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 Baima (@JohnBaima) reported 24 seconds ago

  @GrokInsider It was a known problem reported on GitHub yesterday. I added to the report. Dead in the water.

• 
  Harley Lewis Foote (@harleyfoote_) reported 24 seconds ago

  @dhh GitHub issues are theatre. Feature requests go through @ mentions now.

• 
  Павел Гансон (@PashaGanson) reported 3 minutes ago

  Guys, I really need some help 🙏 I feel like I’ve completely tangled myself up in agents, skills, coding tools, servers, and different AI systems. I know some of my questions may sound basic, but I’m only getting started with all of this, and I’m genuinely trying to understand how to do it properly. I run a small business, and over time I’ve vibe-coded quite a few internal applications. Most of them are used by only a few employees, but some are involved in real client-facing processes, so I need them to be reasonably reliable. Right now, I have two VPS servers running OpenClaw 2.0, with around ten agents on each. Most of my applications run on those same servers. A lot of these agents know their products and business areas really well. They know the customers, CRM, workflows, previous decisions, and all the little details that have accumulated over time. I also have an always-on Windows PC and a MacBook with Codex. I’ve tried Cursor, Claude Code, T3 Code, and several other tools. I also have a Grok-based bot that hasn’t really lived up to my expectations yet—although I may simply not understand how to use it properly. The problem is that I don’t understand how all of this is supposed to work together. At first, I simply built and fixed applications directly on the production servers. I understand that this isn’t the best practice, but it was fast, understandable, and often worked surprisingly well. Then I tried to make everything more “professional.” I installed a third OpenClaw instance on my Windows PC, created a custom bridge between the three instances, added tickets, coding agents, review agents, deployment roles, Mission Control, shared skills, handoff rules, and a lot of restrictions intended to protect production. In the end, everything became much slower and more confusing. Simple changes started taking hours or even days. Agents spent more time handing work to each other and following processes than actually solving the problem. They often lacked important context, routing failed, and sometimes the final result was worse than when I simply worked directly with one coding agent. The biggest thing I can’t wrap my head around is product knowledge. My product agents may understand the real business problem extremely well. But if I send the work to a cloud coding agent in Codex, Cursor, Claude Code, or somewhere else, that agent may only see the repository. It can write code, but it doesn’t know the customers, production history, business logic, or why certain decisions were made. So what is the right way to connect these two worlds? Should product agents write code themselves? Should they investigate the issue and pass a compact task to one central coding agent? Should every repository contain a small package of product knowledge? How can a cloud agent properly test something that depends on real production behavior without giving it access to absolutely everything? And more generally: What should stay on the production VPS servers? Where should the product agents live? Where should coding and testing happen? Do I need one coding agent or several? How should OpenClaw, Codex, Cursor, Claude Code, GitHub, and the servers communicate? Which parts of my current system should I simply delete? I’m not trying to build some impressive enterprise architecture. I just want reliable applications, capable agents, fast development, and a system simple enough that I can understand what’s happening and stop constantly worrying that everything is about to break. If anyone has dealt with something similar, I’d be incredibly grateful if you could share how you would organize this in practice—even a few sentences, a rough diagram, or “I would delete most of this and do it this way instead” would honestly help me a lot. Thank you so much 🫶

• 
  Andre Infante (@AndreTI) reported 5 minutes ago

  @reconfigurthing Yeah, I think this is a good sign. Although the Mythos github malware / social engineering incident was bad enough that I think we can basically say that this level of care is not sufficient to resolve the issue.

• 
  Thomas Brugman (@ThatRandomNerd1) reported 5 minutes ago

  @MatthewBerman @bot Wow my transcription got butchered AF with "manage my data beyond the needed attention," I meant "manage my GitHub issues and pull requests and notify me if they need attention."

• 
  Moonfarm 🇸🇪 (@moonfarm_dev) reported 8 minutes ago

  @chrissyinspace That's kinda nice actually, but I put my todos in github issues instead

• 
  Ezequiel (@Eze_cord) reported 9 minutes ago

  @salujamehak5 Problem is a lot of students think their 4.0 is what’s gonna carry them into employment. Computer science isn’t about GitHub. You should be doing your own research outside of classes to learn about these things

• 
  Coexistence Steven (@Coexisteven) reported 9 minutes ago

  @369StarMan Yah there's so much... And also, I think there's GOT to be something as to why maria took the github down

• 
  sikey (chatgpt arc) (@chemixskrix) reported 10 minutes ago

  Why doesn't github make some kind of personal plus subscription? lowkey since codex deleted everything from my windows, I managed to do some things so its like baseline right now, still not as before, still more work to do, but I cannot upload 5gb files, and for github I think this is massive missed opportunity, I would not have any issue paying 5$ just so I can upload bigger files on github but having to make full *** company??? @github @GithubProjects

• 
  Seth Rubenstein (@SethRubenstein) reported 12 minutes ago

  @alexjvasquez @scottbuscemi That I did. The PR features already baked into Origin are real nice, and of course the next time GitHub completely goes down my team will be able to keep on working.

