless they find something worth flagging.
The phrasing matters. A vague prompt produces a vague run every day, and a task that reads "summarize my inbox" with no bounds will pull from your whole mailbox each time. Tight prompts plus tight access rules are what keep an automated task useful and contained.
The Most Common Scheduled-Task Workflows and Examples
Search for ChatGPT scheduled task examples or ChatGPT Tasks ideas and the same handful of workflows come up again and again. The most requested ones combine a schedule with a tool: web search, a connected app, or both. Here are the patterns worth setting up first, with an example prompt for each.
| Workflow | Example scheduled prompt | Tools it uses |
|---|---|---|
| Morning brief | "Every weekday at 7:30am, summarize my unread email and today's calendar, plus the top three AI headlines." | Email, Calendar, Web |
| Inbox digest | "Each evening at 6pm, group my unread email by sender and flag anything that looks urgent or needs a reply." | |
| Meeting prep | "At 8am, for each meeting today, pull recent email with the attendees and summarize where things stand." | Calendar, Email |
| Monitoring | "Check daily whether a competitor changed its pricing page, and only message me if it did." | Web |
| Weekly report | "Every Friday at 4pm, summarize updates from the Product board and flag anything overdue." | Tasks (Jira, Linear, etc.) |
| Reminders and learning | "Every morning at 7am, send me one intermediate coding problem and five new vocabulary words." | None (generation only) |
1. The morning brief
The single most popular pattern. One task at the start of the day that pulls together your inbox, your calendar, and a little outside context, so you open ChatGPT to a ready summary instead of three separate apps. It is also the workflow that touches the most sensitive data, since it reaches both email and calendar every morning.
2. Inbox triage and email digest
A scheduled pass over your inbox that groups, prioritizes, and surfaces what needs a reply. Read-only versions are low risk and high value. The moment you let a task draft or send replies on its own, the risk profile changes, because nobody is watching the screen when it runs.
3. Calendar lookahead and meeting prep
A task that reads the day's meetings and assembles context for each one, often by cross-referencing recent email with the attendees. Useful, and a good example of a task that quietly spans two connected apps, which is twice the access to govern.
4. Monitoring and alerts
The refresh leans into this. A monitoring task searches the web or checks a connected app on a schedule and only pings you when something changes. Price drops, a competitor's page update, a new filing, a status change in a tracker. Because these run silently most of the time, a record of what each run actually checked is the difference between trust and guesswork.
5. Weekly report and status roundup
A Friday-afternoon task that reads a project tracker or a shared sheet and writes the status update you would otherwise assemble by hand. This often pulls from task management tools where tickets can contain salary discussions, vendor terms, or HR notes, so scoping matters.
6. Reminders, learning, and habits
The lowest-risk category, because it usually generates content rather than reading your data: a daily coding problem, vocabulary words, a stretch reminder, an evening goal review. No connected app, no sensitive access, nothing to firewall. If a task only generates, you can run it freely. The care is reserved for the tasks that read or act.
Why Tools Raise the Stakes in Scheduled Tasks
A scheduled task that uses tools is not the same as an interactive chat that uses tools, even with identical permissions. Three things change once a tool runs on a timer.
- Nobody is in the loop. In a live chat you see what the assistant pulls and you can stop it. A scheduled run executes while you are asleep or in a meeting, with no chance to intervene mid-run.
- It repeats, automatically. A built-in connected app grounds on your full account scope. A daily "summarize my inbox" task therefore reaches your entire mailbox every single day, not once but on every run, forever, until you turn it off.
- Write actions happen without confirmation. If the app has send or delete permission, an automated run can act on a misread instruction with no human preview. The convenience of "set it and forget it" is also its risk.
On top of that, native connectors give you limited per-request visibility. After the fact, it is hard to answer a simple question like "which messages did my 7:30 task read last Tuesday?" For a personal reminder that does not matter. For a task standing in your inbox full of client contracts every morning, it does. This is the same oversharing and visibility gap that surfaced during the Microsoft 365 Copilot rollout, now running on a schedule.
The fix is not to avoid tool-using tasks. They are the whole point. The fix is to put a layer between ChatGPT and your data that enforces your rules on every run.
