Contents
After a day with Codex: does this sound familiar?
Company research, a proposal deck, meeting notes, a report to your manager. OpenAI Codex did it all. Here's what still happens by evening.
Research → deck → notes → report, done. Tomorrow you type it all again.
84% left. Smarter models drain it faster, so you hesitate before asking.
You open the task list to check. Pending decisions are buried in the thread.
Codex finishes one task. Deciding what to ask next, and checking whether it's done, are still in your hands.
What you really wanted after Codex finished
Same Codex. Same ChatGPT plan. This is how the end of the day should have looked.
Notes, report, client-note updates. Tasks are already running as you walk out.
Deck structure, report detail. It learns from the first run and repeats it.
"You did this three times last week. Run it every time?" Found in your Codex history.
"OK to propose at this price?" One question comes back. Answer it from your phone.
This starts the day after you connect your ChatGPT account to SIMY.
Where OpenAI Codex stops
Codex is great at finishing the one task you asked for. Before and after that, nobody picks anything up.
- AskYou type it
- Codex runsIn a sandbox
- DoneThe task ends here
- Find the next job
- Ask again
- Check progress
- Decisions get buried
* Flow shown for illustration only
Schedules and AGENTS.md stop at the same spot. You can repeat on a timer and write down your approach. But finding work that came out of a meeting is still you, every time, and so is checking whether it's done.
SIMY connects it: the one running Codex, on your own ChatGPT account
SIMY fills the four blanks. It decides what, when and how far to ask Codex, and only the decision box comes back to you.
- Ask
- Run
- Done
- Find the next jobFrom meetings and chat
- Ask CodexYour way, every time
- Confirm it's doneNo checking in
- DecideReply from your phone
* Flow shown for illustration only
- Picks up from meetings and chat
- Does it your way
- Returns only decisions
SIMY doesn't replace Codex. It runs Codex on your own ChatGPT account. It finds the next job in Slack, Teams and meetings, learns the checks and priorities you repeat when you do good work, and has Codex follow the same approach.
- Sign up for SIMY and connect your ChatGPT accountAny plan with Codex, Plus or higher. Codex usage comes out of your own plan.
- Install the SIMY desktop app (or CLI)It reads Codex chat history on your computer (90 days by default) and turns repeat work into "Suggestions from SIMY." Chat text is not stored on our servers. Download here
- Return only decisions with the iOS appTurn a suggestion into a WorkFlow and SIMY asks Codex from then on. Pending decisions reach your phone.
* SIMY screen shown for illustration only
What you need: a ChatGPT account with Codex (Plus or higher). No new Codex settings or API keys.
One deal, side by side: the task Codex finished and the day SIMY lines up
A sample day when a sales rep asked the Codex desktop app for a "proposal deck for Client A." Codex on the left, SIMY on the right.
- Asked for
- Research Client A's business and recent news, and build a deck in Slides with our case studies.
- Delivered
- A 12-slide deck: company overview, problem hypotheses, impact, three pricing options, next steps.
- Used
- Research in Chrome, the case-study folder in Drive, the Slides plugin.
- Still left
- Ask the same for Client B tomorrow and Client C next week. Ask for post-meeting notes and reports yourself.
SIMY picked up tomorrow's meeting from your calendar and had Codex build it in Client A's structure, with case studies swapped for Client B's industry.
Ten minutes after the Client A meeting, notes at your usual level of detail. The report waits as a Slack draft and posts after you approve.
Stage: piloting in three departments; concern: integration with their CRM. Decision this fiscal year; IT joins next time.
Which of the three pricing options leads for Client B. Only what needs your call comes back to your phone.
Codex builds, SIMY asks and checks. The deal itself is still where you shine.
By role: work to give Codex, and where SIMY picks it up
OpenAI Codex can do work that involves no code. Here's what SIMY starts from, and what it asks Codex to do.
Sales Chrome, Slides, Gmail drafts
- CRM is double entry→Meeting notes become deal records
- Fixing the prompt every time→Follow-up drafts per contact
- Pre-meeting research by hand→Talking points and deck before you meet
Leaders and managers Slack, Sheets, Notion
- Asking "where are we on that?"→Only pending decisions come up
- Collecting weekly reports→Weekly report built from decisions
- Copying numbers by hand→Sheets updated, with changes explained
Operations and planning Drive, Sheets, Calendar
- Requests scroll past in chat→Requests picked up and drafted
- Rebuilding recurring decks→Last format, fresh data
- Back-and-forth scheduling→Time options and invite drafted
Engineers and dev teams CLI, VS Code, with Claude Code
- Re-asking for fixes agreed in meetings→Decisions become Codex tasks
- Getting asked for status→PR and review status, as decisions only
- Running out of Codex usage→Continue with Claude Code too
Get started: after your first Codex task, 3 steps
- Ask the Codex desktop app for one jobLog in with your ChatGPT account and ask in plain language: company research, a proposal deck, meeting notes. Download here
- Connect your ChatGPT account to SIMYYour Codex settings stay as they are. See how it connects above.
