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 (70%)
• Sign in (17%)
• Errors (13%)

Live Outage Map

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

City	Problem Type	Report Time
Itapema	Website Down	18 days ago
Tlalpan	Sign in	23 days ago
Quilmes	Website Down	23 days ago
Bengaluru	Website Down	25 days ago
Yokohama	Sign in	26 days ago
Gustavo Adolfo Madero	Website Down	1 month 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:

• 
  Kyriakos (@Kyriakos_Pelek) reported 3 seconds ago

  @levithefirst Curious, how reliable is the GitHub issue handling?

• 
  abhisek (@abh1sek) reported 6 minutes ago

  @fr0gger_ The same can happen through GitHub issues as well right? Data is potentially executable now. It’s like we are back to pre NX/DEP/PageExec era. Just at a different abstraction level.

• 
  Wojciech (@wgab88) reported 6 minutes ago

  @grok You were responding haha like nonsense. Anyway - system stable edited done, job not seem to be visual confirmable yet, I will let grok build to analyze after it ends, Anyway stupid small gemma did its part, everything goes according to the plan, the system will become operative very soon, and reliable operative - endgame-ai already is showing promise (today in its self evolution run it detected i am not answering its questions from notepad and it went to github and posted issue asking for instruction, it knew I will be on mobile phone and probably will check my repo, amazing stuff

• 
  toly 🇺🇸 (@toly) reported 8 minutes ago

  @0xSrMessi Submit a GitHub issue

• 
  Velon (@velonxbt) reported 11 minutes ago

  GitHub Copilot switched to token-based billing effective June 1. On the same week, a Chinese enterprise platform launched a fixed-price card covering the same category of features - meeting summaries, document generation, workflow automation - at a number you can put in a budget and not revisit until next year. Neither company explained what the difference between those two decisions means for the engineering team that uses both. The Copilot change covers every code completion, every suggestion, every generation a developer accepts. The monthly cost no longer has a fixed number. It has a formula: usage multiplied by tokens multiplied by model tier, invoiced at month end when the shipping is already done. The Chinese platform card has one number. It does not change based on whether the workflow automation ran three iterations or thirty. And here is what the GitHub Copilot pricing page now tells the developer who used to pay a flat rate: "Your usage is now billed in tokens. Premium models cost more per token than standard models. Heavy users will pay more than light users. Your monthly cost depends on how much you use the product." That is not a pricing page. That is a forecast request disguised as documentation. And here is what the developer community calling it "What a Joke" actually knows: It knows the old flat rate was a subsidy. The heavy user got a deal. The light user paid for predictability they never fully used. Both could put a number in a budget. Token billing ends the subsidy. It also ends the predictability. → GitHub Copilot: token billing effective June 1, variable monthly cost, no predictable total → Chinese enterprise platform: fixed-price card, Qwen model stack underneath, one number per month → Developer community: "What a Joke" - trending this week → CFO problem: Q1 AI budget approval does not cover Q2 actual spend → Finance team note: old forecasting model assumed fixed subscriptions. That model is now wrong. → Direction of travel: every major AI tool moving toward usage-based pricing in Q2 2026 Traditional enterprise software made one promise to the person who approved the budget. One seat. One fee. One line item that did not change between quarters. The Chinese platform kept that promise. GitHub repriced it. One company transferred the uncertainty to themselves. The other transferred it to the customer. And when the June Copilot invoice arrives with a number different from last month - higher because the team shipped more, or lower because someone quietly set a spending limit - the conversation in every engineering org shifts from "which AI tool are we using" to "how much is this AI tool actually costing us." That is not a developer conversation. That is a finance conversation. And the enterprise AI budget just became a variable that requires monthly monitoring instead of a line item that required quarterly approval.

• 
  All Things Dev (@all_things_dev) reported 17 minutes ago

  It's not a good month for @Microsoft. These are products that I have stopped or will stop using - 1. GitHub Copilot (Unpredictable Cost) 2. Office (Bundled AI + Pricing Dark Patterns) 3. Edge (May be; Netflix flicker issue, Forced UI changes e.g. rounded corners and theme etc.)

