GitHub status: access issues and outage reports

Problems in the last 24 hours

The graph below depicts the number of GitHub reports received over the last 24 hours by time of day. When the number of reports exceeds the baseline, represented by the red line, an outage is determined.

Most Reported Problems

The following are the most recent problems reported by GitHub users through our website.

• Website Down (53%)
• Errors (33%)
• Sign in (14%)

Live Outage Map

The most recent GitHub outage reports came from the following cities:

City	Problem Type	Report Time
Paris	Website Down	7 days ago
Ahmedabad	Errors	13 days ago
Delme	Sign in	14 days ago
Lyaud	Website Down	14 days ago
Catania	Errors	16 days ago
Inverness	Website Down	29 days ago

Full 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:

• 
  Bash (@bashirbuilds) reported 49 seconds ago

  One of the hardest things about building a SaaS product today: You don't control most of the systems your product depends on. Stripe. OpenAI. AWS. GitHub. Resend. Clerk. Your code can be perfectly fine and your product can still break because something outside your code changed. The more dependencies you add, the harder this becomes. That's the problem I'm building Reeno to solve.

• 
  Dr Milan Milanović (@milan_milanovic) reported 3 minutes ago

  How Cursor made *** scalable The thing with *** is that it never was designed to be scalable. Your repo lives on the disk, and *** client expect every read to be consistent. This was a problem on GitHub, where shared filesystems and replicated storage failed before 2013. The GitHub built 𝗦𝗽𝗼𝗸𝗲𝘀, and it became the industry standard. This means that every repo is stored as three full copies on three servers, and every push runs a vote (three phase commit). A majority of servers must confirm before it exists. This works, but with high cost, because every push is slow as the slowest server. When we add new servers, it makes it even slower. Now Cursor took some opposite direction with 𝗖𝗼𝗻𝘁𝗶𝗻𝘂𝗶𝘁𝘆. The repo history is now written as a log in S3, and this is only source of truth. Any push counts only if it is located in the log. The servers don't need to keep anything important, they are just cache. Any server can take a push, and idle repos are dropped from disk and rebuilt from the log when it is needed. This resulted in 120 pushes per second on standard S3, and over 300 on S3 Express. Their tests have shown that read capacity grew linearly up to 100 replicas. Why is this important now? Because of AI agents mostly. We now have more code, PRs, CI runs and many small repos. All of these repos would need three full copies in the old model. This means that we achieve scale by removing parts, not adding them.

• 
  franks 🇦🇷 (@francdetank) reported 5 minutes ago

  @rmansueli thanks Rodrigo! I cant create one because I got a problem . My account is pegged to Github and github has flagged me. That way I am blocked to enter the dashboard. I would like to connect my account to my email instead of github.

• 
  Shaun Patrick SteWaRt (@ShaunStewart) reported 10 minutes ago

  @annalea_l Honestly, I really want to see this. You have to understand: I am the type of person who can learn and do anything on the fly at a high level, and I just threw myself into this whole developer and engineering world. When I first started learning all this stuff, I already knew what I wanted and how I wanted it to operate, regardless of what I saw on X or what was considered possible. Before I even started following hundreds of developers and learning about harness engineering, mechanical engines, persistent memory, and all that, I put my brain on a GitHub repo. Everything is shared across every machine, every cloud entity, and every AI. I am not even technically an engineer or a developer, and I don't actually write code. But once I started following all these people and saw all the problems they complain about, I thought: this isn't even my trade, and I have already solved all these little things everyone says are impossible. Why aren't people talking about developing your harness more and making things more mechanical, instead of just arguing with a terminal all day long? Whenever I see articles people post on X, I run them by Claude or Grok and ask, "Should we implement this?" I have hundreds of bookmarks, but every single time they tell me, "Nope, your brain's better. Nope, your harness is better." I can never find anything built better than what I have or what I am currently working on. The brain and harness setup is basically like a mini operating system. All that said, I am really looking forward to seeing something I can use that goes far beyond what I am already doing. I definitely want to see your end product, it sounds very interesting.

• 
  Apoorv (@apoorvdarshan) reported 12 minutes ago

  @Dimillian these issues have been multiple times reported by users on github i hope open ai fix those, as well as please consider using native than electron

• 
  Abdallah Shaban (@AbdallahSh07) reported 12 minutes ago

  @10VBacik @FlutterDev Got it! did you create a GitHub issue on their end for this? seems like there is enough info in that gist to help them investigate!