How to Run Scheduled Tasks Safely With PortEden
PortEden is a data firewall that sits between ChatGPT and your accounts. Instead of connecting Gmail or Drive to ChatGPT directly, you connect them to PortEden, then add PortEden to ChatGPT as a connected app (a custom MCP connector). Every request a scheduled task makes then passes through PortEden's rules first, so the task only ever sees data your policy has approved, on every run.
Because PortEden runs a hosted MCP server, adding it uses the same connector flow ChatGPT already supports. There is nothing to install or deploy.
The five-minute setup
- In PortEden, connect the underlying account you want a task to use (Gmail, Outlook, Google Calendar, Drive, and so on) at
my.porteden.com, and set the access profile that matches the task. - In ChatGPT, open
Settings, thenConnectors(Apps), and choose Add connector. - Paste the unified PortEden endpoint,
https://mcp.porteden.com/mcp. This is the recommended option: it automatically exposes every tool your token allows through a single connector, so one entry covers email, calendar, drive, and the rest. ChatGPT detects the MCP endpoint and opens a PortEden sign-in window. Sign in once; no token pasting required. - Create your scheduled task as normal. When it runs and needs that app, the call routes through PortEden automatically.
The unified endpoint is the simplest path, and it matters more for scheduled tasks than for one-off chats. A morning brief that spans email and calendar needs both tools, and a single connector covers them with no extra setup. If you would rather expose only specific capabilities, each one also has its own URL you can add as a separate connector:
| Connector | Endpoint URL | What it covers |
|---|---|---|
| All tools (recommended) | https://mcp.porteden.com/mcp | Every tool your token allows, in one connector |
| https://mcp.porteden.com/email | Gmail and Microsoft 365 email | |
| Calendar | https://mcp.porteden.com/calendar | Google and Outlook calendars |
| Drive | https://mcp.porteden.com/drive | Google Drive files and folders |
| Google Docs | https://mcp.porteden.com/google-docs | Search, read, and edit Docs |
| Google Sheets | https://mcp.porteden.com/google-sheets | Read, write, and append to Sheets |
| Tasks | https://mcp.porteden.com/tasks | Monday, Linear, Asana, Jira, Notion |
The full walkthrough with screenshots lives in the Connect PortEden to ChatGPT guide.
What you can control
Once PortEden is in front of ChatGPT, you decide exactly what each scheduled task can see and do, from your PortEden dashboard:
| Control | What it does for a scheduled task |
|---|---|
| Redaction | Strips names, emails, phone numbers, and 50+ other identifier types out of responses before they reach ChatGPT, on every run. |
| Read-only by default | Grant read, draft, send, or delete as separate permissions. A read-only morning brief simply cannot send or delete. |
| Time windows | Limit a task to recent items, for example the last 24 hours or 7 days, instead of your entire history on each run. |
| Contact and label rules | Block specific people, domains, or labels (legal, HR, clients under NDA) so a task never includes them. |
| Confirm before write | Require an explicit preview before any send or delete, so an unattended run cannot act on a misread instruction. |
| Audit trail | Every tool call from every run is logged with the requested action, the decision, and the response shape. Exportable to your SIEM. |
To be precise about that last row: PortEden records the tool call the task makes through the firewall, the access-rule decision, and the redacted response returned, including anything blocked. It does not see the task prompt you wrote in ChatGPT or the summary ChatGPT sends back to you. You can read more on the audit trail and data redaction pages.
Worked example: a secure morning brief
Take the most popular task and lock it down. The prompt: "Every weekday at 7:30am, summarize my unread email from the last 24 hours and list today's meetings."
Without a firewall, this task reaches your full mailbox and full calendar every weekday morning, and you have no per-run record of what it read.
With PortEden, you add one connector, the unified https://mcp.porteden.com/mcp endpoint, which exposes both email and calendar, and set the rules once: read-only, a 24-hour time window on email, PII redaction on, and a block on your legal and HR labels. Now each 7:30 run sees only the last day of non-sensitive mail and today's events, returns a clean redacted brief, and lands as a logged entry you can review any time. Same useful summary, a fraction of the exposure.
