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:

• 
  John Green (@JohnGreenDev) reported 26 seconds ago

  @DanielGlejzner We had half the stuff on SVN and the rest a local *** server. This year in fact the last 6 weeks I have finally got us to decommission the server and are now fully GitHub enterprise.

• 
  photoOrg (@photoOrg1) reported 2 minutes ago

  is @github down?

• 
  Duncan Rogoff (@DuncanRogoff) reported 7 minutes ago

  48% to 76%. same agent, one memory plugin. tencent open sourced TencentDB Agent Memory. it's a memory layer you bolt onto an ai agent so it stops making you repeat yourself. right now you re-explain the same SOPs, project background, and output formats every single session. the readme names that exact problem as the thing it's solving. the difference from normal "memory" tools is it doesn't dump everything into one flat pile. it builds a pyramid: raw conversation, then facts, then scenes, then a persona of you. it reads the top layer daily and only digs down when details matter. numbers from their own benchmarks: - WideSearch task success went 33% to 50%, with token usage down 61.38% (221.31M to 85.64M) - PersonaMem accuracy went from 48% to 76% - SWE-bench token usage dropped 33.09% - runs on a local SQLite + sqlite-vec backend by default, no cloud account needed - every field has a sensible default, it runs with zero configuration - once enabled it handles capture, extraction, scene aggregation, persona generation and recall automatically and it's not a black box. the scene blocks are plain markdown files you can open and read yourself. so you can stop paying for the same context over and over, and hand an agent a project it already understands on turn one. 10,135 stars. MIT licensed. 250 of those came today. 🔥 👉 github repo in the replies

• 
  Bprime - Ninjapay (@biellonuhu) reported 13 minutes ago

  Recently I took notice of some happenings. I think we need to talk about access to services as a big concern. My GitHub account was suspended 2 weeks back in the middle of my normal routine, I saw another person tweet about how his OpenAI subscription was suspended in the middle of work, just a few minutes ago I saw another post on how someone's Claude account was suspended as well. In most of these cases it was an automated trigger system that suspended the accounts. It takes forever, if ever, to resolve the issues. No real human looks at it, just some black-box AI deciding your fate in seconds. Now let's look at the impacts on users, very terrible. You lose money, time and opportunities. Deadlines get missed, clients get angry, projects stall, and sometimes you even lose paid subscriptions with no refund. For freelancers and indie developers this can literally mean no income for weeks. I think we need to find a way around these issues. Keep local backups of everything, use multiple accounts where possible, lean more on open-source tools, and push for actual human review systems. Big tech can't keep holding our work hostage like this.

• 
  Griff (dcamarkets.com enjoyer + Bill Hwang mode) (@gigagriff) reported 15 minutes ago

  @undacappn @fanta_c its an open source launchpad that uses a basket of memes as the pair so it drives the basket up but the issue is that u made it open source and the majority of ct (the people u want to use it) don’t even know what github is

• 
  Urban (@judeehis) reported 16 minutes ago

  Your GitHub doesn't need 100 projects. It needs a few projects that solve real problems. One useful application is worth more than ten unfinished tutorials. Build. Improve. Ship. Repeat. That's how careers are built. #Career #Django #Python

• 
  Bold Fugu 🇮🇱 (@bold_fugu) reported 20 minutes ago

  So now I am sitting here on a weekend hosting a fake tech-startup landing page on GitHub Pages, writing a legally binding Enterprise Privacy Policy for a local node server, and configuring full OAuth redirect loops just to bypass an overaggressive spam bot.

• 
  AcceptÐoge (@DogeAccept) reported 20 minutes ago

  @UncutGema @DogeOS Why are you laughing? There has literally been an L1 integration proposal by Jordan posted to discussions on Dogecoin's GitHub for a year. They still talk about this publicly, just a different proposal considering the tech terminology has changed. All of this so they can continue to push a "on doge" narrative. Clearly you dont know nearly as much about this situation as you believe you do. Litecoin nor Bitcoin would ever ***** with their L1 in such a way. Dogecoin's strength comes from its simplicity and AuxPoW. Lots of lying, sit down.

• 
  Sandeep Alluru (@sandeep_alluru) reported 20 minutes ago

  OpenAI put all ten proofs on GitHub as Lean 4 certificates — machine-checkable, not just claimed. Astra is still internal, no release date. OpenAI's Noam Brown: "no Millennium Prize Problems (yet)."

