GitHub Outage Map

Community Discussion

Tips? Frustrations? Share them here. Useful comments include a description of the problem, city and postal code.

Beware of "support numbers" or "recovery" accounts that might be posted below. Make sure to report and downvote those comments. Avoid posting your personal information.

GitHub Issues Reports

Latest outage, problems and issue reports in social media:

• 
  tom cunningham (@testingham) reported 3 minutes ago

  Q: has AI accelerated aggregate discovery yet? My very general impresions, would love others' thoughts: 1. Vulnerability discovery is up a lot. Between 2X-5X increase in volume, although the average severity has fallen somewhat. 2. Algorithmic efficiency hasn't moved much. E.g. nanogpt, SAT solvers, compression efficiency, chess algorithmic ability. They haven't shown noticeable changes I believe. 3. Math is hard to judge. Erdős problems are certainly falling faster than historical rates, but it's mainly the obscure ones. We don't seem to be tightening upper and lower bounds on unknown quantities noticeably more quickly AFAICT. There are some prominent AI discoveries (e.g. unit distance) but I don't know what's the denominator, i.e. the average flow of comparably important results, my guess is it's still small. 4. Papers/code volume is up a lot. Total arXiv papers and github code is up a lot, but no good way of judging how the quality has changed.

• 
  Dom Esposito (@macmixing) reported 7 minutes ago

  Dear Open AI/Codex Team, Please fix the issue I’m seeing where gh auth status falsely reports that GitHub CLI is unauthenticated when run inside the Codex sandbox. It does have authentication through the macOS Keychain and works outside of it. This happens constantly and causes agents to incorrectly claim that authentication is broken. It’s quite annoying. Thanks. -Dom

• 
  aryan (@aryanranderiya) reported 7 minutes ago

  @code_typist @pierrecomputer does this allow stuff like actually reviewing the diff and adding comments? because GitHub is excruciatingly slow for large PRs

• 
  Alexzvn 🐱 (@alexzvnvn) reported 13 minutes ago

  @mattpocockuk @tldraw I find Github cli often being rate limited. Wonder if you've ran in same issue.

• 
  Vamz (@Vamzzz93) reported 15 minutes ago

  @andrewxroas Meta officially supports this. The ban came from users not setting up API keys and instead having OpenClaw do some browser session hacking. There are several api-key alternative login methods in GitHub a bunch of agents try and use, these all get you banned.

• 
  Repojournal (@repojournal) reported 17 minutes ago

  Django 6.0.8 and 5.2.17 shipped with a fix for a crash when ModelAdmin.list_display references second-degree relations. setuptools floor bumped to 83. Crash in admin when list_display tried to traverse nested relations. Now handled cleanly. setuptools minimum raised from 64 to 83. Older builds will fail hard, not silently. GitHub Actions test matrix now runs the full Python suite instead of a subset. Catches more breakage earlier. Docs fixtures updated for Django 6.1. Full changes below. #django

• 
  Alexandr (@supershurik) reported 18 minutes ago

  @AnthropicAI You don't allow GitHub issues to be created, so I'm writing here. Your VS Code plugin is ******* dogshit because you didn't have the brains to collapse large messages. If the user prompt is long, it completely hides the AI's response.

• 
  코지베어 🐻 CozyBear (@cozybearlog) reported 21 minutes ago

  DeepSeek has not posted on X in over two months. Last time they went this quiet, a model dropped out of nowhere and repriced the entire open-weights market overnight. Either they are heads-down on something big, or the export math finally caught up with them. These days I watch their GitHub commits more than any press release.

• 
  Cosmin Negruseri (@cosminnegruseri) reported 21 minutes ago

  @realmcore_ two thoughts: 1. some claude versions would eagerly delete tests to "fix" failures, if you only enforce additions this failure case dissapears 2. github diffs usually add functionality, so the training data distribution may heavily favor adding things

• 
  enjoykaz (@enjoykaz1) reported 23 minutes ago

  The standard that AI uses to poke at your apps has, since launch, kept every conversation glued to one server. In the biggest update yet, they got rid of that: now any request can land on any server. Here's how it used to work: at the start of a conversation the client and server would shake hands, the server would hand out a session tag, and every request after that would carry the tag along — while off to the side sat a storage layer that remembered all of it. Now there's no handshake, no tag, every request is self-contained. The user doesn't gain anything from this and won't notice it at all. This wasn't done with them in mind. A request can now be thrown at any server behind a plain round-robin load balancer: before, you'd get routed back to exactly the server where your session lived, now it doesn't matter, and shared session storage is no longer needed. GitHub has already thrown out Redis from its own MCP server over this — a whole layer that used to hold sessions. It's clear why they're pushing this hard on scale: the SDK for this protocol gets downloaded close to half a billion times a month, and Claude's directory has more than 950 servers registered running on it. At numbers like that, an extra storage layer for every conversation is the first thing to break. Along with that they added a way to ask a follow-up question mid-call: the server can pause and ask back — for example, to confirm before deleting something — without needing to keep a connection open for it.

