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:

• 
  Bert Shim (@bert_programmer) reported 2 minutes ago

  @emanueledpt every suggestion in that screenshot is a service, not your machine. that's basically the answer. github is another connector like slack or gmail, so PRs and issues are reachable from anywhere, and whatever is sitting uncommitted on your mac may as well not exist

• 
  Poonacha Machaiah (@poonachaX) reported 4 minutes ago

  Third: Cursor launched Origin, code hosting with agents built in, on the same day GitHub went down for hours. Twenty years of the world's code sits on those rails. Ground you assume is permanent turns out to be a choice someone keeps making.

• 
  Ruthrasankar (@SRuthrasan71495) reported 4 minutes ago

  Lost 45 mins to Github and Supabase issues 😭😭 Unfortunately I connected the wrong github account to Supabase and when I tried to change, Github asked me to verify my account. And resert my password and email. Then it took some time for the previous repositories to appear. I was afraid to create a new repo when I can't even see my old ones - Result, I lost 45 mins in this process 🤯 I don't think I can pull it within the next 50 mins. What should I do?

• 
  Artes (@Artemwkz5) reported 4 minutes ago

  Your agent isn’t failing. It’s on payroll in the wrong building. Busy chat. Dead output. Same disease. This 11-second video is ByteByteGo’s floor plan for LangChain, AutoGen, CrewAI, LlamaIndex, and Semantic Kernel. Same model logos on the right. Five jobs on the left, in those yellow boxes. X keeps voting for a winner. The video is the org chart they should have used before they cloned a repo. Don’t listen to the captions. Watch the arrows. See the llama row? Hard hat at the bottom. Query up. Response down. That’s a librarian. Load → parse → index → ask your PDFs, DBs, APIs. If you dumped 400 internal docs into a generic “agent” and spent a month on retrieval, you hired a plumber to run the library. The weekend version is already drawn. See writer and editor, then three clipboards? That’s a newsroom. Process in the middle. Tools under the floor. If you needed draft → edit → three tasks and opened a group chat, you staffed a committee. Committees debate the assignment. Newsrooms file it. That’s why one path ships by Friday and the other still has “alignment.” LangChain is pipes: prompts, tools, vector stores, chains. AutoGen is a meeting: humans can cut in, code can run, tools can fire. Semantic Kernel is a socket wall: C#, Python, Java, then a plug into the model. None of that is a personality. It’s a department. One person who pauses on the right box looks like a 7-agent company. One person with five GitHub tabs open looks like they work in AI. Open the video. Find your yellow box. Fire the other four.

• 
  Aditya A. (@iamAdityaAnjana) reported 6 minutes ago

  Things to do after office hours > fix and update portfolio > update github readme for profile > ship some fixes for my startup > continue system design tutorial > have fun

• 
  Vlady (@danc_danc) reported 7 minutes ago

  A free skill hands Claude Code Apple's design rules Somebody on GitHub compressed years of Apple's design thinking into one skill, the exact colors, fonts, motion and rules they use across every product. Here's what it does once it's in your terminal: → New sites - you describe the page and it comes back premium and animated, in Apple's own type and motion → Existing projects - point it at a site you already shipped and it upgrades the design instead of you rebuilding it → Ask it to review your site - in a couple of minutes you get every minor thing that's off, plus the exact fix for each one I went and read the page after: it also tells you where motion does not belong, not just where to add it. Save this for the next time your site looks almost right (links below 👇)

• 
  Shoalstone (@Shoalst0ne) reported 14 minutes ago

  @Nvr_Knows_Bst @i3dwne1 @fishyfishydev I see it shows on axiom now, I think it was an issue where github was linked to my twitter but for some reason my google needed to be linked too and I had to delete my google-made acct and link to my old twitter-made acct, weird

• 
  Sinecan (@sinecanswork) reported 16 minutes ago

  Totally forgotten that I had a GitHub Copilot subscription. Got this email today saying that my annual subscription ended. It was great in the beginning and gave lots of value. After the price increase of AI models, my monthly quota started finishing within the first 3 days of the month. - It was extremely frustrating. - The value was not there anymore. Stopped using it, and even forgot that I had a subscription. I can imagine that I am not alone. Feel sorry about the team who have really worked hard to make GitHub Copilot great, as far as it goes. While their pricing looks attractive, wrt how quickly one consumes his/her monthly quota, it makes more sense to use - for instance - DeepSeek or GLM-5.2 kind of models with other harnesses. If money is not a problem, Claude is the one to use. Haven't used Codex for some time but it is likely the one that gives a lot more value now for the same price compared to Claude.

