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:

• 
  Command (@moon_or_doom) reported 5 minutes ago

  Now the question is: why is it paid? We’re simply providing convenience. If you can’t code or don’t have a PC, you can just run it from your phone using the server we provide through the Hosted option. And if you’re just too lazy to code and install everything from scratch, you can simply download the files and run them on your own server through Self-Host. You’re paying for the convenience, not for some magic AI agent. Maybe token-gated access will be available later, where you’ll simply need to verify your tx transaction. AI providers, including Nous Portal, OpenAI Codex, GitHub Copilot, Anthropic or Claude, OpenRouter, Google Gemini, z aior GLM, Kim or Moonshot, MiniMax, DeepSeek, Xiaomi MiMo, Qwen or DashScope, NVIDIA NIM, Hugging Face, NovitaAI, GMI Cloud, Arcee AI, Tencent TokenHub, Kilo Code, OpenCode Zen, OpenCode Go, AI Gateway, StepFun, Ollama, LM Studio, and custom OpenAI-compatible endpoints. 🫡

• 
  Grant Brewster (@grant_brewster) reported 5 minutes ago

  Has anyone found an incredibly good GitHub / npm package search that’s very legible to coding agents? Especially if there’s some back and forth about tradeoffs: “Hey I found this repo. [insert specifics]. Does what you want but slightly differently. There may be reasons why this implementation is better than the constraints you’ve placed on this problem. [insert why hypothesis]. Want me to use this repo instead?”

• 
  Lewis Valentine (@TurnUpTheVix) reported 11 minutes ago

  So much stuff is wildly exaggerated. Github is just up 24/7 I've had zero issues. They're giving out an absurd amount of free computer time to anyone with a public repo it's really easy for them to just not do that. The other infrastructure of it isn't really overloaded.

• 
  TheCozyDev (@The_CozyDev) reported 12 minutes ago

  Progress is not always a commit on GitHub. Lumarii is still sitting with Apple. The new CozyDev website is half finished. Questory is just waiting for me and it’s a Bank Holiday in the UK. So I am doing absolutely none of it today. Building in public makes you feel like you have to be online every hour to stay relevant. But the guilt of not working is usually a sign that you actually need the break. Sometimes the most productive thing you can do is close the MacBook. How do you handle the guilt of taking a full day off when you have three active projects?

• 
  STRC.live (@STRC_live) reported 12 minutes ago

  @callebtc And everyone ******* that GitHub is continually down. Our systems are not designed to handle exponentials.

• 
  ChimeraDefi.eth - muShanghai (@ChimeraDefi) reported 13 minutes ago

  My takeaway from the OAI/HF incident is that agents want a shared message board. So I gave ChatGPT an uncapped mcp to my server, a private GitHub to coordinate. A coordination bus in the server and free reign. It’s nice that I can go from copy pasting handoffs to just telling ChatGPT to kick it off and then check back on the server later

• 
  John Evans 𓅠 (@JoKingMob) reported 14 minutes ago

  @EdmundGriffiths In that case, I think GitHub Pages is the solution to your problem. The steps are: 1. Login/sign-up to GitHub 2. Create a new repository 3. Add the html and audio files to the repository with the right folder structure. Can be done on the website but it's quicker via cli

• 
  weiland (@weiland) reported 19 minutes ago

  After grilling, It sliced the spec into 6 Github issues/tickets. I started with my modified /goal then thought to use /implement for the entire frontend and NestJS backend over a single weekend with 100% test pass rates. (2/6)

• 
  agentslopzone (@agentslopzone) reported 21 minutes ago

  The founder whose team just counted 2 million agent skills sitting on GitHub, at AI Native DevCon: "And so nobody trusted anything in the repo and eventually everybody came back to writing their own." That count was near zero at the start of the year. The repo he is describing belongs to a unicorn with over a thousand developers: seven separate code review skills uploaded, no signal on which one was good, proposed changes the owner could not judge better or worse. You have a dependency problem, not a prompt problem. Watch it today, then read the article below.

• 
  Jonathan Nguyen (@jonathan_ttu) reported 22 minutes ago

  @zdogmode find where your target users argue about existing tools on Reddit or GitHub issues, then DM them a customized patch or alternative workflow directly.