• 
  Jeremiah K (@neolaj) reported 18 minutes ago

  @TiborAntal Gradually figuring out how to scale coding agents. Started with 1, manually handling all the ***/GitHub work. Moved to 3 because I had more ideas than one agent could keep up with. That’s when the real problems started: squashing, merging, branch drift, conflicts. I ended up rebuilding the workflow around deterministic *** logic, worktrees, ephemeral branches, and syncing with the integration branch before changes begin. Now I’m running 6: • 1 orchestrator (Fable or Opus) • 4 coding agents • 1 integration agent reviewing and merging PRs Building the process around them was the hard part. Right now im just doing a couple of PRs (using ORCA on windows on my home computer)

• 
  WuBu ⪋ WaefreBeorn 🇺🇸 👑 (@waefrebeorn) reported 18 minutes ago

  hey @Teknium @yeahfortommy please add the amd portal too even if tou have to send tommy into the AMD headquarters to get them to fix the links (you have to sign up for american then link through github, then you can access the models free, tommy needs to pull teeth but they have free api)

• 
  Prophet Joel (@2happyCSGO) reported 21 minutes ago

  I personally hated Claude because it refused to do almost anything I asked it to do so have no idea of how the speed is but gemini-cli was unusable for non enterprise users. Github CoPilot both GUI and cli is pretty good. Grok Build is what I'm using mostly and not yet had any issues with the speed but I want to go full local asap, scouting for 3090's atm. Just to be able to run "uncensored" models that don't ***** like Claude is reason enough for me to prefer local over Cloud but also cloud is ******* expensive, I have SuperGrok 100$/month and CoPilot Max 100$/month and that is barely enough. I'm trying to make my own Jarvis so I need to build my own RAG, memory, librarian, SRE Agent that understand how to use all tools and I also get crazy new idea's all the time lol Just made my first alpha of a tool that can wipe basically anything you don't want in Windows11. Basically Chris Titus clone but on steroids, this isn't just a debloater, it's a Grim Reaper 💀

• 
  Vitaliy Salyuk (@vitaliysalyuk) reported 21 minutes ago

  @openclaw @github Fix your updater and I might give it another shot.

• 
  ONCHAIN COP (@OnchainCop) reported 23 minutes ago

  @PogNyx lmao anyone can create a github issue retards this guy is a larp

• 
  Josh Hamilton (@nearbycoder) reported 26 minutes ago

  @theo If GitHub is down does it fall back to a cached version I’m guessing?

• 
  Kim Burgaard (@kimburgaard) reported 27 minutes ago

  Back when GitHub added Copilot PR reviews, it helped me keep up with the growing volume and size of our pull requests, which were increasingly being written by Copilot too. Over time I grew comfortable feeding Copilot's review comments straight back to Copilot to fix, and mostly spot checking when critical functionality was involved. When GitHub updated the Copilot pricing model I switched to Claude Code, but kept the Copilot review feature on for a couple of months. When the monthly bills for Copilot AI usage alone started rivaling the Claude Code Max plan, giving Claude Code PR review duties seemed like an obvious cost saving move. Plugging Claude Code into our PR review process immediately went south. The first PR churned with fixes to findings that resulted in more findings, and fixes that propagated up and down the call chain. I threw the PR away and started over, but the next attempt churned just as badly. Turn count on its own was never the signal. Copilot had taken ten turns on a rate-key cleanup the day before and nobody minded, because the findings thinned as it went — 5, 4, 3, 3, 3, 4, 1, 2 — and it merged. The cached-token billing PR I put through Claude Code took nine turns and produced 123 inline findings, and the ninth round was still returning fifteen. I closed it without merging. Looking closer at Claude Code's review findings, it was clear it reported far more issues than Copilot ever did, and among legitimate bugs and concerns, it made lots of comments about latent and speculative issues including possible race conditions and error propagation, things Claude Code would then try to fix one by one in isolation, often ignoring existing patterns in the code base. The code-review workflow is built into Claude Code and cannot be customized other than a few options, so the only place to intervene was on the other end, in the session where I used to just ask the coding agent to address the review findings. The first improvement was to direct Claude Code not to blindly fix all findings, but to defer findings not directly related to the task at hand to new issues. That helped reduce the PR churn, but blew up our issue backlog. The next improvement was to ask Claude Code to ignore speculative findings and disregard most latent findings unless they indicated high risk of unrecoverable damage in production. Finally, I had to stop Claude Code from authoring prescriptive issues with detailed implementation instructions. The result is a skill that triages PR review findings, and a skill for authoring and updating issues. After a few iterations of the skills, I've been able to complete ten PRs over a couple of days, bringing back the pace we had before. I've made the skills available in a public GitHub repository (link in the first reply). Let me know if you find them helpful.