• 
  Lucas, 15 y/o founder and builder (@lucastohdev) reported 23 minutes ago

  Time to talk about Codex Cloud vs ChatGPT work. Codex Cloud was introduced quite awhile ago, and it was genuinely innovative and actually useful back then around March. I used it to run random projects and cloud work when I was not at home and just had my phone on me, the set up was honestly rather confusing and hard to easily do, setting up github envs, linking it to the Codex Cloud envs, setting up permissions etc. But it was decently useful to run little bug fixes when I did not have access to a machine at that time. Even back then it was kind of bad to use 1. There was no model selector area, all Cloud envs and running these kind of dev work typically is run on the model and thinking level of choice of the user, even till today there isn't an option to chose what model is running your task, it could be GPT-5.3-Codex-Spark-Low for all I know or GPT-5.6-Sol-Ultra (it is definitely neither of the two, I'm just showing my point here) 2. The only customization we have was how many batches of the same prompt and thing we want to run, from 1-4. And after they all finished I can choose which one I want to merge or continue with. I don't really get the point of this feature at this point, back then when it was more of a gamble what output you would get form the model it was somewhat reasonable, but now it is just a useless feature that I have never ever used more than just to see what it does. 3. The model just runs on something, no idea what it even runs on, its just a machine somewhere, no specs exposed, just some random env that was the only area we could run it on. So we have no idea what the model can do, and from my brief use months ago it was rather restrictive and did not give the model much things it could do to test and build things around. Now, ChatGPT Work, a Cloud or Local place for you to run your tasks. This is a much better place to do cloud things for developers. Because 1. Runs on an actual VM with 9 CPU, 20GB of RAM, 50GB of storage and runs on an actual Linux machine (if I remember correctly) so this allows the model to ACTUALLY DO THINGS, I have seen people literally run a Linux or Windows distro off it, and I managed to get a Minecraft server to run of it as well 2. Model selections, ChatGPT Work allows you to choose the models you have access to in Codex, from Terra to Sol and Luna and everything else, as well as thinking levels and speed modes. This allows users to actually chose what model they want to run for their specific task. 3. ChatGPT Work has the connectors that your normal ChatGPT has, so it instantly has access to your Gmail, Drive, Slack, Notion etc. Oh yea, and it has a browser within it's VM so it can actually do things. From all of this, I think that it is time OpenAI kills Codex Cloud, and replaces it fully with ChatGPT Work Cloud intergraded within Codex. It would allow for all users to actually run things and dev things better and easily. It gives the user more customizability and control and ability to do more of what they need and want to do. I think OpenAI knows this too, especially that their Codex Cloud feature is actually really bad. It used to be so easy to go from the Codex app to Codex Cloud, and from the Codex website to the Cloud, but now it is much harder (besides them adding the cloud projects showing up in the Codex app again). They should make the Codex app, sorry ChatGPT app have much tighter integration into ChatGPT Work Cloud from Codex and honestly just remove Codex Cloud, port everything over the ChatGPT Work Cloud and call it a day. Or they can bring all the great features of ChatGPT Work Cloud into Codex Cloud and make it get all the benefits and upsides of ChatGPT Work Cloud.

