Field notes
ChatGPT dots field notes
Posts from people using dots, newest first. The words and credit stay with the author; we add a one-line takeaway and flag anything OpenAI hasn't confirmed.
-
719,687-byte files succeeded without prompting, while 720,000-byte files triggered approval.
Takeaway: With the GitHub plugin set to Allow all actions, writes of files around 720 KB and larger still asked for approval; smaller writes went through. Tests ran from the dot, Codex workers and scheduled tasks. The reporter doesn't know if that's an intended safeguard. One user's tests, not confirmed by OpenAI.
See also: Scheduled tasks · Connect apps
-
The same continuing permission is accepted inconsistently, blocking previously authorized read-only browser navigation and checks.
Takeaway: A dot with standing permission for recurring read-only website checks was sometimes refused for missing authorization, allowed on a retry, then refused again. The permission was never revoked. One user's report, not confirmed by OpenAI.
See also: Scheduled tasks · Custom rules
-
The requested operation remains blocked on approval, but no actionable approval prompt or link surfaced on mobile.
Takeaway: A task their dot created on a connected Mac didn't appear in the iOS task list, and approving in the main conversation didn't unblock it. They couldn't open or approve the task from the phone. One user's report, not confirmed by OpenAI.
See also: Local computer access · Custom rules
-
Explicit approvals aren’t carrying through reliably, and I’d like to understand the correct setup.
Takeaway: A user's later approval was treated as untrusted by a delegated task that began read-only; posting the approval inside that task got it through. OpenAI's docs say a new task has its own conversation and doesn't see everything you told your dot. One user's report, not confirmed by OpenAI.
See also: Custom rules · Scheduled tasks
-
The dot keeps running, but I cannot type anything or send further instructions to it on either device while the message is showing.
Takeaway: After an Auto-review precaution pause, the message blocked typing on desktop and mobile while the dot kept working. On desktop, Resume your dot failed with Something went wrong. Restarting the dot brought it back for this user. One user's report, not confirmed by OpenAI; it matches an earlier GitHub report of the same pause problem.
See also: Scheduled tasks · Custom rules
-
I was actively making calls with Dots, but I ended up hitting the limit.
Takeaway: A Pro user saw a preview-limit message after about four hours of calls, and access came back the next day. Another user in the thread reported a daily stop too. OpenAI hasn't documented a dots limit yet.
See also: Pricing & usage · Remind me later
-
the safety system said the Dot was paused, the human operator was effectively blocked from the Dot, but the Dot continued executing autonomous work.
Takeaway: An open bug report: after an Auto-review precaution, Resume and Delete failed and the dot kept working on its task. One user's report, not confirmed by OpenAI. OpenAI's docs say Pause stops the main task only, so check Activity and Scheduled too.
See also: Scheduled tasks · Custom rules
-
WeChat had been installed on the cloud computer, but it was no longer present when I returned.
Takeaway: A user found an installed app gone from their dot's cloud computer, and WhatsApp Web asked them to sign in again, with no reset they knew of. OpenAI's docs say the cloud computer can keep its state, and a site session lasts until you sign out or the site expires it. One user's report, not confirmed by OpenAI.
See also: Local computer access · Connect apps
-
i got to try openai dots a week ahead of the launch as part of their early access community … it's closer to Muse than Grokbot. one bot holding all your context, not multi-agent coordination … the interface feels like iMessage: chat or call your dots. SMS only for now, RCS and iMessage coming soon … now you invite your dot into Slack or Teams and it's already there
Takeaway: A pre-launch user's read: dots are one agent that holds your context, not a team of bots, and meeting you inside Slack or Teams is the real change. The note on SMS now and RCS and iMessage later is this user's report; OpenAI's docs describe texting only as a limited US beta.
See also: Dots vs Grok Bot · Dots vs Meta Muse · Route updates by channel
-
虚拟机的配置比较一般:Debian Linux,10G 内存,32G 硬盘 … 内置了大量高端生产力软件 … 例如blender,FreeCAD等 … 你可以手工操作虚拟机,也可以让dots连接本地电脑
Translation: The VM specs are modest: Debian Linux, 10 GB of memory, a 32 GB disk. It comes with a lot of serious productivity software preinstalled, such as Blender and FreeCAD. You can operate the VM by hand, or let your dot connect to your own computer.
Takeaway: A user's look inside the dot's cloud computer. The specs and preinstalled software are this user's report; OpenAI hasn't published them. OpenAI does document that you can take over the cloud computer yourself and optionally connect one personal computer.
See also: Local computer access · Fix reported bugs for review
What are field notes?
Posts from people using ChatGPT dots, quoted and credited, each with a one-line takeaway. They are user reports, so they stay here and never feed our reference guides until OpenAI documents the same thing.
Field notes: FAQ
Can I send you a field note?
Yes. Send the post link through the Submit page. We quote it with credit and add a takeaway.
Why aren't field notes on your guides?
Our guides only state what OpenAI has documented. User reports can be wrong or out of date, so they stay here, labeled as reports.