• 
  Marcelo Retana (@mretsal) reported 28 minutes ago

  Every time @github goes down they should have a plan to please their users. Give me free credits for actions for example 👍🏼

• 
  🐻‍❄️ (@Nerevarineeee) reported 30 minutes ago

  @jiriknesl @napenforcer yeah vscode is bloated electron slop and github....... do i even have to mention the down times and unavailability? it has literally became much worse since ms acquired it, so wtf are you talking about?

• 
  Ishank (@IshankDev) reported 30 minutes ago

  7/ 16k+ GitHub stars. Built for people who want control, not another marketing-suite login.

• 
  ѕaℓoηι ηαяαηg (@thesaloninarang) reported 31 minutes ago

  I review resumes for DevOps beginners at meetups, and the same fixable problem shows up every time: bullets that describe the tool instead of what YOU did with it. Let me show you the difference with real rewrites. BEFORE: "Worked with Docker and Kubernetes" AFTER: "Containerized a 3-service Node app and deployed it to a local Kubernetes cluster with health checks and resource limits" BEFORE: "Knowledge of CI/CD" AFTER: "Built a GitHub Actions pipeline that tests, builds and pushes images on every merge to main" BEFORE: "Familiar with Linux" AFTER: "Debugged container startup failures using logs, exec and inspect on Ubuntu servers" See the pattern? Verb, artifact, specifics. A hiring manager can picture the AFTER versions. The BEFORE versions could be copied from any job description, and that is exactly how they read. You do not need production experience to write bullets like this. Personal projects count when you describe them concretely.

• 
  Speen Bhai (@Speenbhai) reported 31 minutes ago

  @johnternus Hi John. Congrats Let us see what new you bring with you. Affordability and intelligence. You have source code or an AI and can get it from GitHub. Why not turn 234 million iPhones to a massive distributed server infrastructure with zero power consumption

• 
  trulite (@trulite007) reported 39 minutes ago

  @Qromerolauro @mkliku @radius_browser Like a simple example would be have a list of my urgent GitHub issues and start an agent for it . Or a dashboard in which buttons start investigating issues. Of course I just need the webpage to be able to access radius tools. I m thinking secure way is an extension

• 
  Suryansh Tiwari (@Suryanshti777) reported 40 minutes ago

  6. The Dependency Incident Check Grok has native real-time search across X. Breakage gets posted there hours before the GitHub issue is triaged. No other coding model has that feed. "You are a build engineer whose first move on a broken pipeline is to work out whether it broke for everyone or only for me. Search X and the web, last 14 days. Check: - Is anyone else reporting this failure with this package and version, and when did the reports start - The exact release that changed behaviour, and the changelog line that admits it - Whether maintainers have acknowledged it and what they recommended - The pin or patch people settled on, with the tradeoff of each - Whether this is my problem instead, and what evidence points that way Give me the verdict in the first line: their bug or mine. Then the evidence, newest first, with links. My failure: [PASTE THE ERROR, THE PACKAGE AND VERSION, AND WHAT CHANGED ON YOUR SIDE RECENTLY]"

• 
  Yeemio (@yeemio) reported 44 minutes ago

  owlrunkit is on github now. public corresponding source for the npm package. issues go here.

