ChatGPT integration
Resolved 💬 43 comments Opened Aug 11, 2025 by aehlke Closed Jul 9, 2026
💡 Likely answer: A maintainer (etraut-openai, contributor)
responded on this thread — see the highlighted reply below.
I'd love to be able to move a codex session over to ChatGPT to brainstorm and use web search features and so on in a nice UI when I'm just trying to research a task and experiment with ideas. And then bring the result of that (or of a chat I started in ChatGPT) into codex CLI easily so that I can then prompt it to take action based on that context.
Ideally this would be seamless with the macOS app.
43 Comments
Why do you love this discussion-less thread so much? You keep shutting down discussion to link your favorite thread. I guess it's easier than actually working on the solution.
@ChazWinterYT, we generally use upvotes to inform prioritization on the Codex team. If there are a dozen open issues that all speak to roughly the same feature, it dilutes the signal. This thread already has many upvotes, so if you'd like to discuss this feature, please comment in this thread rather than creating a new one.
Note that now my biggest use case for this would be to seamlessly bring the 5.2 Pro model in for helping with a Codex session, or for brainstorming with Pro first and then bringing that to Codex (so, either direction)
Bringing chats and projects from ChatGPT to Codex and vice-versa is essential. Right now the only way to do that is manually transcribing everything to bring it back and forth as context. Not ideal (and very time consuming) but it works.
I thought that once I am logged-in, Codex (VSCode ext.) would have access to my past ChatGPT chats and projects, but it doesn't. I have a full project where I am designing something new technically and ChatGPT offers me the kind of brainstorming (voice included) that Codex can't, that's why I use both and constantly switch back and forth but the UX would be much better if they are able to work together.
That's why both are needed to work more closely together IMHO.
Yes. I agree, i also brainstorm in ChatGPT but have no way of having that context in codex seamlessly. Did you manually copy and paste between them to include the context in codex?
Just merge the memory/context of all of your products already. As users, we are here for the extended, long-form intelligence and memory that ChatGPT gives us. The fact that Codex, Images, ChatGPT and Pulse aren't just modalities of a common memory is a recipe for falling behind. Asking ChatGPT or a Pulse suggestion to go off and create a prototype is just natural. You shouldn't need upvotes to put this as number 1 on the feature sheet.
If this gets implemented, it would be nice if it didn't automatically happen, but was something I could opt into on a session-by-session basis.
+1
Interesting proposal! — I’d love this integration too!!
This would be especially helpful when I’m away from my dev setup (e.g., on the go, or even while traveling) — I often want to research and brainstorm in ChatGPT’s UI first(nice UI/UX, web search, etc.) , then later hand off the resulting context to Codex to execute when I’m back at a terminal/IDE.
I tried a workable stopgap with what _we have today._
The ChatGPT → Codex direction is partially doable in a semi-automated way:
conversations.json(easy with a small script),(cont...)
(cont.)
The main friction is that the export is a “full history ZIP”, so there’s still manual work, and the ZIP can get pretty large — not great for day-to-day use...
Ideally, ChatGPT would support exporting ***only the currently open conversation* (e.g., an “Export this conversation” button) — ideally as Markdown plus an assets folder (images/attachments) in a single (small) ZIP.
That would be extremely useful on its own (archiving, sharing, offline use, importing into other knowledge bases), and it would also provide a clean, reliable ChatGPT → Codex handoff without scraping the DOM.
Ideally, there would also be an even more primitive capability: an API that lets users fetch their own conversations directly, which would enable broader reuse and automation.
Even better, if Codex could accept a chat URL (or share token) and load the conversation directly — no JSON parsing — that would get very close to a seamless back-and-forth workflow.
To fully automate the ChatGPT → Codex handoff end-to-end, I also considered parsing the ChatGPT web UI (rendered HTML), but since the content is loaded dynamically and the DOM structure can change, it felt brittle and I couldn’t get it working reliably...
Either way, collaboration with the ChatGPT team will be crucial.
Hey, I need this feature badly. Is there any update on this?
I'd love having something like this in place asap.
Yesterday I had the idea to enrich the business context for a repo with all the recordings we had available. They all were mp4s without transcripts so I checked in ChatGPT how to use whisper.cpp so I could get them, summarize them, and add the curated result to my agent's docs.
I though, this is gonna be super quick, I'll let ChatGPT give me the steps and I will adapt it quickly. Long story-short, after explaining the problem and giving enough details in the conversation, I realized I needed quicker interaction to iterate with the script I generated and make it available for other teams.
Exactly at that moment, I would've loved having a button like "Continue this work in Codex" and boom.
This would save me taking my planning in ChatGPT over to Claude or Antigravity.
Big +1. I'm constantly doing this. There should be a very easy way of saying to Codex:
Currently I'm creating a shared link for the ChatGPT conversation thread, which I then give to Codex. But I think that's really not ideal, because then Codex attempts to
curlthe chat thread, which probably is not in an ideal format for it to consume. There really should be an easy way for Codex to consume ChatGPT conversations, with the ability to prioritize certain messages (so it doesn't fill context)Yes!! I have so many projects where I just export the entire conversation in Markdown to the repo as CONVERSATION.md and tell Codex (in AGENTS) to use that as the basis for what the project is. This would make everything SO much simpler!
How do you export a chatGPT conversation to markdown?
@3f6a Go to the ChatGPT website, under settings go to data controls, export all data, and eventually you'll get an email with a link to download all your data in a zip (link expires 24 hours after it's sent though), and if you open
chat.htmlyou can copy an entire conversation. It gives the raw unformatted responses unlike the regular website which formats the model's Markdown, so you get the formatting too. The page does have ALL your conversations, though, so you have to specifically copy just the one you want, and also you have to redownload all your data whenever you want to update it for CodexWould be nice to integrate Deep Research queries from Codex CLI to ChatGPT
@jonahdk Thanks. Though I think we can agree this is not ideal. A nicer integration between chatGPT and Codex would be a very nice addition to the ecosystem.