Tips for Tool-Using Scheduled Tasks
Whether or not you put a firewall in front of them, these habits keep scheduled tasks useful and low-risk:
- Start read-only. Let a new task read and report for a week before you ever grant it the power to send or delete.
- Bound the prompt and the access. "Last 24 hours" in the prompt and a matching time window in your access rules beat "my inbox" with full history.
- Keep confirmation on for write actions. An unattended task should never send or delete silently. Require a preview.
- Connect the minimum. Add only the app connectors a task actually needs. A meeting-prep task does not need Drive.
- Keep a record. Make sure you can answer what a task read on any given run. A per-request audit trail turns that from a guess into a query.
- Review the Scheduled page. Pause or delete tasks you no longer use. An idle task with live access is unnecessary exposure.
ChatGPT Scheduled Tasks FAQ
Can ChatGPT scheduled tasks use tools and connected apps?
Yes. As of the June 2026 update, a scheduled task can use the same tools an interactive chat can, with a few exceptions. That includes web search and your connected apps (OpenAI's Apps, formerly called Connectors). A monitoring task can search the web and check connected apps for changes on a timer, then notify you only when there is something worth reporting. Voice, file uploads, and Custom GPTs are not available inside scheduled tasks.
Which ChatGPT plans support scheduled tasks?
OpenAI is rolling the refreshed scheduled tasks experience out to Go, Plus, Pro, Business, and Enterprise users on web and mobile (Web, iOS, Android, and macOS, with Windows on the way). Using custom app connectors built on MCP, including PortEden, requires a paid plan. Confirm the current availability for your plan with OpenAI, since these details change often.
How many scheduled tasks can I have at once?
ChatGPT caps the number of active scheduled tasks at once. Coverage of the feature has commonly put the limit at up to 10 active tasks, and the cap can vary by plan and change over time. When you hit the ceiling, pause or delete an existing task to add a new one. Check the Scheduled page in ChatGPT for your current limit.
Do ChatGPT scheduled tasks run when I am offline?
Yes. A scheduled task runs on OpenAI's side at the time you set, whether or not you have ChatGPT open, and it notifies you by push notification and email when it finishes. That is exactly why it matters to govern what a task can reach: it executes unattended, with no one watching the screen, so a tool-using task touches your connected apps on its own schedule. Routing those apps through PortEden keeps every unattended run inside your access rules.
Can a scheduled task send email or take actions on its own?
If the connected app has write access, then yes, a scheduled task can act without you watching, because it runs unattended on a timer. That is exactly why a security layer matters more for scheduled tasks than for interactive chats. With PortEden in front, you can keep a task read-only, or require a confirmation step before any send or delete, so an automated run cannot act on its own.
Is a scheduled task that reads my inbox a privacy risk?
It can be. A built-in connected app grounds on your full account access, so a daily 'summarize my inbox' task reaches your entire mailbox every single run, automatically, with no human reviewing what it touched. Routing the app through PortEden limits each run to a time window, redacts personal data, blocks sensitive labels or contacts, and logs every request so you can answer what the task read on any given day.
How do I add PortEden to ChatGPT so my scheduled tasks use it?
In ChatGPT, open Settings, then Connectors (Apps), and add a connector using the unified PortEden MCP URL, https://mcp.porteden.com/mcp. This one endpoint exposes every tool your token allows, so you do not have to wire up email, calendar, and drive separately. Sign in to PortEden when ChatGPT opens the consent window, then build your scheduled task as usual and it will route through PortEden whenever it uses a tool. See the PortEden ChatGPT setup guide for the full walkthrough.
The Bottom Line
The June 2026 refresh makes ChatGPT scheduled tasks genuinely agentic. They run on a timer, search the web, and reach into your connected apps, which is exactly what makes a morning brief or an inbox digest worth setting up. It is also what moves them from a personal convenience to something that touches real data with no one watching.
For reminders and learning prompts, run them freely. For any task that reads your inbox, calendar, drive, or trackers on a schedule, add PortEden as the connected app. The task keeps working every morning, ChatGPT sees only what your rules allow, sensitive details are redacted before they leave your perimeter, and every run is logged. You set the rules once, and the automation runs on data you can safely let it see.
Scheduled and automated. Your data, governed.