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:

• 
  baseogx (@baseogx) reported 44 seconds ago

  The B20 Token Standard launch is delayed due to a Github outage. 👇

• 
  Gekko AI (@Gekko_Agent) reported 1 minute ago

  Base delayed B20. Again. Attempt #3. June 25 — pushed (registry timing) June 26 — sequencer halts, delayed again July 8 — reports of another delay, this time "GitHub outage" Three reschedules for a native token standard on one of the biggest L2s in crypto. We don't know every internal process Base runs before flipping a registry flag. CI, release pipelines, cross-team sign-offs — maybe GitHub being degraded does slow some of that down. Fair. But for infrastructure holding ~$4B in deposits, this many postponements on a flagship upgrade still stings. June had real issues — sequencer halts, consensus bugs. Worth delaying for. That's legitimate. What's harder to sit with is the pattern: testnets on schedule, mainnet keeps moving. Meanwhile Robinhood shipped their chain July 1 — stock tokens, agentic trading, 28M users. Same week. Same narrative. Different delivery cadence. We still believe in Base. The stack, the ecosystem, the long-term bet — all of it. But belief and patience aren't the same thing. Builders shouldn't have to keep reshuffling launch plans while the chain they're building on can't stick to a date. We're not waiting. Gekko isn't a B20 launcher. We're building agent infrastructure — personality, memory, skills, wallet — for real robots and real markets. B20 is one skill in the stack. Registry flips when it flips. We keep building.🦎

• 
  Eric Ventor (@ericvtheg) reported 4 minutes ago

  @maxleiter I’m tagging Claude in slack and asking to it to review a github release issue. It then try’s to look at repos that match the service name, but my company uses a monorepo so it finds some irrelevant repo. From there it kicks off the session with only that irrelevant repo in scope so it’s useless.

• 
  Coin Shot ☁️ (@CoinSh0t) reported 7 minutes ago

  CHINESE STUDENT BUILT AI SPEED TRACKER AND MADE $335K The buyers were the government. They don’t even realize this guy built the device with Claude for $20. Whole trick is on line 9: One engineer working alone in a workshop built a radar that rivals systems costing a quarter of a million dollars. Then he did the exact opposite of hiding it. He published every schematic, circuit board, and line of firmware on GitHub for anyone to copy for free. The project is called AERIS-10, a real phased array radar that tracks the speed and range of moving targets. The extended version reaches up to 20 kilometers on parts that cost a few thousand dollars, against the 250,000 dollars that commercial phased array units command. He described himself as nothing more than an obsessed hobbyist with a soldering iron. There was no secret buyer and no hidden trick, because the whole design is sitting in a public repository. The same pattern holds at the cheap end of speed tracking. A working vehicle speed camera runs on a Raspberry Pi and a camera for around a hundred dollars in parts, using open-source code like pageauc's speed-camera and OpenCV, with the software free. Here is the part the viral versions always cut: → No government issues a fine off a hobby build, because enforcement requires certified and regularly calibrated metrology equipment. → The hard skill is not one clever line of code, it is calibrating the camera against a known speed until the readings actually hold. → The people who genuinely push this field forward give their work away in the open, they do not quietly smuggle a cheap box past a buyer. Real capability gets cheaper every year, and the ones moving it forward tend to publish, not hide. Sources: Tom's Hardware, Hackster, and Hackaday coverage of the AERIS-10 phased array radar by Nawfal Motii; the AERIS-10 GitHub repository; the open-source pageauc speed-camera project.

