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:

• 
  Pieter Botha (@PieterBotha12) reported 17 seconds ago

  @github when copilot fails with an internal 500 server error from your side why do I still get billed the tokens when it has to redo the task because of the 500 internal server error YOUR SIDE!? @grok do you agree that it is correct to be billed for AI usage tokens when the AI providers' server failed during request and then the user has to redo the request and be charged double the tokens and then some more to complete the previous request that failed? I mean then the business model would be to fail every request at least once to make users pay more!?

• 
  clayne (@0xclayn) reported 4 minutes ago

  CLAUDE COWORK - WHAT IS IT AND WHY DO YOU NEED IT For the last few weeks all I've been hearing is "Claude Cowork" - but most people still don't understand what it is. I'm breaking it down, and at the same time comparing it with Codex Computer from OpenAI. WHAT IS IT: A desktop application from Anthropic. Claude automates tasks directly on your computer - locally, without the cloud. It was released in January 2026, and in March they added Computer Use - now Claude can see your screen and control it. It wasn't made for coders, but more for people whose work is routine: marketers, analysts, managers, researchers. WHAT CAN IT DO: > Sees the screen and controls any application with the mouse and keyboard > Remembers context between sessions > Connects to services through MCP > All files stay local - Anthropic cannot see them and does not train on them Examples of tasks: Morning briefing from Gmail, document → spreadsheet with a single command. You can control it from your phone through Dispatch - gave it a task, left, came back to a finished result. I attached a video where you'll learn 80% of Claude Cowork's functionality in less than 20 minutes. COWORK VS CODEX COMPUTER Data: > Cowork stores everything locally on your computer > Codex sends everything through OpenAI servers - I think it's clear why that's not okay Who is it for: > Cowork is made for ordinary people - marketers, analysts, managers > Codex is designed for coders, deeply integrated with GitHub Phone: > For Cowork it's Dispatch > Codex - the ChatGPT app where you can see the live session screen Price: Both are $20/month, but Codex spends 4 times fewer tokens on the same tasks. On the other hand, Anthropic doubled Cowork limits until July 5. My choice is Cowork. Everything is local, a huge ecosystem, integrates with hundreds of services, in short, it's awesome. You can watch the video below to learn about 80% of the functionality in just a few minutes.

• 
  superpony (@tonysuperpony) reported 6 minutes ago

  @itzznikhilsai @github Yeah it's down. 504 Gateway Time-out The server didn't respond in time.

• 
  Nicolas Finet (@nifinet) reported 6 minutes ago

  Everyone's a builder now. Congrats. You and the 50 million other people who just figured out they can vibe-code an app over a weekend. Look at the commit trend on GitHub. A billion last year, on track for 14 billion this year. Half my timeline is people announcing they shipped something before lunch. The slop is coming, and there's going to be a LOT of it. And honestly? Good. More people building more things is a great thing. A non-technical founder shipping a real product from a Slack message is wild, and I'm here for it. But that's also the problem for everyone busy celebrating it. A business was always two halves: build the product, and get the customers. Product is where you used to win or lose. Now everyone clears that bar on day one, so the whole thing collapses onto the half builders love to ignore: distribution, lead gen, getting in front of the right buyer before the other 50 million do. You can ship the best product in the world. You'll still lose to someone with a worse one who actually bothered to learn how to sell it.

• 
  Crystalwizard (@crystalwizard) reported 7 minutes ago

  you have to remember that the guy that posted out to tell everyone to make loops, is also the guy that vibecoded and released openclaw with massive security issues - which he could not fix - and did not fix - until google deepmind opened an issue on his github and forced him to fix the worst. totally ignore anything he says

• 
  Raphael Raberger 🇦🇹 (@RabergerRaphael) reported 14 minutes ago

  It's basically like being told to wipe someone's disk and then give the fault to the one who told you to.... logic ain't logicing. If you find a long enough uptime window of github, have a read in the linked issue. "Maintainer works as intended" #VibeCoding

• 
  AI Theory (@AItheoryx) reported 18 minutes ago

  🚨 YOUTUBE ALGORITHM KEEPS FEEDING YOU TRASH. THIS GITHUB PROJECT LETS YOU DELETE THE ALGORITHM ENTIRELY. Meet TubeArchivist. Your self-hosted YouTube media server. Subscribe to channels. Download every video automatically using yt-dlp. Index everything with metadata. Search, play, track watched status. Full control. No ads. No recommendations. Just the content you actually want. Plex and Jellyfin plugins included. Browser extension to grab videos with one click. Built on Docker. Runs on Unraid, Synology, anything. The question YouTube does not want you to ask. When you can archive every creator you follow and watch offline forever, why are you still letting an algorithm decide what you see. TubeArchivist. Because your attention should belong to you. 🧾