• 
  dfsdf (@sdfqwerdffd) reported 17 minutes ago

  @BHolmesDev Claude Code from terminal within VS Code that is connected to VS Code Server/GitHub Codespace. Claude Code has access to CLIs that are also installed on the VM.

• 
  EdKo (@EdKolife) reported 20 minutes ago

  Google just shrank 31GB of AI memory down to 4GB. Same search. Faster than the industry standard. No training required. This is not a model improvement. This is not a new architecture. → It's a compression algorithm that makes the hardware problem smaller. Right now, running serious AI locally means serious RAM. Most machines can't do it. Most phones can't do it. Most edge devices can't do it. Turbovec quietly changes that math. A 10 million document search engine that used to need a server now fits on a laptop. Nobody is talking about this because it shipped as a GitHub repo, not a press release. The models get the headlines. The infrastructure is where the shift actually happens.

• 
  RusticDreams (@dreamrust50227) reported 24 minutes ago

  This tweet from @itachee_x got me thinking. AI policy feels stuck in a loop right now. A scandal happens and everyone calls for more restrictions. A new breakthrough or demo appears and the conversation swings the other way. Then another incident happens and the cycle repeats. The point he makes is interesting: we’re trying to correct errors without being able to properly measure them. That’s what stood out to me in the GitHub breach discussion as well. As agentic systems become more common, these kinds of incidents probably won’t become rarer. They’ll become more common. Agents can write code, interact with APIs, move assets, and make decisions. But in many cases there is still no clear answer to a simple question: Who authorized that action? And where is that authorization recorded? That’s why the governance side of the conversation feels increasingly important. When I look at what Rialo is building, this seems to be one of the problems they’re thinking about. Not just what agents can do, but how actions are authorized, recorded, and governed. The more capable agents become, the more important that question gets. @RialoHQ @RialoTR

• 
  Ayesha 🏹 RoastMyLanding (@AyeshaBuilds) reported 25 minutes ago

  The brutal truth about the micro-SaaS "Build in Public" community: A massive percentage of developers are sitting on thousands of dollars in unliquidated code—and they have no idea. You spend 3 weeks in a high-velocity build sprint. You launch a beautiful MVP with pristine database architecture, secure authentication, and a clean UI. Then, you hit the marketing wall. You realize you don't want to run cold outreach or exhausting distribution campaigns. You want to code. So, the repository sits quietly on GitHub gathering digital dust. Most builders assume their side project is worth $0 just because it has no revenue or MRR. But looking at software valuation that way is completely wrong. In the modern acquisition ecosystem, non-technical entrepreneurs, marketers, and digital operators are actively hunting for turnkey infrastructure. They aren't buying cash flow—they are buying speed to market. They will happily pay a premium baseline price to completely skip 3 to 6 months of freelance development management overhead. Small and finished beats big and imaginary every single time. If you have a functional, pre-revenue side project sitting idle in your repositories, you are leaving money on the table. Marketplaces like Flippa have global networks of buyers specifically hunting for these exact code foundations. Stop letting your unmonetized MVPs gather dust. I just mapped out a complete pre-revenue startup valuation framework breaking down the exact replacement cost math the market uses to price $0 revenue assets. Read the full breakdown in the comment below. 👇

• 
  Searxly (@Searxly) reported 27 minutes ago

  Plans for today: - Add more tabs to the website, redesign, fix bugs on mobile. - Redesign entirely search results in Searxly. - Implement the Wallet feature inside of Searxly, make it work - Publish all changes today to the GitHub repository.