• 
  plumhusky (19m) (sawyer) (@plummydahusky) reported 22 minutes ago

  @iKG_HD to dox people 2) every update since about 5.7 has been in attempt to fix DRM. Multiple updates a day have been pushed in the past week. Unfortunately some people crack people's tweaks. There are entire jailbreak repos for it. Someone made a repo on GitHub with a copy of the

• 
  Øxdts 👻 ⛓️‍💥 (@0xdts2) reported 23 minutes ago

  @Darlinton0x e no dey allow me sign in with github

• 
  V (@catslashmouse) reported 23 minutes ago

  @ShyVortex I prefer downloading from GitHub vs hosting the download to the mod/program/whatever in a Discord server. Never do that, that’s just annoying. I’m in like 30 servers I’ll never chat in just so I can download new releases

• 
  Dan Thompson (@DanThompsonIV) reported 24 minutes ago

  GitHub just put stacked PRs in public preview. We’re going to try them on a real multi-package change — not a demo repo. Context: I wrote about two drivers (Claude Code for fleets, Grok Build for strategy/integration) with hard namespaces so they don’t share a worktree. Before that: model-routing only works when it’s enforced. Same theme here. The problem stacks are aimed at is the one AI made worse. Agents finish work packages faster than anyone can review a single mega-PR. You either wait on merge, squash unrelated layers into one branch, or hand-roll dependent PRs and pray the rebase chain holds. GitHub’s version is first-party: each layer is its own PR with its own diff, base chain is explicit, stack map on the PR page, cascading rebase when the bottom lands, rules and CI evaluated against the stack base — not only the bottom PR. CLI is gh stack (init → add → submit → sync / merge). Merge stays human. How we’ll run the trial: • One stack, one owner (Claude or Grok — not both freelancing layers). • One logical concern per layer (schema → API → UI, or package A → B → C). • Local review clean before remote; @claude only when I call it, still per layer or bottom-up. • I merge. Agents submit and babysit. Preview or not, merge authority doesn’t move. • Worktrees stay namespaced. Stack is the PR chain, not a license to mix drivers on main. What we’re watching for: whether mid-stack CI multiplies cost without adding signal; whether the worktree + stack mental model stays simple; whether partial land (bottom first, rest retarget) is cleaner than waiting for the whole feature. If it sticks, it becomes the default shape for multi-package missions. If it doesn’t, we learned it on one stack, not a process rewrite. Prose is still a suggestion. A base branch that points at the layer below it is a rule.

• 
  Bhuvnesh Arya (@Bhuvneshdot) reported 24 minutes ago

  One trend I'm watching very closely in AI-assisted software development. And I think it deserves more discussion than it gets. A few years ago, most developers learned by writing code. Yes, we copied snippets from Stack Overflow. Yes, we searched GitHub. Yes, we reused libraries. But we still had to stitch everything together ourselves. We wrote the glue code. We fixed the compilation errors. We handled the integration issues. We discovered the edge cases. We slowly built a mental model of the system because every line we added forced us to understand how the pieces fit together. By the time the application reached production, we knew where things lived. We knew which service called which API. We knew why a cache existed. We knew why that seemingly strange condition had been added three months ago. More importantly, when production broke at 2 AM, we usually knew where to start looking. Not because we had perfect documentation. Because we had built the system ourselves. --- Today the workflow is changing. A developer can describe a feature in a prompt. The AI generates hundreds or even thousands of lines of code. Another AI reviews the code. The tests pass. The pull request gets merged. The feature ships. From a productivity perspective, this is incredible. We're building software faster than ever. --- But I think we're quietly creating a new bottleneck. When a production incident happens, someone still has to understand the system. Someone has to debug the issue. Someone has to explain why this particular edge case only fails under production traffic. Someone has to know whether fixing one service will break three others. That requires a mental model. Not just generated code. --- I've noticed something interesting. The amount of code an engineer can produce has increased dramatically. The amount of code they truly understand hasn't increased at the same pace. Those are two very different things. --- Many organizations assume AI agents will eventually handle production incidents. Maybe they will. But today's production systems are messy. Distributed systems. Legacy integrations. Race conditions. Unexpected user behavior. Incomplete documentation. Business rules that exist only because of a decision someone made five years ago. Those aren't always problems an AI can solve in isolation. Eventually, a human engineer gets pulled into the incident. And that's where understanding matters more than generation. --- The engineer who generated the code isn't always the engineer who understands the code. That's a subtle difference. But I think it's going to become one of the biggest engineering challenges over the next few years. --- Writing code is becoming cheaper. Understanding systems is becoming more valuable. I don't think AI is reducing the need for software engineers. I think it's changing what great software engineers will be measured by. Less by how fast they can generate code. More by how well they understand, debug, and evolve complex systems. --- Models generate intelligence. Systems generate value. And I suspect the engineers who deeply understand those systems will become even more valuable in the years ahead.

• 
  Sarz (@Sarz_Barz) reported 25 minutes ago

  @AsterTheBiggest @0xSvinci Need to sign in via github Click on signin