ChatGPT Codex Automations + Gmail Plugin: How to Schedule AI Tasks That Run Without You
Codex can send Gmail summaries on a schedule with no manual input. Here's how to set up Automations and the Gmail plugin step by step.
Your Friday Report Writes Itself: Codex Automations and the Gmail Plugin
Codex Automations — accessible via sidebar > automations > new automation — can send a Gmail summary of your project folder’s activity every Friday morning without you touching a keyboard. That’s the outcome. Everything below is how you get there.
Most people who discover Codex treat it as a faster way to write code or generate a document. That’s underselling it by roughly half. The Automations feature combined with the Gmail plugin turns Codex into something closer to a scheduled agent: it wakes up on a cadence you define, reads files in a folder, synthesizes them, and emails the result to whoever needs it. No cron jobs. No API keys. No Python script running on a server somewhere.
If you’ve ever spent twenty minutes on a Friday afternoon assembling a status update that could have been assembled by a machine, this is the feature you’ve been waiting for.
What You’re Actually Building Here
Before the steps, a concrete picture of the end state.
You will have a Codex automation that runs on a schedule — daily, weekly, whatever you choose. It reads files inside a project folder you’ve designated. It synthesizes those files into a summary. Then it uses the Gmail plugin to send that summary to an email address you specify.
The demo that illustrates this most clearly: a “marketing email automation” that reads transcript files from a working folder, summarizes the video topics covered that week, and sends the result via Gmail to a specified recipient. The whole setup takes under ten minutes. The automation then runs without any further input.
There are two distinct capabilities being combined here. Automations handle scheduling and task definition. The Gmail plugin handles the email delivery leg. They’re configured separately and then composed together in a single automation prompt. Understanding that separation makes the setup much less confusing.
One thing worth flagging early: Codex runs on GPT-5.5, and the agent burns through tokens while working. A weekly summary automation on a small folder is cheap. An automation that reads hundreds of large files daily is not. You’ll want to check your usage before setting aggressive schedules — more on that in the troubleshooting section.
What You Need Before Starting
A Codex subscription with sufficient credits. Automations consume tokens every time they run. Check your current balance at File > Settings > Usage tab before you build anything scheduled. Running an automation against a large folder on high reasoning could drain credits faster than you expect.
A Gmail account you’re willing to connect. The Gmail plugin uses OAuth. You’ll authorize Codex to read and send email on behalf of that account. Use an account you actually control and monitor — not a shared team inbox unless you’ve thought through the access implications.
A project folder with files in it. Automations that summarize work need something to summarize. Create a folder on your local machine, put relevant files in it (transcripts, notes, drafts, reports — whatever represents your working output), and know its path. The demo used a folder called “Codex demo” containing two transcript text files.
Basic familiarity with Codex’s interface. If you’ve never opened Codex before, spend fifteen minutes with it first. The sidebar, the project folder selector, and the chat interface are all things you’ll navigate during this setup.
Setting Up the Gmail Plugin
Step 1: Install the Plugin
Open Codex. Toggle the sidebar open. Navigate to the plugins tab. Find Gmail in the list and click “install plugin.”
Codex will open your browser to an OAuth authorization screen. Sign in with the Gmail account you want to connect. Grant the requested permissions — Codex needs to be able to read your email and send on your behalf for the automation to work.
Return to Codex. You should see Gmail listed as connected in the plugins tab.
Now you have a Gmail-connected Codex instance. You can verify it works by opening a new chat and asking something like “what’s the latest unread email in my inbox?” If it returns real email data, the plugin is live.
Step 2: Verify the Connection Does What You Expect
Before building an automation around Gmail, test the plugin manually. Ask Codex to summarize your last five emails, or to tell you what’s urgent. The plugin can read, categorize, and draft replies — but for the automation use case, you primarily need it to send.
Not a coding agent. A product manager.
Remy doesn't type the next file. Remy runs the project — manages the agents, coordinates the layers, ships the app.
Ask it explicitly: “Draft and send a test email to [your own address] saying ‘Gmail plugin test.’” Confirm the email arrives. This rules out permission issues before you build a scheduled workflow on top of a broken connection.
Now you have a confirmed, working Gmail plugin that can both read and send.
Building the Automation
Step 3: Create a New Automation
Open the sidebar. Click “automations.” Click “new automation.”
You’ll see a prompt to either use a template or configure from scratch. The templates skew heavily toward code-based tasks. For a summary-and-email workflow, configure it yourself — it takes about two minutes and gives you more control.
Give the automation a name. Something like “Weekly Project Summary Email” is descriptive enough that you’ll know what it does when you come back to it in three weeks.
Step 4: Attach a Project Folder
In the automation configuration, attach it to the folder containing your working files. This is the same folder mechanism used elsewhere in Codex — you’re pointing the automation at a local directory so it has context to work from.
The automation will read whatever files are in that folder when it runs. If you add files to the folder between runs, the next run will include them. This is the behavior you want for a weekly summary: the folder accumulates work throughout the week, and the automation synthesizes it at the end.
Step 5: Write the Automation Prompt
This is where most of the value lives, and it’s also where most people underspecify.
A weak prompt: “Summarize my work this week.”
A better prompt: “Summarize the files inside this folder. Identify the main topics covered, any notable outputs or deliverables, and anything that appears incomplete or needs follow-up. Then use @Gmail to send this summary to [recipient email address] with the subject line ‘Weekly Project Summary — [current date].’”
The @Gmail mention is how you invoke the plugin inside an automation prompt. Codex treats it as an instruction to use the connected Gmail integration for the delivery step.