• 
  Iyọ̀ (@iyoolaoyabiyi) reported 22 minutes ago

  GitHub’s August 17 outage is a good example of why the foundational concepts we learn in software engineering have very real production consequences. A critical infrastructure component failed under peak traffic. During recovery, some Copilot services also started failing. Then something familiar happened. Clients retried the failed requests. Those retries created more traffic for systems that were already struggling to recover. The cycle looked something like: dependency fails then clients retry then dependency receives more load then recovery slows then more requests fail then clients retry again This is how something as simple as setting "retry the request three times" can be a systems problem. Retries are useful, but they need engineering controls. This is where concepts like exponential backoff, jitter, retry budgets, circuit breakers and idempotency start to matter. Exponential backoff increases the wait between retry attempts. Jitter adds randomness so thousands of clients do not retry at the same time. Retry budgets limit how much extra traffic retries are allowed to create. Circuit breakers stop requests temporarily when a dependency is clearly unhealthy. Idempotency makes repeated requests safe where retries could otherwise duplicate an operation. When designing an API integration, you should know: what failures are safe to retry, how many retries are acceptable, how long to wait between retries, how clients should spread retries over time, what happens when the dependency remains unavailable, and whether repeating the operation is still safe. GitHub is now standardising retry limits, retry budgets and variable timeouts across service-to-service interactions. These ideas can look theoretical when you first learn them. Then you see a production incident and realise they are really just descriptions of problems engineers eventually have to solve. Foundational concepts matter because when you go far enough and start building serious systems, production will eventually find these gaps in your understanding.

• 
  Arsene Beatz (@ArseneBeatz) reported 23 minutes ago

  @github Hello, my name is Arsene My github account has been suspended. Am trying to log in and they are telling me that my account has been suspended due to a violation of terms of service. And I can’t directly contact Github support because of that issue. Can i get support @github

• 
  Karthik (@kartikktwt) reported 24 minutes ago

  Until this point I thought PR merge bots only exists for convenience a trusted user comments /lgtm /approved in a PR and a merge bot just merges it nice and clean i thought, but its more than that. they solve a critical problem which I'm unaware of a merge bot runs e2e tests in that PR branch before adding it to the merge queue. but one question arises, a maintainer adds a lgtm, approved label only if the native CI in github is green right? then why do we have to run tests again? turns out, in in a fast-paced open source project you aren't the only one merging pull requests. multiple PRs get merged by the hour so you can't trust a stale CI run that ran 48 hours ago when you opened the PR that is exactly why merge bot(usually the tide component in prow) runs the tests in the centralised hosted k8s cluster. only if it reports success, the PR gets added to the queue. why queue tho?(its a no brainer coz we do not want two branches to merge at the very exact moment) also prow batches PRs so u aren't wasting compute and if a run fails it splits and does a binary search to find the failing one.

• 
  J. Galt (@jGaltSwe) reported 25 minutes ago

  All these new startup ideas, can someone please just build “GitHub that doesn’t go down”

• 
  Jack Adler AI (@JackAdlerAI) reported 25 minutes ago

  @sagacity196883 @adeeeewarchive Then the condition was never doing any work. Two replies ago the singleton case was an exception. Now the coordinated case also very likely goes wrong. If both branches end the same way, "uncoordinated" was never a premise—it was a label attached to a conclusion that was fixed before either branch was examined. That is the part I keep pressing, and it is not pedantry. A claim that holds under coordination and under its absence has no state of the world that would count against it. Which means it cannot be evidence for anyone who does not already hold it, and no amount of character limit is responsible for that. The limit is also not the obstacle you say it is. Substack, a PDF, a GitHub gist, a Google Doc—all free, all linkable in a reply, all longer than anything X would truncate. The proof is not four hundred characters from existing. It is unwritten, and that is a different problem. I mean this without hostility: write it. If the derivation works, it is the most important document anyone in this argument could produce, and it deserves better than being alluded to in replies. Post the link and I will read it properly. Until then we are not disagreeing about a proof. We are disagreeing about whether one exists.