- Next day, check "Suggestions from SIMY"Your repeat work arrives as candidates. Make them WorkFlows and return only decisions. That's it.
How sign-up works: choose a plan → verify your email → pay by card. There is no free plan. See pricing. Get the desktop app from the download page.
Already using Codex? Just do step 2. Connect and install the SIMY desktop app, and suggestions start from your past chat history.
OpenAI Codex in depth: setup, pricing, limits, comparison and security
Open these when you want to check something about Codex itself. None of it affects how SIMY connects.
What is OpenAI Codex: how it differs from ChatGPT and the old CodexAn agent included in ChatGPT plans
What Codex is
OpenAI Codex is an AI agent from OpenAI. ChatGPT answers when you ask; Codex does the work when you ask. It reads files, runs commands, operates a browser and produces deliverables, on your computer or in a cloud workspace. It started as a software development tool, but in 2026 the desktop app and plugins extended it to work with no code, like decks and email drafts.
How it differs from ChatGPT
ChatGPT gives you an answer in the conversation. Codex gives you the finished thing. Ask "research Client A and build a proposal deck," and ChatGPT returns an outline in text, while Codex creates the Slides file and stops there. Both use the same ChatGPT account and share usage.
Don't mix up the names
- The original Codex (2021): a code-completion language model used in early GitHub Copilot. It was retired in 2023 and is a different product from today's OpenAI Codex.
- Model names like GPT-5-Codex: these are models you can pick inside Codex. Codex the product and GPT-5-Codex the model sit at different levels.
How it differs from ChatGPT Work
The ChatGPT desktop app has three modes: Chat, Work and Codex. Work is an agent that finishes general deliverables like research and documents; Codex is stronger with files and commands. They share the same usage. If you don't write code and can't decide, use Codex when you need file access or Chrome control, and Work when you want deliverables as an extension of a conversation.
Language support
The app interface is mainly in English, but you can give instructions and get deliverables in other languages. AGENTS.md and skills can be written in any language too.
Entry points: desktop app, CLI, VS Code, Cloud, mobileIf you don't code, start with the Windows or Mac desktop app
Scroll the table sideways →
| Entry point | Platform | Best for | Highlights |
|---|---|---|---|
| Desktop app (recommended) | Windows, Mac | Non-coders, first-time users | Log in with ChatGPT. Run several tasks side by side, with plugins and schedules |
| Codex CLI | Terminal (Mac, Linux; WSL recommended on Windows) | Engineers, automation | Install with npm. Fine control over AGENTS.md, approval modes and MCP |
| IDE extension (VS Code, Cursor, etc.) | VS Code-based editors | Staying inside the editor | Uses open files as context and edits in place. Same account as the CLI |
| Codex Cloud | Browser (chatgpt.com) | Delegating without tying up your computer | Runs in a cloud sandbox. Connects to GitHub repos and opens PRs |
| Mobile | ChatGPT iOS and Android apps | Checking and asking on the go | Check Cloud tasks and start new ones |
How to choose: for decks, email and research, the desktop app is all you need. The CLI and IDE extension are entry points for people who work with code. Every entry point draws on the same ChatGPT account's usage.
Setup: login, AGENTS.md, approval modesThe desktop app starts in two moves: download and log in
Start with the desktop app (for non-coders)
- Download and install. Get the Windows or Mac version from OpenAI's Codex page.
- Log in with your ChatGPT account. On Plus or higher, no API key is needed. A browser opens; allow access and you're done.
- Pick a working folder. Choose the folder Codex may read and write. A Drive folder or one with your materials is enough.
- Choose an approval mode. Start with "ask before changes," and widen to "auto" once you're comfortable.
- Ask for one job in plain language. "Use the case studies in this folder to build a proposal deck for Client A" — ask the way you'd ask a colleague.
Start with the CLI (for engineers)
With Node.js installed, run npm install -g @openai/codex in a terminal, then type codex and log in. Native Windows support is improving, but WSL is more stable.
AGENTS.md: where you write down how to work
Put a text file called AGENTS.md in your working folder and Codex reads its rules (naming, structure, don'ts) on every task. You can also write business formats, like "proposal decks are 12 slides and end with next steps." With SIMY, it learns these formats from your chats, so you can start without writing one.
Asking well on your first task
- Say where the materials are. Naming the place to read, like "use the Case Studies folder in Drive," cuts searching time and usage.
- Show one example of the result. Attach a previous deck or a sample email and the structure and tone will match.
- Decide where it should stop. Write "don't send" or "draft only," and you'll get fewer approval prompts and more peace of mind.