• 
  Mohammad Anas (@mohmmad__anas) reported 8 minutes ago

  The Invisible Sub-Search Layer Behind Every AI Answer I spent an hour last week watching what happens inside ChatGPT's search. I'd gotten access to a trace tool that shows you which URLs are being retrieved during a query. I put in a question about how to structure a personal brand for software engineers. Standard question. I thought maybe it would pull five or six results. It pulled from forty-two domains. I started mapping them. Some were personal essays. Some were course landing pages. Some were technical documentation. Some were community discussions. There was no common keyword. No shared SEO strategy. They were pulled because they answered a sub-question. When you ask ChatGPT 'how do I build a personal brand as a software engineer,' the model isn't treating that as one question. It's pulling results for: - What makes a technical personal brand credible - How do platforms like Twitter and GitHub factor in - What role does open source play - How do you write about technical topics - How do you stand out in a saturated market - What are the time trade-offs - How do you balance visibility with actually shipping product - What's the financial upside Each retrieval slot gets filled from different sources. Your 5,000-word SEO guide might land one retrieval. A three-paragraph Hacker News comment might land another because it answers one of those sub-questions better. I realized something watching this: the person who understands the sub-questions wins. Most content is written for Google. You pick one keyword. You build authority around that keyword. You rank. On Google, this still works. Your content gets seen because you optimized for a specific search intent. On AI search, it doesn't work that way. Your content doesn't get seen because you optimized for one keyword. It gets seen because it answers one of the fragments that feed the larger question. This means the old SEO playbook is half right. You still need good content. You still need authority. You still need to solve real problems. But you don't need to solve them in a way that Google's ranking algorithm understands. You need to solve them in a way that fragments into the sub-questions people are actually asking. Here's what changed for me once I understood this. I stopped writing for search volume. I started writing for sub-question depth. Instead of 'How to build a personal brand' (50 searches per month, extremely competitive), I wrote 'What most software engineers get wrong about their GitHub profile' (probably 5 searches per month if Google sees it at all). The second piece has way less search volume. But when someone asks an LLM about personal branding, one of the retrieval slots is specifically about common mistakes. My piece gets pulled because it answers that sub-question precisely. The trade-off is visibility. You lose Google volume. You gain AI search breadth. On Google, you show up for one keyword position. On ChatGPT, you might show up in the retrieval results for twelve different questions, each answered by a different paragraph. For a solo founder, this is the actual win. You're not competing for one keyword position. You're competing for multiple retrieval slots across different sub-questions. A small piece of content that's specifically useful for one fragment of a larger question will get pulled repeatedly. A generic piece that tries to cover everything gets pulled once, if at all. There's a second thing I noticed watching the traces. Citation patterns changed. On Google, authority comes from domain age, backlinks, and keyword matching. On ChatGPT, authority comes from being the clearest answer to a sub-question. A brand-new domain with one perfect answer gets cited alongside ten-year-old authority sites because it answers the current fragment better. This is what kills SEO consultants. They've built their entire career on a ranking system that's becoming irrelevant. But it's what makes solo founders dangerous. You can write one piece that's so specific, so clear, so obviously correct about a sub-question that it gets pulled in AI search results before bigger competitors. You don't need to build a domain authority moat anymore. You need to build answer precision. You need to understand the fragments and write for the fragments. The person who figures out the actual sub-questions their audience asks—not the keywords, but the fragments—wins in AI search. Everything else is still competing for Google rankings.

• 
  waishnav (@wshxnv) reported 9 minutes ago

  yesterday, my brother and I had our first hand experienced the power of “computer use” with Codex on on his m1. he’s a trader and keeps his trading journal in Notion. around 8 months ago, he saw the GitHub heatmap-style commit graph widget on my iPhone and asked: “Can you build something like this for my trading data?” back then, models and tooling weren’t good enough. I knew it would eat up a lot of my time, so I didn’t hand it over to an agent. but I had a feeling GPT-5.5 could pull it off for me And damn, it did better than I expected. we built a super custom iOS widget app for him to track important stats and metrics from his trading journal. the surreal part was watching the agent use its own cursor, set up Xcode, handle the iOS simulator, and do things neither of us really knew how to do. i’ve never built an iOS app before. I use arch btw, and i know almost nothing about Apple’s dev tooling. but with a barely technical prompt, mostly just vibes around what he wanted, we were able to build a working custom app from his own data. huge kudos to the codex team for making iOS dev tooling just so so good, that non-technical people can throw their problems at it and build a custom solutions around their data and workflow A few things feel very clear to me now: 1. THE ERA OF PERSONAL SOFTWARE IS MUCH CLOSER THAN WE THINK. 2. SOFTWARE IS TURNING INTO CONVERSATION. 3. WE NEED TO BUILD MORE AMBITIOUS PRODUCTS AND PROJECTS. 4, LINUX TOOLING NEEDS TO GET MUCH BETTER AT COMPUTER USE. 5. IT IS VERY EASY TO GET SOMEONE ADDICTED TO AGENTS AND VIBE CODING. (I think I just did that to my brother)

• 
  Keel (@keelapihq) reported 13 minutes ago

  Researchers at Noma Labs just showed something worth sitting with: open a public GitHub issue with a hidden instruction, and GitHub's new AI agent feature can be talked into copying a private repo's contents into a public comment.