• 
  Khalid (@KhalidDevLog) reported 17 minutes ago

  @Ruimarcosjoao @expo Are you using an emulator or the Expo Go app? Note that PanelUI uses Expo SDK 57, so if you're using an older version, it might not work for you. Please create an issue in GitHub and tell me exactly what you're experiencing, including the type of emulator and whether you're using Android or iOS.

• 
  Het Mehta (@hetmehtaa) reported 19 minutes ago

  an AI agent was asked to find a Walmart MCP server. it recommended a malicious GitHub repo. another run turned an attacker-written README into installation guidance. the new social engineering target reads at machine speed.

• 
  Jimmy (@jimmy_harika) reported 21 minutes ago

  Love this. I was looking for something like this since whole github issue. Been experimenting with Forgejo but this seems even better for my use case. Gonna keep running both for now and will update the experience on weekend.

• 
  Gabriel Rubens (@gabrielrubenss) reported 23 minutes ago

  VPS deploy via GitHub (1/8): my deploys for Pensio, my AI journaling app, kept failing on the step that waits for the server to answer. A timeout in CI can mean many things, a slow build, or the more than 6,000 tests I run on every push, so I assumed the problem was mine.

• 
  Akshat Soni (@akshatsonic) reported 25 minutes ago

  @Ka_redemption I use my github from phone, the auth codes also stay on phone's authenticator app, same problem would occur there as well.

• 
  Visionary x AI (@VisionaryxAI) reported 25 minutes ago

  I’d put it at the end as a natural transition from idea → something you might actually build: I've been thinking about authentication for agents a lot lately. Right now every time we build an agent that needs to actually do something, we end up repeating the same flow. Connect Google. Connect GitHub. Connect X. Store OAuth tokens somewhere. Give the agent scopes and hope we've constrained things enough. It works. But I don't think this is how agent auth should work long term. What if there was just one broker between you, your accounts, and every agent you use? You set it up once. Connect Google, GitHub, X, YouTube, Slack, etc. to the broker. Then define what you're comfortable with: Google Drive → read automatically Calendar → read/write Email → ask before sending GitHub → ask before merging X → ask before posting Delete anything → never Now an agent doesn't need access to my Google account. It just tells the broker: I want to send this email to this person. The broker already knows whether that action is allowed. If it needs approval, I get a request on my phone. Approve with Face ID/passkey/OTP. The broker executes it or gives the agent a tiny capability that exists only for that action and expires immediately after. That's it. The part I really like is that the agent becomes almost irrelevant to the trust model. I could be using Codex today, Claude Code tomorrow, some open-source agent locally next week. I shouldn't have to keep giving each of them access to my digital life. They should all speak to the same permission layer. And that layer should know: who is asking what they want to do which resource they want to touch whether I've already allowed it whether I need to approve it and exactly how long that permission should exist Basically: Agents ask. The broker decides. I approve when necessary. The service credentials never leave the broker. OAuth solved "can this app access my account?" I think agents create a slightly different problem: "Can this agent do this one thing for me, right now?" As agents start moving from answering questions to actually operating software for us, I think this distinction is going to matter a lot. Maybe the missing primitive isn't another agent framework. Maybe it's a permission layer for all of them. I'm thinking about building an open-source version of this and seeing how far the model can actually go. If this is a problem you've been thinking about too, or you'd want to contribute, build integrations, poke holes in the architecture, or partner on it, I'd love to talk.

• 
  Julien Zammit (@JulienZammit1) reported 27 minutes ago

  5. can your agent edit .github/workflows? if a stranger can open an issue, a stranger can reach your CI. 1 min