Why not just use the Codex app?
no pro model
Not available on Linux
Does the Codex app have some magic connection to ChatGPT for sharing context? If there is, I haven't been able to find it yet.
No, the point is that Codex app offers a chat experience similar to ChatGPT app, so you don't need to use ChatGPT in the first place. Not having the pro model is a valid point. Maybe that's something the OpenAI team can consider.
Related to agent auto-accept workflows: if anyone here is running extended agent sessions and getting tired of manually clicking Accept/Run/Continue, I built an extension that automates this at the OS level:
Antigravity Autopilot — OS Level (GitHub)
Uses Windows UI Automation (not CDP/DevTools) to auto-click agent approval buttons. Works across VS Code, Cursor, Windsurf, and Antigravity. Fully configurable patterns let you control which buttons get clicked.
Windows-only currently, macOS support planned.
Fair enough– although for me it's not that the experience isn't similar (it is) but that I use ChatGPT on mobile, and often prefer it for fuzzier conversations that may end up driving code changes but may not (and aren't about a specific code repo, notably). For the case where some part of the conversation naturally leads into code work, it would be great if Codex could pick up the thread that was started in ChatGPT.
One issue is that Codex app doesn't render latex, whereas chatgpt app does. I'm often working on math heavy project, where for example I first prototype some algorithm or simulation pipeline with chatgpt (this usually involves discussing various equations back and forth), and then want Codex to implement that final idea.
If Codex was able to discuss rendering LaTex correctly then yes, maybe I could just use Codex.
@cadam11
+1
That’s exactly right.
I also tend to use ChatGPT on mobile for everyday questions or tool-related issues. From there, ideas naturally expand, and I often end up thinking, “this could be turned into an app” or “I could build a tool for this.”
Another big point is that ChatGPT lets you branch conversations or go back to much earlier points and continue from there. Codex App mainly supports fixing the most recent message, so this kind of usage is harder.
When brainstorming, I don’t want to narrow things down too quickly. I want to explore multiple directions and let ideas spread. For that, being able to branch is really important (If you keep going in a single thread without branching, the long context can get “polluted.” Different topics start to mix together, and the quality of responses can drop as a result.).
Also, being able to easily drop in images, text, and other information during a conversation makes ChatGPT much better suited for brainstorming.
This kind of integration is important. And it would also be great if the Codex app could support conversation branching like ChatGPT.
Let’s give the issue a 👍thumbs up and upvote it so OpenAI can prioritize it.
Reading this thread made me realise how important it is to be able to have one central source of truth for AI chat. That way you can jump between different providers (e.g. OAI, Anthropic) and clients (e.g. Codex, OpenCode).
So I started building OpenChat (https://github.com/P0u4a/openchat). I'm planning for it to become a full on protocol, but for now I've solved my own problem by having a chrome extension that automatically saves my conversations locally, and an MCP server that can read those conversations and expose them via resources. So you can reference them in codex or any other client that supports MCP.
Keen for feedback.
up for this, I think should have integration between
chatgpt.comandcodexsoon cc @etraut-openaithis will be like
teleportfrom other featureit's likely by design. Codex (And other providers) threads get very large. Sometimes more than 10mb. I have about 500 chats just from the last 2 weeks. I have had to build around this issue with KaizenAI since keeping them in memory for fast loading just gets out of hand quickly.
I do agree though there could be some kind of last 20-50 threads transfered to a cloud provider and then available across systems.
Ideally the user gets to choose their storage backend. Kind of like Obsidian notes, I can use GitHub or OneDrive to sync. And storage is cheap anyway so storing all threads should be fine.
Adding a concrete use case here instead of opening another duplicate.
Uploaded Codex CLI feedback thread ID:
019d9a46-fe44-7491-91dc-7690b2298431What I want is not just generic integration, but an intentional handoff flow in both directions:
I am not asking for raw transcript mirroring. An explicit handoff capsule would be a much better design:
Related public discussion:
https://community.openai.com/t/feature-request-native-chatgpt-codex-thread-bridge/1379162
i can't beleive there is no movement on this.... hope it is soon. copy / paste is brain-dead.
Still no progress???
they announced that the codex app is going away and will be consolidated into chatgpt. so wait for that to see.
yeah instead of integrating chatgpt into codex, codex should be merged into chatgpt. this will increase the codex usage more and agentic AI will be on default for all chatgpt app users.
Hopefully when they merge the two apps, they will have figured out how to merge the history, something they've not been able to do for 9 months according to this thread.
Adding a related reference, not an official OpenAI solution and not a proposed Codex PR: a friend pointed me to this independent MCP bridge that explores a similar workflow: https://github.com/velhud/PatchBay
It seems to cover part of this workflow: ChatGPT can coordinate local Codex CLI workers and receive reports/diffs/results back in the chat.
like I said - it happened today
Solved in today's update now that they've replaced ChatGPT with Codex:
<img width="234" height="79" alt="Image" src="https://github.com/user-attachments/assets/214115cf-54ea-475e-b9d6-489af11fccec" />
And so the monkey's paw finger curls... ChatGPT Work quota is shared with Codex quota 😔
<img width="907" height="560" alt="Image" src="https://github.com/user-attachments/assets/0846e8c2-3eb0-4e1d-8589-1a8694f0227c" />
@ignatremizov Chats are not though. nothing changed
Ohh I see can you tell something about a recent task where you faces this problem??