Set your reasoning level here. For a summary-and-send task, medium reasoning is sufficient and won’t burn through credits unnecessarily. Save high and extra-high for tasks that involve code generation or complex analysis.
Step 6: Set the Schedule
Choose your cadence. The demo used Sunday at 9:00 a.m. for a weekly summary. Daily options are available if your use case warrants it.
Be deliberate about timing. An automation that runs at 6:00 a.m. Monday delivers a summary before your workday starts. An automation that runs Friday at 5:00 p.m. delivers a summary when no one is likely to act on it. Match the schedule to how the output will actually be used.
Step 7: Set Permissions
Codex offers three permission modes: default, auto-review, and full access. For an automation you’re running unattended, full access is the practical choice — otherwise the automation may pause waiting for your approval on file reads or email sends, which defeats the purpose of scheduling it.
Full access means Codex can read your project folder and send email without prompting you each time. Understand what you’re granting before enabling it. For a folder containing work documents and a Gmail account used for notifications, this is a reasonable tradeoff. For a folder containing sensitive financial data or a primary business inbox, think harder about it.
Now you have a scheduled automation that will run without your involvement, read your project folder, and email a summary to your specified recipient.
Combining This With Skills
One underused pattern: pair an automation with a Codex Skill to enforce consistent output formatting.
Skills are reusable task-specific capabilities created via sidebar > plugins > create skill. The demo example was a Twitter skill enforcing a 240-character limit, plain language, no hashtags or emojis. The same logic applies to email summaries.
If you want every automated summary to follow a specific structure — executive summary first, then bullet points by project, then open items — encode that as a skill. Then reference the skill in your automation prompt. The automation will apply the formatting rules every time it runs, without you re-specifying them.
This matters more than it sounds. An automation that produces inconsistently formatted output trains recipients to ignore it. An automation that produces a clean, predictable structure every week gets read.
What Goes Wrong (and How to Fix It)
The automation runs but no email arrives. Most likely cause: the Gmail plugin lost its OAuth authorization. Tokens expire. Go back to the plugins tab, disconnect Gmail, and reconnect it. Then run the automation manually to confirm delivery before relying on the schedule again.
The automation pauses waiting for approval. You’re running on default or auto-review permissions. Switch to full access in the automation settings if you want unattended execution.
The summary is vague or misses important files. Two possible causes. First, your prompt is underspecified — add explicit instructions about what to look for and how to structure the output. Second, the folder contains too many files for the reasoning level you’ve set. Try bumping from medium to high reasoning for the automation, and monitor whether the credit cost is acceptable.
Credits are draining faster than expected. Check File > Settings > Usage tab. If an automation is running against a large folder with high reasoning enabled, the token cost per run adds up. Either reduce the folder size, lower the reasoning level, or reduce the run frequency. A weekly automation is almost always cheaper than a daily one for the same folder.
The automation creates files but doesn’t send the email. The @Gmail mention in your prompt may be malformed or missing. Codex needs the explicit plugin invocation to route the delivery step through Gmail. Rewrite the prompt to include @Gmail immediately before the send instruction.
Where to Take This Further
The summary-and-email pattern is the obvious starting point, but it’s not the ceiling.
You can chain more steps into the automation prompt. Summarize the folder, identify the three most important items, draft a short action list, and send all of it in a single email. The automation executes the whole sequence on schedule.
Seven tools to build an app. Or just Remy.
Editor, preview, AI agents, deploy — all in one tab. Nothing to install.
You can run multiple automations against the same folder with different cadences. A daily automation that flags anything marked urgent. A weekly automation that sends the full summary. A monthly automation that synthesizes the month’s work into a longer report. Each is a separate automation configuration pointing at the same folder.
The concurrent execution capability is worth understanding here. Unlike standard ChatGPT, which handles one task at a time sequentially, Codex can run multiple tasks across separate chat windows simultaneously. This means your scheduled automation can run in the background while you’re actively working in another Codex session — they don’t block each other.
If you’re thinking about how this kind of agent-plus-integration pattern scales to more complex workflows, MindStudio handles that orchestration layer: 200+ models, 1,000+ integrations, and a visual builder for chaining agents and workflows when the task outgrows what a single tool’s plugin ecosystem can support.
The local web app demo — a valuation tracker for OpenAI vs. Dropbox funding rates — points toward another direction: using Codex to build small data tools that live alongside your project folder. If you find yourself wanting a more durable version of that kind of output, tools like Remy take a different approach: you write a spec in annotated markdown, and it compiles a complete TypeScript backend, database, auth, and frontend from it — the spec is the source of truth, the code is derived output.
For teams already using Claude-based tooling, the Google Workspace CLI with Claude Code covers a parallel approach to Gmail and Drive automation that’s worth comparing against the Codex plugin model — different tradeoffs on setup complexity and control.
The comparison between Claude Code and Codex is also worth reading if you’re deciding where to invest time. They overlap more than their marketing suggests, but they diverge meaningfully on how they handle file system access and scheduling.
And if you’re thinking about keeping agents running continuously rather than on a schedule, the patterns in keeping a Claude Code agent running 24/7 apply to any local agent setup, Codex included.
The practical recommendation: start with one automation. Pick the weekly summary use case, configure it against a folder you actually use, connect Gmail, and let it run for two weeks. If the output is useful, you’ll know exactly what to build next. If it isn’t, you’ll have learned something specific about what your prompt needs to say differently — which is more valuable than any amount of theorizing about what automations could do.