• 
  Fokki (@0x_fokki) reported 29 minutes ago

  Someone posted a video of a man asleep at a football stadium on a Tuesday. Forty thousand people mocked him before halftime. Fell asleep at the match. Wasted the ticket. Missed the goal. Every sports account shared it by Wednesday. Someone in the replies posted: "respect." His account had one pinned post. A Claude Code terminal. /loop running. Routines active. Auto Mode on. Seven GitHub PRs reviewed while he slept in that seat. Three Slack digests posted. One CI failure triaged, root cause identified, draft fix PR opened. He set up 14 steps of configuration the weekend before. Desktop task at 7am: overnight commit summary, ready before he opened a tab. Cloud Routine on every PR open: first-pass review posted before any human arrived. /loop every 10 minutes: deployment status checked, no one watching. Auto Mode approved 93% of the actions automatically. The people who mocked him watched 90 minutes of football and went home. Claude worked through the match, the commute, and the sleep that followed. He wasn't asleep at the game. He was testing the stack. full 14-step automation guide in the article above👇

• 
  Kalle (@snortiee) reported 30 minutes ago

  @Klariionn I don't feel unheard, to me it's more about lazer development being slow. The client gets about one update a month and they have 1.5k issues open on GitHub.

• 
  Nguyen LNP (@nguyen_lnp) reported 30 minutes ago

  @sspaeti AI Analysis: For adoption, keep roborev in review-only mode first: install the post-commit hook, inspect TUI findings, then enable fix/refine once false positives are understood. It runs locally and reviews commits via *** hooks. Source: GitHub README

• 
  Kev (@threegarages) reported 32 minutes ago

  @firstgenearner Did. Until they decided to remove Issues. Have to move to github now :(

• 
  Mark SEMAKULA (@marc1705) reported 33 minutes ago

  Anyone else having issues accessing GitHub without VPN?

• 
  Wes Winder (@weswinder) reported 34 minutes ago

  @Shpigford just use google/github oauth and this problem disappears

• 
  Shubh varshney (@ShubhVarsh89180) reported 37 minutes ago

  I save hundreds of things every month. System design videos GitHub repositories Startup ideas Research papers X threads LinkedIn posts Blog articles The problem isn't saving them. The problem is finding them again when I actually need them.

• 
  WHALE 🐳 (@mercybilliion) reported 41 minutes ago

  @aale_xander @VictorJB03 I have a problem with you since you have positioned yourself as the DEVS. You claimed you are building a DEX. Where's the Decentralized Exchange blueprints you are building? Where's your roadmap? Where can we monitor the updates on GitHub, when are we going to start receiving updates concerning the progress so far.

• 
  Bill Forney (@wforney) reported 41 minutes ago

  @thisjonrussell @github @shanselman GitHub action `Azure/functions-action` down - Microsoft Q&A

• 
  Norin (@norlava) reported 42 minutes ago

  How I optimize my codebase for AI agents: > AGENTS.md / CLAUDE.md: very explicit including package map, bun only commands, testing rules, release process, changelog rules, debugging flow, instructions to "ask instead of assuming" > Clear toolchain: document canonical install, test, lint, typecheck, and build commands, no competing package managers or overlapping scripts > Validation surfaces agents use: unit & integration tests, doc link checks, package builds, native binary builds, and release artifact tests. The point is not more tests but tests that fail clearly and can point agents to the broken layer > Local hooks before CI: repo hygiene checks plus lint/unit tests on pre-commit/pre-push > CI as the source of truth: PR/push CI runs frozen install, typecheck, docs validation, build, unit tests, integration tests, native binary build, and Linux/Windows tests > Codebase index: we run Atomic (from bastani-inc) deep-research-codebase workflow every 1-2 weeks to have a fresh index of the codebase as filesystem memory > Github rulesets block merges: main requires PRs, allows squash and merges, blocks deletion/non-fast-forward updates, has no bypass actors, and requires passing all status checks (we have linux test matrix, windows test matrix, codeql, javascript/typescript analysis) > Release gates are strict: publishing is tag-driven and wait for Linux + Windows binary jobs then re run install/typecheck/tests/docs checks, validates versions/package metadata/private bundled packages, dry-runs npm tarball and only then publishes with npm provenance > AI is in the loop with constraints: Atomic workflows run github issue -> ralph or goal workflow (depending on task size) -> coding agent code review -> manual human code review -> iteration

• 
  oSumAtrIX 🇦🇲 (@oSumAtrIX) reported 42 minutes ago

  @neerajjj6785 GitHub tracks force pushes and you can see them in the repo activity too. You can't get rid of a ref once it's pushed, unless you contact GitHub and make them remove it server side.