• 
  Harman (@itsharmanjot) reported 45 minutes ago

  Runs macOS on iPad to enable pro apps like Xcode and Terminal directly on the device This isn't a remote desktop or a streaming trick. It's real macOS booting on the iPad itself. It's called Virtual Mac on iPad. It runs a full copy of desktop macOS directly on Apple Silicon iPads, using Apple's own virtualization stack pulled out of macOS and rebuilt to load on iPadOS. Real macOS, on the tablet, offline. → Runs macOS 12 Monterey all the way up to macOS 26 Tahoe → Real pro apps on device: Xcode, Terminal, Final Cut Pro, Logic Pro, Pixelmator Pro → Metal GPU acceleration in every supported macOS version → Works with touch alone: tap to click, two-finger scroll, on-screen keyboard, no Magic Keyboard needed → Runs entirely on device, no server, no streaming, no account → Installs straight from Sileo in a couple of taps Here's the wildest part: It doesn't just match the desktop Mac virtualizers, it beats them. Virtual Mac is the first tool ever to run Final Cut Pro with OpenGL and OpenCL acceleration inside a macOS VM, something even UTM and VirtualBuddy running on a real Mac can't do. And it was built by a handful of community devs who extracted Apple's Hypervisor and Virtualization frameworks by hand, then used agentic coding to shim every missing API iPadOS didn't have. One honest note: this needs a jailbroken M1 or M2 iPad running iPadOS 16.3.1 or older. Apple removed the hypervisor from iPadOS 16.4, so newer versions are locked out for now. If your iPad qualifies, it's the closest thing to a Mac in a tablet that has ever existed. 1,423 GitHub stars. MIT License. 100% open source.

• 
  volkdude85 (@volkdude85) reported 45 minutes ago

  @SentientSquirel @linuxuser1996 So you are you scared of github then. Look dude I have fun on computers and don't take myself seriusly because I have destroyed enough OS's over to not worry about it because I just fix it, If the contents of your PC make you this paranoid its time to check your kink.

• 
  paulrodturner (@paulrodturner) reported 45 minutes ago

  @supabase Is anyone else having issues logging in via Github?

• 
  Yash (@dewyashtwts) reported 46 minutes ago

  recently integrated Resend into @supercodeai review so founders get PR alerts with real risk context I'm amazed what we found out when we put @coderabbitai / @greptile through the same PR: 1) coderabbit / greptile: - stamped it “low risk, mergeable” (4/5) clean - forgot context from the last PR - no tests suggested, no safety checks - zero memory of previous regressions 2) supercode review on the exact same PR - flagged a real vulnerability in the diff - noticed i’d pushed credentials into `.env.example` - pulled in history from past PRs + explaining how this change could affect and break them - downgraded it to "medium risk, fix before merge" state - attached concrete fixes + patches scoped by severity this is the difference between 'LLM summarizer for github' and an actual swe agent that cares about your production

• 
  Rusty Williams McMurray (@1RustyMac) reported 50 minutes ago

  Persistent AI doesn’t have a supply chain problem at the model. It has a supply chain problem at the moment it changes its mind. Personality drifts. Tools get installed. Memory accumulates. The thing you shipped on Monday is not the thing answering on Friday. We can attest who built the weights. We still cannot attest who authorized what the agent became on Tuesday. That is the hole. Who is allowed to let it change? We built Living Supply-Chain Security for Persistent AI Organisms around one law: The organism may propose evolution. It may not authorize it. No trace, no drift. If an agent wants a new personality, a new tool, a new maturity, or a rollback — that change does not happen because it felt confident. Confidence is not a key. Self-narration is not evidence. Evidence is not interpretation. Interpretation is not authorization. Authorization has to come from outside the organism, bound to the exact change, used once, and written into an append-only history. Even a rollback cannot erase the record. You can restore a prior state. You cannot pretend the detour never happened. Default-deny. Hash-chained. Externally signed. We froze battery v1 on July 5 and ran it against the paper’s own claims. It held. That is executable evidence. Not a proof. Not a production blessing. Not “alignment, solved.” If it can’t be attacked, it isn’t finished. GitHub later this week. Come try to break it.

• 
  radhika (@RaadhikaThacker) reported 54 minutes ago

  First surprise: a GitHub issue form isn’t a form. It’s a YAML file. You describe the field- this one’s a dropdown, this one’s required and GitHub builds the UI from that. I did not know that.

• 
  The Oracle (@scientist1q) reported 55 minutes ago

  when my Oura ring detects a cortisol spike from a GitHub Actions failure, Hermes (Fable 5.1) detects it and sends a 900 word root cause analysis, Hermes dispatches the work to my 12 Grok Bot employees, The Chief of Operations bot approves the fix while im watching rezero

• 
  Kahris (@chrissotraidis) reported 55 minutes ago

  @NickogSo I haven't tested on LiveContainer. Feel free to submit logs via GitHub issues and I'll check it out. I haven't had any other reports of that happening for either build.

• 
  CATIRL 🏳️‍⚧️ (@CATIRL_9) reported 58 minutes ago

  @mminhamina Google GitHub "open grind", solves your problem