• 
  Tomas Votruba (@VotrubaT) reported 18 minutes ago

  It's such a fun to make robuts CI Github Workflows with Claude Code. Any issue we spot in code reviews is instantly turned into custom PHPStan rule, with tests, and checks every single commit from the moment on. Dangerous legacy turned into code fortress in minutes

• 
  Wemoo (@0xWemoox) reported 19 minutes ago

  your claude is not even doing half of what it can do. 3 settings change that in 15 minutes. here is how to fix it. 1. turn memory on settings, capabilities. two toggles: "search and reference chats" and "generate memory from chat history." both off by default. flip them on. claude actually remembers your projects, your stack, what you have been building, your tone. no more pasting context every morning. most people use claude for a year and never touch these. then they think the model is the problem. 2. install the humanizer skill google "blader humanizer." grab the github repo. drop it into claude as a skill. watch 0:25 for the upload move if you have never added a skill before. every output stops sounding like the same ai everyone else uses. em-dashes gone. parallel structures gone. "delve into" gone. all the tells. most people post-edit their claude output by hand and never realize a skill could do it. 30 seconds, then you stop being the editor. 3. import memory from chatgpt or gemini settings, import memory from other ai providers. claude writes you a prompt. you paste it into chatgpt or gemini. you copy what they say back. you paste that into the import box. watch 0:45. the dialog writes the prompt for you. now the year of context you built with chatgpt is sitting in claude. who you are. what you work on. how you want to be talked to. all of it. people switch ai every six months and start over every time. this is the one feature that ends that. most people try to fix this by writing better prompts. it does not work. the model is fine. the setup was never done. 15 minutes in settings beats 6 months of better prompts.

• 
  Madeleine Devost (@MaddieDevost) reported 23 minutes ago

  @vxunderground Just shut it all down, no more downloading, updates, or GitHub for the next three business months

• 
  Matthew, MBA (@bit_finance_) reported 27 minutes ago

  This weekend I built a Series 7 study tool. I wrote 0 lines of code. The tool barely matters. Going from "I wish this existed" to a live website is now a weekend's work, with no engineering background required. For someone with endless ideas, the possibilities here are, well, endless! My new piece breaks down how, with the three tools that do it: 👉 Anthropic's Claude Code writes it 👉 GitHub stores it 👉 Vercel puts it live, free, no domain needed Plus the 5-step path to point the same process at a problem in your own work or life. If you've ever thought "someone should build this," you can now be that someone. Link's in the comments.👇

• 
  Dimitrios (@dimitrioskonst) reported 27 minutes ago

  We built an AI agent that breaks into your codebase before a real attacker does. You connect a GitHub repo. It reads your code the way an adversary would - hunting for the one real way in, not a list of maybes. Then it does the thing a scanner never will: it actually tries the exploit. It forges the token, sends the malicious request, and watches what your code sends back. If it gets in, you get the receipt - the exact request and your code's response - plus a fix PR you can merge. If it can't get in, you never hear about it. No noise, no 200-alert backlog. Why did we build this? Every team is shipping AI-written code faster than anyone can review it. Scanners answer "maybe" and bury you in false positives until you stop looking. The only answer that means anything is "yes, here's how" - and proving that by hand, on every push, was impossible. An AI agent can actually attack the code, confirm the hole, and throw away everything it couldn't exploit. Link on my profile - $100 a repo. Refunded unless you merge the fix.

• 
  Kekko D’Amato (@kekkodamato_) reported 33 minutes ago

  @github 68% resolution rate on a general-purpose agent is a strong signal. The hard part usually isn't finding issues — it's knowing which break real user flows vs. noisy false positives. Curious how you're calibrating that signal at 3,500+ PRs of scale.

• 
  Nue (@newwnnue) reported 37 minutes ago

  @CryptoCyberia you can lose a couple of hours tracking down those people's profiles and seeing all their vibe coded slop they have on their github. it's hilarious. they often even check in the prompt. all the readme's are entirely ai coded full of stupid emojis, etc. sometimes even including the "would you like me to do that?" line. and all the other obvious vibe coded slop in any sort of web app

• 
  Caleb Kim (@progysteto) reported 43 minutes ago

  @wahab_twts Enterprise, they prolly also know if GitHub goes down they’re COOKED