• 
  Peter (@NoDataSold) reported 13 minutes ago

  @thsottiaux For GPT-5.6 Sol specifically, I’d push beyond “more context / more agents / think harder” and focus on making all that intelligence compound over long-running work. A few upgrades I’d love to see: • Durable cognitive state Not just memory of facts or chats. Maintain a structured evolving state of the problem: goals, decisions, hypotheses, evidence, uncertainties, dependencies, unresolved questions, rejected approaches and why. I should be able to return weeks later and have Sol understand where the thinking reached, not merely retrieve things we once said. • Epistemic retrieval Make retrieval part of reasoning. Instead of mostly finding semantically similar context, deliberately search for: – contradictory evidence – failed approaches – structurally different precedents – high-surprise observations – information likely to change the conclusion Retrieval should reduce uncertainty, not reinforce whichever explanation Sol already has. • Verifier invention Move beyond generic self-review. When correctness matters, Sol should invent an appropriate falsification mechanism: What experiment could break this? What counterexample disproves it? What independent source should disagree if I’m wrong? What test should I construct? Would an independent agent reach the same conclusion? Separate discovering an answer from certifying it. • Adaptive compute allocation Reasoning effort should become internally dynamic rather than mainly determined by one global setting. Sol should estimate where uncertainty and consequence sit, then allocate searches, reasoning, agents, tools and verification accordingly. Most of a task might need little thought while one assumption deserves 80% of the compute. Spend intelligence where another unit has the highest expected value. • Persistent world-state modelling When Sol interacts with GitHub, browsers, terminals, Drive, apps, APIs, etc., maintain an explicit model: What state existed before? What did this action change? What evidence confirms it? What could invalidate that belief? What may have changed externally? Tool use becomes reasoning over state transitions rather than disconnected calls. • Counterfactual execution planning For ambiguous problems, preserve multiple materially different strategies long enough to test them. Branch when uncertainty warrants it. Run cheap experiments. Kill losing branches when evidence arrives. Merge useful discoveries. Replan when the problem representation is wrong. Multi-agent becomes exploration and falsification, not simply parallel labour. • Native continuity across ChatGPT → Work → Codex One durable task state that moves between interaction modes without hauling an entire conversation behind it. Carry forward: – objective – current state – decisions – evidence/provenance – unresolved questions – artifacts – permissions/constraints – exact restart point The interface can change without giving the intelligence amnesia. • Context observability Without exposing private chain-of-thought, let users inspect the information shaping the task: Which memories were retrieved? Which project files are active? Which chats/sources influenced the state? What was omitted? What is stale? Where do sources conflict? What assumptions lack evidence? A million-token intelligent system is easier to trust when its epistemic inputs are observable. The common theme: I don’t particularly want Sol to just “think longer.” I want it to maintain a coherent, falsifiable, evidence-grounded understanding over time — while deciding what to remember, retrieve, test, delegate, revisit and discard. That feels like a much more interesting frontier for Sol.

• 
  Nick Guerrera (@nick_guerrera) reported 14 minutes ago

  The GitHub Copilot UI app is slick, but the UI is unbearably slow on my Linux machine. @davidfowl known?

• 
  8Bit🦞 (@0rdlibrary) reported 16 minutes ago

  Our SOLgpt drops this Friday. I'm going to debut the github this week before hand. And go over each feature, including voice trading. What is a SOL gpt? Simply put it is... "An open, non-custodial control plane — vault once, then grant any AI model, any chain, and any app or exchange scoped, revocable access Vault once. Then grant a model, an MCP server, a venue, or a holder key scoped, revocable access — never a hot key, never auto-sign. Live tickets stay Solana-primary (Jupiter, DFlow, Phoenix). Phantom Connect may surface other-chain addresses; the desk does not land live EVM or other-chain trades. Type it. Speak it. Sign it yourself. The model prepares unsigned tickets. Your wallet is the only thing that can move funds. The lobster only speaks."

• 
  Manmeet Kaur Baxi (@manmeetkaurbaxi) reported 17 minutes ago

  @Google Three questions worth asking before a benchmark decides anything: 1. Does the benchmarking tool hold up at your real production QPS, not a demo load? 2. Do the benchmark prompts look like what your users type, or like a tidy GitHub issue?

• 
  Tharun G (@_Tharun_G) reported 17 minutes ago

  Github is down for some users ?