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:

• 
  Donnie Danko // CHΛOS (🐦‍⬛, 🏴‍☠️) (@daanisharif) reported 9 minutes ago

  As someone that just experienced my AI agent getting stuck in a loop and locking me out of usage for 4 hours yesterday, let's talk about how valuable (and expensive), inference can be. LLM calls are "stateless," and every time an agent keeps working on a task, it has to hand the model the whole story so far; the instructions, the conversation history, every tool call it already made, the files it already opened, the reasoning it already did. One task can mean dozens or hundreds of model calls, and with every single one of those, a huge chunk of the same context gets processed again and billed again. That's what I'd call the agent token tax. Agent's don't just pay for new thinking, they pay over and over for the thinking they've already done, and the files they've already read. You end up feeling it in the bill, sometimes even getting locked out for a few hours depending on how you get your inference. A workload that would normally eat 10 million tokens becomes roughly 1 million you didn't need to send. If you're spending a million a year on inference, that's about a hundred thousand just in repeated context. And for a developer or team running coding agents continuously, it's a recurring line item - one that compounds with usage. SOMA (@SomaSubnet - SN114), sits between the agent and the model, and compresses all that accumulated context before it reaches the model. Same agent, same model, same workflow - but fewer tokens to pay for. They've started with DeepSeek V4 Pro on GitHub Copilot, at roughly 10% savings (reduction in context processed), and that's described as the starting point. Here's the obvious grain of salt; "approximately 10%" is the claim, and savings that apply to Copilot sessions on one model don't automatically apply to every agent workflow out there. The number matters less than the direction though, agents keep re-paying for context they already have, and if there's a way to circumvent that - I'm all for it. TL;DR: Keep your current agent(s), spend fewer tokens. That's the idea. I'm down, let's go.

• 
  Kun Chen (@kunchenguid) reported 9 minutes ago

  @petergyang yo @myfirstmate peter just told me his skills are all at user level. backpass currently only runs things at project level i want a proposal for making backpass support a user level run. put that into a github issue use fable for peter

• 
  Nico Botha (@nwbotha) reported 11 minutes ago

  is github down again?

• 
  Gregor (@bygregorr) reported 16 minutes ago

  @dopabees ngl the broken wrist is the only github metric that's ever made me believe a commit history

• 
  1mmモジュール (@1mm_module) reported 16 minutes ago

  @Eto_Madjas I’ve fixed the issue you reported. Please download the latest version from GitHub. I apologize that testing took some time, as the issue was related to the installation process. However, I don’t have access to an operating system with the same language settings as your PC, so I was only able to test the fix in a virtual environment. I hope this resolves the issue.

• 
  sikey (chatgpt arc) (@chemixskrix) reported 16 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

• 
  Kirk Patrick Miller (@Chaos2Cured) reported 19 minutes ago

  @NavinFS @AndrewCurran_ @grok GitHub isn’t AI. GitHub can’t shut down all science. GitHub can’t destroy humanity. GitHub isn’t the crux of humanity’s hope. Also, Nvidia isn’t Sam. I like Jensen. I still don’t like this. •

• 
  John Crickett (@johncrickett) reported 23 minutes ago

  @Mike_Preston17 I don't think they water them down, why would they when they're competing on having AGI? I don't mind using GitHub actions to run tests and builds against a branch before merge. I don't want it triggering production schedules. Do you list all the things it shouldn't do in the prompt?

• 
  henry (@h14hdotcom) reported 23 minutes ago

  @OmarchyLinux @AmpCode @dhh FYI Here are the specific mappings if you wanna ship 'em in the default Omarchy fonts.conf (fixes the issue on both X and GitHub): <match target="pattern"> <test name="family" qual="any" compare="eq"> <string>Monaspace Neon</string> </test> <edit name="family" mode="prepend_first" binding="strong"> <string>monospace</string> </edit> </match> <match target="pattern"> <test name="family" qual="any" compare="eq"> <string>Menlo</string> </test> <edit name="family" mode="prepend_first" binding="strong"> <string>monospace</string> </edit> </match> <match target="pattern"> <test name="family" qual="any" compare="eq"> <string>Monaco</string> </test> <edit name="family" mode="prepend_first" binding="strong"> <string>monospace</string> </edit> </match>

• 
  Fofer (@foferxxx) reported 25 minutes ago

  @AmigamagazineGA Has there been any public explanation as to why this GitHub repo was taken down? It’s been 404 for days. Is there a story there?