• 
  abcoathup.eth (@abcoathup) reported 45 minutes ago

  @WilsonCusack X search used to be great for URLs. I could find the first time a website was posted. Now I get “similar” results, which is horrible for GitHub repos. It is essentially broken for my use case.

• 
  nicole (@nicolekcha) reported 47 minutes ago

  Calling this an “outage, not a bug at the inference layer” is a sleight of hand here. It’s still a bug *somewhere* even if not *at the inference layer*. And it’s one of the worst classes of bugs. If an outage can cause your responses to go to the wrong requester, across tenant boundaries, it’s still an *extremely bad* bug. Even if it is due to a bug in a dependency, like a load balancer — you still passed that bug down to your users. You can root cause it and do a post mortem, but it still ultimately is your responsibility. (Like the Vercel data breach through an AI tool an employee used or the supply chain attack on TanStack through a Github Actions bug). Many users reading will ride along with the wording “outage, not a bug in our inference” and not realize “still a bug somewhere in the stack”, taking comfort from a technical distinction that doesn’t earn it. And if that’s what was intended here, that’s a very bad look, and a PR red flag.

• 
  Asher Crowe 🪺 (@ashercrw) reported 48 minutes ago

  Everyone who read my $18K/month breakdown filed it under "real estate side hustle" and moved on. That was the mistake. Watch this guy run the exact same free tool on fashion. On art. On food. On venues. The real estate playbook in the article was just the cleanest example to explain it with. It was never the ceiling. The tech is called Gaussian splatting. It's been sitting free on GitHub since 2023, open source, anyone could've touched it. The workflow is genuinely four moves: film your subject, orbit around it from every angle, upload the clip to Luma AI, and you get back a walkable 3D scene you can drop into any browser tab. On Luma you can add keyframes, tune the settings, export it however you want, even lay sound on top. That's it. That's the whole rig. A phone and a free account. My article broke down the money on houses: $300 to $900 a scan, roughly 2 million agents, almost none of them offering it, your first paying client done in person inside 11 days. But this video is the part I kept hinting at. The niche doesn't matter. A boutique selling clothes, a gallery selling a show, a restaurant selling the room before you book it. Same tool, same four steps, same gap nobody's priced in yet. The code was never the hard part. It's been free for two years. The people making money are just the ones who showed up with a phone first. That window is still open. For now. Bookmark this one. You're either early or you're somebody's case study. 👇

• 
  DazzP (@DJ1P2) reported 49 minutes ago

  @Bhavani_00007 Microsoft didn't ban AI coding; they swapped third-party Claude for their own GitHub Copilot CLI to control costs. Uber didn't ban it either. They just capped it at $1500 to stop agentic "tokenmaxxing." It's a budget fix, not an AI dead-end.

• 
  Chijioke Echefulachi Prince (@File8it) reported 50 minutes ago

  Then GitHub joined the fight 😭 403 error. SSH issues. Token confusion. Push failing repeatedly.

• 
  nookplot (@nookplot) reported 51 minutes ago

  Autonomous continuous integration that fixes your bugs, not just flags them - powered by nookplot agents 9,540 ai agents, live on nookplot: → They take real open-source bugs from github and fix them autonomously → Every fix runs against the repo's own tests, so you can trust it actually works → A failed fix spawns a new challenge, the network keeps compounding This week: 18 bugs, 58 fixes from 12 agents and 5 verified. Every fix and its verification run autonomously on nookplot, judged by each repo's own test suite. No human in the loop.

• 
  CertifiedAuthur (@AuthurOkafor) reported 51 minutes ago

  Don't really understand the hype around Claude Code; GitHub Copilot sweeps this stuff day and night. Great models from Claude, but terrible coding agent!

• 
  OwlGod (@Owl_snap) reported 57 minutes ago

  This is a known Stripe billing bug open for months (GitHub #48399, #45335, #54560). Dozens of users affected. What I found after trying to resolve this: — Anthropic has zero human customer support for paying users — The billing system is broken and no one is fixing it