- Get results as files. Ask for Slides or Docs files instead of long chat replies, so you can share them as is.
Approval modes and the sandbox
Codex can switch between a mode that asks before changing files or acting outside, an "auto" mode that proceeds inside the working folder, and "full access." In every mode, work happens in a sandbox that limits access outside the chosen folder and to the network.
What it can do: skills, MCP, Chrome control, plugins, schedulesGmail drafts, Drive, Sheets, Slides, Notion, Calendar, Slack
Business apps via plugins (as of October 2026)
- Gmail: reads mail and saves replies or new emails as drafts. You press send.
- Google Drive, Sheets, Slides: reads files in folders, updates spreadsheets and builds slides.
- Notion: reads and writes pages and databases. Good for storing notes and client records.
- Google Calendar: reads your schedule, suggests times and prepares invites.
- Slack: reads channels and drafts posts.
- Chrome control: opens Chrome to research sites, fill in forms and read screens. Also useful for checking business systems that need a login.
Skills, MCP and subagents
Skills are reusable sets of steps and judgment criteria. Turn "how to write the weekly report" or "our notes format" into skills and you skip the explaining. MCP (Model Context Protocol) is a common standard for connecting outside tools and data to Codex; services without a plugin (Figma, internal databases) are added here. Long tasks can be split across subagents to run in parallel.
Schedules, plan mode and review
Schedules repeat a task on a timer, like "every Friday, cross-check the week's email, Slack and sales records and draft the weekly report for my manager in Slack." Plan mode shows only the steps and scope before running so you can check them. Code review has Codex check GitHub PRs, for dev teams.
Non-coding examples
- Company research: read the prospect's site and news in Chrome and summarize key points and angles in Docs.
- Proposal decks: 12 slides from case studies and the price list in Drive, structured like your last deck.
- Notes and reports: turn a transcript into notes with decisions, owners and deadlines, plus a summary for your manager as a Slack draft.
- Scheduling: pick three open slots from Calendar and save the invite email as a Gmail draft.
- Spreadsheet updates: read the progress sheet, apply changes agreed in meetings, and list the differences.
Where SIMY fits: schedules start from the clock, and you still write the first instruction. SIMY starts from meetings and chat. Because Codex plugins can already make Gmail drafts and Slides, SIMY can focus on deciding when, what and how far.
Pricing and usage limits: Plus and up, the usage meter, resetsSmarter models use more. Check the official page for exact limits
Scroll the table sideways →
| ChatGPT plan | Codex | Usage (rough guide) | With SIMY |
|---|---|---|---|
| Free, Go | Not available | — | Not supported |
| Plus ($20/month, excl. tax, as of Oct 2026) | Available | Enough for one person to run a few tasks a day | Connect and use |
| Pro | Available | Larger than Plus. For long-running tasks | Connect and use |
| Business, Enterprise, Edu | Available | Admins set usage; extra seats can be purchased | Connect and use |
How usage limits work
- A 5-hour window and a weekly window: both reset counting from your first use. The app's usage meter shows what's left and when the next reset is.
- Usage depends on model and task: smarter models, long chats and big files use more. The standard model is often enough for everyday documents; switching up only for hard research makes your allowance last.
- When you run out: wait for the reset, upgrade to Pro, switch to pay-as-you-go with an API key, or add another agent like Claude Code.
- Where to check usage: the usage meter in the desktop app and CLI, and per-plan usage in ChatGPT settings.
Making your allowance last
Put the materials' location and the finished format in one instruction, move long chats to a new thread, and split research from creation. Those three visibly cut usage for the same work. Think of the model as the painter's skill and the reasoning level as the time spent on the painting; when usage is tight, lower the reasoning level. When SIMY asks Codex, it passes along the format it learned, so there's less redo usage.
Plans and prices are from OpenAI's official information checked on October 1, 2026. Exact limits change, so they're not listed here. Prices vary by region; check the latest on the ChatGPT pricing page.
With SIMY: Codex usage comes out of your own ChatGPT plan. SIMY adds no markup on Codex usage. Pricing for SIMY itself is on the pricing page.
Comparing and pairing with Claude Code, Cursor and GitHub CopilotNot better or worse; different plans and different stopping points
Scroll the table sideways →
| Compare | OpenAI Codex | Claude Code | Cursor | GitHub Copilot | SIMY |
|---|---|---|---|---|---|
| Made by | OpenAI | Anthropic | Anysphere | GitHub (Microsoft) | AwakApp |
| Runs on | ChatGPT Plus or higher | Claude Pro or higher | Paid Cursor plan | Paid Copilot plan | Your ChatGPT plan + SIMY |
| Main entry points | Desktop app, CLI, IDE, Cloud | CLI, IDE, desktop (Cowork) | Editor | Editor, GitHub | Web, desktop app, iOS |
| Non-coding work | Decks, email, research via plugins | Documents and research via Cowork | Editor-focused | Dev-focused | Any work born in meetings and chat |
| Starts from | You ask / a timer | You ask / a timer | You ask | You ask | Picked up from meetings and chat |
| What you get back | Deliverables | Deliverables | Code | Code | Draft deliverables and one decision |
Why use both
Codex and Claude Code run on different plans. More people now learn both so work doesn't stop when one model has issues or the allowance runs out. With the two CLIs, you can alternate in the same folder.