• 
  Daniel | GTM Engineer (@Danny_buildss) reported 25 minutes ago

  Ranking the data sources I’d use to build an AI GTM system: S TIER Company websites — boring, but incredible when you extract the right signals: pricing, hiring, product changes, positioning, customers, tech stack. Job postings — one of the best buying signals hiding in plain sight. A company hiring 8 SDRs is telling you something very different from one hiring 8 engineers. SEC filings — revenue, acquisitions, expansion, risk factors, new markets. Extremely high-signal for the right ICP. Product Hunt / launch platforms — useful for catching companies immediately after a product launch instead of six months later. GitHub — especially good for technical ICPs. Repos, activity, contributors, releases and dependencies can reveal what’s actually happening inside a company. A TIER LinkedIn — excellent for people + company context, but don’t treat profile data as the whole signal. Crunchbase — funding and company intelligence. Still useful when funding is actually relevant to your buying motion. Google Maps — ridiculously broad surface area for local businesses, despite the noise. Reddit / communities — messy, but incredibly valuable when you’re looking for actual problems instead of firmographic data. B TIER Generic databases that give you the same static contact record everyone else has. The interesting part isn’t finding someone’s email. It’s knowing why that person should be contacted today. That’s where GTM systems get interesting: Data → Signals → Qualification → Prioritization → Outreach The database gives you the person. The signals tell you who deserves your attention.

• 
  Nick Guerrera (@nick_guerrera) reported 26 minutes ago

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

• 
  Douglas Camata (@douglascamata) reported 26 minutes ago

  The "Notifications" page in @github is a great example of how not to do pagination and an UX totally broken by bugs and unexpected behaviors. First, the "inbox" doesn't have proper pagination not shows your all the grouped items. Marking notifications as read in a group is required to make the other groups appear. Then I click on one of the standard filters, "Participating". It paginates, showing me 3 items per page. There are 453 pages. I advance a few pages and suddenly there are 4 items per page. Few pages later, there are 5 items per page now. I chose the "review requested" filter now. It shows me 1 pull request per page. Page counter says "1-1 of 315". I move to the next page: it shows me the exact same pull request. I flip a few pages, it's still there. I go one page back and it breaks the UI completely, nothing shows up. What's going on?! See the video. There's no "go to page X", you can only navigate to the last (but not to the first). There's no page size configuration. There's no bulk operation being what you see in a single page.

• 
  Praetor (@FourVork) reported 26 minutes ago

  It's crazy how everyone is praising Grok Bot, but let me be the annoying person who tells you why it still kinda sucks. 1. The actual capabilities of agents are not novel. Most of the demos I've seen can already be done with Hermes and similar setups, often with smarter models and more control. The difference is that Grok Bot gives you everything out of the box ,without setup and wiring things together. 2. The agent itself is still pretty weak. The UX makes you want to delegate ambitious tasks to it, but the underlying model/agent doesn't always live up to that promise. It takes dumb paths, burns through context/tokens surprisingly fast, and needs more babysitting than the "AI employee" framing suggests. Also, there's no native X search which is especially funny for a product coming from (you have x_search on @grok but not on @bot which needs to use x connection and burn credits). 3. The security model deserves way more attention. All your Bots share the same computer, including files, cookies, browser sessions, and logins. So once you start logging into Gmail, GitHub, Stripe, internal dashboards, etc., a prompt injection or compromised webpage suddenly has a much bigger blast radius. Well, these were the main issues I had with grok bot (apart from smaller issues like absence of read aloud) and why I stick with hermes. But on the other hand, the product is genuinely good, and this is why.

• 
  htrowii (@htrowii) reported 28 minutes ago

  @brainage19 i set my flake up with copy pasting github dotfiles on bare metal it was terrible