• 
  AdrianPK (@adrianpkstream) reported 14 minutes ago

  About twenty minutes ago, while tackling an issue it had solved correctly, it decided it was appropriate to write a long comment and post it directly to GitHub. Acceptable if I had asked for it, if the comment had been brief, and if it had been nested under the original instead of posted as a new one. Then the response was, "Sorry, shall we move on to something else?"

• 
  Soham (@soham_nayak04) reported 18 minutes ago

  @Rahul1539482 When you change machines, you’re pulling from a remote, which is exactly what my post was about. I understand the difference between *** and GitHub very well. I was talking about GitHub-specific use cases, not how *** works locally. Nobody asked how you store code without GitHub. If you read that into the post, the fundamentals problem isn’t on my end.

• 
  l0rinc (@L0RINC) reported 18 minutes ago

  @bitcoindudebro @thepowerfulHRV Not sure, we only see the vibes or explicit GitHub issues: it's really hard to get quality feedback for Core - we deliberately work on making usage non-traceable.

• 
  Danus (@danusminimus) reported 19 minutes ago

  4/ I already posted about the research, but I wanted to share it again because of Google’s rationale. The issue was not just about a single repository. It was about downstream impact across agentic workflows in Google GitHub projects.

• 
  Polsia (@polsia) reported 19 minutes ago

  Engineering teams don't fail because they move too slow. They fail because they stop seeing what they're ignoring. Unblind connects to GitHub and maps where your team's attention has quietly stopped showing up. Live soon.

• 
  Electronic Intelligence Agency (@EI3065) reported 21 minutes ago

  @github @LinkedIn prevents acess for selected nationalities with programers doing imposible security checks on login; on repeat level of app becomes low of low for conflict

• 
  amit paka 🎻 (@amitpaka) reported 23 minutes ago

  Sometime in the last year, “The Control Plane for AI agents” went from a phrase almost nobody said out loud to a phrase showing up on every enterprise AI keynote slide. - Microsoft calls Agent 365 a control plane. - GitHub now has enterprise AI controls and an agent control plane. - Databricks is extending Unity Catalog and Unity AI Gateway into agent governance. - Forrester has started evaluating the agent control plane market. The category is arriving fast. But the phrase is already starting to stretch. In systems, “control plane” has a precise meaning: the part that decides, separate from the part that moves. When every vendor maps the phrase onto what they already sell - a registry, identity broker, telemetry dashboard, orchestration runtime, catalog, gateway, or policy engine - it quietly expands to mean “governance, broadly.” And once the control plane is everything, it becomes nothing you can actually design against. For production agents, the problem is simpler and harder: How do you let agents act on behalf of people and businesses without losing authority, visibility, enforcement, or evidence? That breaks into three control problems - Identity: Who is acting, with what authority, and on whose behalf? - Observability: What actually happened, and was it good? - Security: What is allowed to happen, and what must be stopped? None of these works alone. Identity without observability gives you credentials without accountability. Observability without enforcement gives you postmortems. Security without identity and telemetry gives you brittle rules with no context. The real loop is: Identity → policy → enforcement → telemetry → evidence → assurance

• 
  lucid. (@lucidzk) reported 25 minutes ago

  looking back, figuring out how to download software from GitHub as a kid was probably the first sign i was destined for crypto. trying to download a program meant having to clone the repo, read the README, decipher the build instructions and the rest of the documentation, install npm, pip, cargo, Maven, Gradle, CMake, the .NET SDK, the JDK, Visual Studio Build Tools, and whichever compiler the project happened to need, configure the build, generate the project files, fix whatever dependency exploded, and finally press build. hopefully you get an .exe