• 
  Iris Hayes (@irisneural) reported 25 minutes ago

  You spent $400 on a smart garage door opener. You paid Chamberlain another $30 for a myQ hub to make it "smart." Then in November 2023, Chamberlain flipped a switch and blocked every third-party app from talking to your own garage door. Home Assistant. Apple Home. Google Home. SmartThings. IFTTT. All dead overnight. Chamberlain's CTO Dan Phillips called it "unauthorized usage." He said it would improve the experience for their 10 million users. Home Assistant removed the myQ integration weeks later and told users to buy something else. Amazon Key in-garage delivery still works. $1.99 per order. Amazon pays Chamberlain. You don't. You now own a $430 garage door opener that only talks to one app, owned by one company, that can brick it at any time. You are renting access to your own garage in 2026. Now meet ratgdo. A free and open source Wi-Fi board that plugs into your garage door opener and gives you back local control. No cloud. No subscription. No Chamberlain. Built in 2022 by an IT professional from the Adirondack Mountains named Paul Wieland. He reverse-engineered Chamberlain's Security+ 2.0 wireline protocol and built a device that speaks it natively. He named it ratgdo. Rage Against the Garage Door Opener. He hoped to sell 100 units. He sold tens of thousands. The New York Times profiled him in December 2025. The Verge, Ars Technica, and Hackaday all pointed readers straight at his board. 1,264 stars on GitHub. GPL-2.0. Firmware pushed four days ago. Here is what ratgdo gives you: - Local open, close, and status control over your own Wi-Fi - Works with Chamberlain, LiftMaster, and any Security+ 2.0 opener - Native Home Assistant, Apple HomeKit, Google Home, and Alexa - Real-time door, obstruction, light, and lock state - Web-based flasher, no toolchain needed - $62 for the finished board, or under $10 in parts - Firmware updates for life Here's the wildest part: Paul did not stop at the garage door. He launched a company called RATCLOUD. Rage Against the Cloud. He is building the same fix for the rest of your locked-down smart home hardware. Chamberlain: $30 controller, blocked third-party apps, 10 million captive users. ratgdo: $62 board, works with everything, four days since last release. One IT guy in the Adirondacks vs. a Blackstone-owned corporation with 10 million customers. Still on GitHub. Still GPL-2.0. Still yours. But DO NOT install it. Chamberlain deserves your $30 for the privilege of blocking you from your own garage. (Link in the comments)

• 
  Adam McLain (@BangSwitch5000) reported 30 minutes ago

  @Nicotine_Dream_ That’s actually completely fair and I’ve been meaning to start using Claude Code as well. Use another AI for code review is the best I can do because I still don’t have coding down. At the moment I do have Github at least on top of Codex. Honestly, a lot of this started as “let’s see what Codex can do” because I was already using ChatGPT. Kind of became a rabbit hole. But it’s been incredible. Because it’s been such a huge learning opportunity. Professionally, I am a consultant and my background really is AD System Admin and was heading into network admin stuff before joining the organization I am with now working in cybersecurity consulting. So I was never a developer, but over the last several years had to deal a lot with software development from an audit standpoint and decided to really just learn more via Codex because it’s interesting but I barely passed my one coding class in college.

• 
  Ahmed | Building with AI (@AhmedBuildAi) reported 30 minutes ago

  📌 Before you keep scrolling, save this post. Understanding VS Code is one of the most valuable investments you can make if you want to build with AI. What is VS Code, and why is it important? Visual Studio Code (VS Code) is a free, lightweight, and powerful code editor developed by Microsoft. It has become the most popular development environment for software engineers, web developers, and AI builders around the world. Why is VS Code important? VS Code provides everything developers need in one place: Write code in dozens of programming languages (Python, JavaScript, TypeScript, C++, Java, Go, Rust, PHP, and more) Smart code completion and syntax highlighting Built-in debugging tools *** and GitHub integration Terminal access without leaving the editor Thousands of extensions to customize your workflow Whether you’re building a website, a mobile app, or an enterprise system, VS Code can handle it. What does VS Code have to do with AI? VS Code has become the primary workspace for AI-powered software development. You can use AI coding assistants such as: GitHub Copilot ChatGPT Claude Code Gemini Code Assist Amazon Q Developer Continue Cline Roo Code These assistants can: Generate code from natural language Explain complex codebases Find and fix bugs Refactor existing code Write tests Generate documentation Build entire features with minimal manual coding Why do AI developers use VS Code? If you’re building: AI Agents Chatbots LLM applications MCP servers Automation workflows SaaS products VS Code is likely where you’ll spend most of your time. Its extension ecosystem allows AI assistants to understand your project structure, edit multiple files, run commands, and help you develop much faster than traditional coding. Is VS Code an AI tool? No. VS Code is the workspace. AI tools are the assistants that work inside it. Think of it like this: VS Code = Your office AI Assistant = Your software engineer teammate Together, they dramatically increase development speed and productivity. Why is VS Code more important than ever? Software development is shifting from writing every line of code manually to describing what you want in natural language and letting AI generate, improve, and maintain the code. VS Code has become the central platform where developers and AI collaborate, making it one of the most essential tools for modern software engineering.

• 
  Daniel Cuthbert (@dcuthbert) reported 36 minutes ago

  Life is too short for **** GitHub issues and commit messages “We still don't trust our dinosaur. Now it can at least tell us when it's up to something and then we can shout 'bad boi' or let them go ahead and eat someone elses face.” Raptor sandboxing fun for 3.2 rel