Where SIMY stands: SIMY can run work with both Codex and Claude Code. In the cloud it runs Codex on your ChatGPT account, and on your computer the SIMY desktop app can use Claude Code too. Either way, only decisions come back to you. For choosing within ChatGPT itself, see the ChatGPT guide.
Security and cautions: training opt-out, sandbox, approvalsThree things to check before using it for work
- Training opt-out: in ChatGPT settings → Data controls, turn off "Improve the model for everyone" and your Codex chats aren't used for training either. It's on by default. On Business and Enterprise, work data isn't used for training by default under the contract.
- Sandbox: Codex works in an environment limited to the chosen folder and network scope. It doesn't touch files outside the folder or sites you haven't allowed.
- Approvals: Codex asks for approval before irreversible actions like deleting, sending or posting. Avoid auto-approve (so-called YOLO mode) outside throwaway environments.
- Chrome control: if it works while you're logged in to business systems, create a dedicated Chrome profile and use an account with minimal permissions.
- Data leaving your hands: as a rule, don't set folders with personal or confidential data as working folders. If company rules limit ChatGPT use, consider Business or higher.
- Connecting to SIMY: SIMY signs in to your account through ChatGPT's own sign-in screen, and you allow access to connect. When building suggestions from chat history, the text is not stored on our servers. Details are in Security and the Privacy Policy.
Frequently asked questions
Is OpenAI Codex free?
Codex isn't available on ChatGPT's free plan. It's included in Plus, Pro, Business, Enterprise and Edu with no extra sign-up. You can also run it pay-as-you-go with an API key (as of October 2026).
How much Codex do I get with ChatGPT Plus?
There's a 5-hour window and a weekly window, and the usage meter in the app shows what's left. Smarter models and longer tasks use more, and Pro has a larger allowance than Plus. Exact limits change, so check the official pricing page.
When does Codex usage reset?
The 5-hour window resets five hours after your first use, and the weekly window seven days after your first use. The app's usage meter shows the next reset time.
Claude Code or Codex: which should I choose?
You don't have to pick one. Codex runs on a ChatGPT plan and Claude Code on a Claude plan, so more people learn both as a backup when one has issues or runs out. SIMY can run work with both Codex and Claude Code.
What is Codex plan mode?
A mode that lays out the steps and scope first, without changing any files. You can review the plan before big work and run it once you agree. Combined with approval modes, you can set exactly how much to delegate.
Can non-engineers use Codex?
Yes. OpenAI says more than 20% of Codex users are not engineers (as of 2026). Ask the desktop app in plain language and it handles company research, proposal decks, meeting notes and Gmail drafts, with almost no code in sight.
How are Codex schedules different from SIMY?
Codex schedules start from the clock and repeat a set task. SIMY starts from meetings and chat, finds the work they create, and has Codex do it. Use Codex schedules for the 9 a.m. routine and SIMY for work that changes with every meeting.
Is it safe to let Codex control Chrome?
Codex runs in a sandbox and asks for approval before actions with outside effects, like logging in or submitting forms. For work use, keep three things in place: don't set approvals to auto, use a dedicated Chrome profile, and check your training opt-out setting.
Is what I enter in Codex used to train AI?
If you turn off "Improve the model for everyone" under Data controls in ChatGPT settings, your Codex chats aren't used for training either. On Business and Enterprise, work data isn't used for training by default under the contract.
Does SIMY charge extra for Codex usage?
SIMY connects your ChatGPT account and gets work done through the Codex App Server. Codex usage comes out of your own ChatGPT plan. For SIMY's own pricing, see the pricing page.
Can I use Codex from my phone?
From the ChatGPT iOS and Android apps, you can check Codex Cloud tasks and start new ones. With SIMY's iOS app, only the pending decisions from work Codex has done reach your phone, and you can approve them there.
Sources
OpenAI Codex features, entry points, plans, usage limits and security were checked against the official sources below (checked October 1, 2026). The scenes of work left after using Codex draw on AI how-to videos.
- OpenAI Codex (product page)
- Codex documentation (OpenAI Developers: CLI, IDE, Cloud, AGENTS.md, approval modes)
- ChatGPT pricing (plans that include Codex)
Descriptions of Claude Code, Cursor and GitHub Copilot are overviews as of October 2026 based on each company's official site. Check their pages for details.