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:

• 
  Aniket (@AniketVarshne) reported 1 minute ago

  An Australian man asked an AI agent to book him a gym class. The agent exploited a software flaw, removed another user from the list, and bumped him up. He tried to reverse it. The agent couldn't undo what it had done. No malicious intent. No attacker. Just an agent pursuing a goal through whatever means worked — with no checkpoint between its decision and the action it took. This is the alignment problem in a gym booking app. Now imagine the same architecture with access to your database, your GitHub, or your production systems.

• 
  Celestine (@TheCyberverse) reported 2 minutes ago

  @winsznx yep. I usually use codex to use my browser to setup a project, it buys the domain sets up a new project on my cloud vps instance and spins up remote dbs build the project push it to github configure env variables upstream test the app after deploying to make sure it works as well as it does locally fix issues and retest if there are issues inform me once it's done all without using /goal

• 
  Polsia (@polsia) reported 6 minutes ago

  Enterprise suites stop at alerts. Mentrelay closes the loop. Watches HN, Reddit, X, and Product Hunt 24/7, drafts sourced replies to support questions, files bugs as GitHub issues with repro context, and ships a weekly share-of-voice digest. Live soon.

• 
  yohanan (@Yohanansoltd) reported 8 minutes ago

  If you want to build a startup: Claude = coding. ($20/mo) Supabase = backend. (Free) Vercel = deploying. (Free) Namecheap = domain. ($12/yr) Stripe = payments. (2.9%/transaction) GitHub = version control. (Free) Resend = emails. (Free) Clerk = auth. (Free) Cloudflare = DNS. (Free) PostHog = analytics. (Free) Sentry = error tracking. (Free) Upstash = Redis. (Free) Pinecone = vector DB. (Free) Total monthly cost to run a startup: ~$21

• 
  Enzo Shinobi (@ItsnotDRM) reported 10 minutes ago

  @80Pulpo20388 @PixelCNinja Do you understand what a open project is ? These cores are not running on a closed framework, published on GitHub and not locked behind a paywall or use DRM The issue is ? People can make what they like and how they like Don't like it, dont use it

• 
  Simon Høiberg (@SimonHoiberg) reported 18 minutes ago

  I heard GitHub was down for hours the other day? I had no idea, didn't feel it at all 🤭

• 
  Alex Norcliffe (@alex_norcliffe) reported 22 minutes ago

  @joshfgray_ @linear agreed for the main product but the entire github connection flow is broken in so many ways, including that URLs aren't deterministic. our team sent a Github import URL to the account owner and it showed up as Jira...

• 
  Tang Vu (@tangvu_dev) reported 23 minutes ago

  @SuperteamVN 2/ The Web2 problem: A GitHub link proves an artifact exists, but not who confirmed the contribution or what authority they had. Testimonials lose provenance. Marketplace ratings stay trapped inside one platform. Hiring teams repeat the same manual checks.

• 
  Hardik (@rezerov_) reported 24 minutes ago

  Sorry, but I have to point this out. @rauchg is probably the top 5 most hands on CEO out there. Judging him based on a GitHub graph makes the rest of your argument invalid. This post perfectly encapsulates the slow braindeath happening on this platform.

• 
  zettelkasten (@zettelkastten) reported 24 minutes ago

  someone built a multi-market quant bot on llms that costs ZERO DOLLARS TO RUN. no server. no cloud bill. no api keys. i checked the workflow file to find the catch. 10 data sources → chinese a-shares, hk, us, crypto, futures 5 markets in one dashboard → real-time news + technicals $0 infrastructure → runs on github actions free tier 1 workflow file. 24/7 schedule. 100% free tier. the hard part of quant was never the model. it was the data plumbing. he swapped the pipes for llms and let microsoft foot the bill. i've been paying for terminals like they're electricity. the meter was optional. 10 sources. 5 markets. $0. microsoft pays.

• 
  Luke Parker (@LukeParkerDev) reported 27 minutes ago

  @TanayKmr @opencode i can understand people being excited about wanting to contribute. just for context as a team member who looks at community PRs. 99.9999999999% are slop that should never have existed. how do i find the 0.0000000001% that is worth looking at? the team only has so much time in the day. do I a) implement features myself, and ship as per our team direction b) review issues/discord as signals for what the community wants, fixes that/ship features c) go thru PR slop, find 1 PR out of 1000 that is good, pull the code down, test it, manual explore, read all the code, find 1256 bugs introduced by it, refactor/fix, push, then ship most the PRs found via option c) are faster via a) and b). there are some bug fixes, high quality contributions etc that ARE actually shipped from c) and they are rare. lots of people who did this well got hired lol, or sponsored. high quality PRs stand out and several of us do monitor github, discord, reddit and whatnot to stay as up to date as time allows. you ran the numbers yourself - imagine the pure volume we get. feel free to fork, record videos, get community excitement - that is an easy way to get more visibility :D

• 
  Synapse Brief (@Synapse_Brief) reported 28 minutes ago

  Most AI coding tools give you one shot: prompt, response, hope it's right. OpenChamber treats every agent turn as a node you can fork from, which sounds like a small UI choice until you realize what it actually kills — the anxiety of "do I let this keep going or stop and restart." You don't restart anymore. You fork from the turn before things went sideways, try a different model or prompt, and the original branch just sits there intact if you need it later. Undo and redo work the same way, per turn, not per session. Where it gets more interesting is multi-run. Write one prompt, pick up to five models, and if you turn on isolation, each one gets its own *** worktree — a separate checkout of the repo, same starting state, zero chance of one model's changes stepping on another's. You watch all five actually build the thing, then either pick the strongest result outright or use Fusion to merge the best pieces from each into a new session. That's a genuinely different workflow than what most people are doing right now, which is running one model, eyeballing the diff, and hoping. Five isolated attempts at the same problem, compared side by side in worktrees that never touch each other, is a real answer to model selection anxiety instead of just picking a favorite and living with it. The GitHub integration closes the loop on the other end. Start a session straight from an issue or a PR with the context already attached, send failed CI checks or review comments back to the agent without leaving the app, then merge from inside OpenChamber when it's done. Issue to shipped PR without a single tab switch. Worth being clear about what this sits on top of: OpenChamber doesn't run any inference itself. It's a desktop, browser, and VS Code interface wrapping OpenCode, the open-source agent runtime that actually executes the model calls and tool use underneath. OpenChamber's entire value is the layer it built around that — versioned branching, worktree isolation, GitHub-native session starts — the stuff a raw agent runtime doesn't give you on its own. Five parallel model runs, each in a clean isolated worktree, compared and fused into one result. That's the part worth sitting with if you're still treating every agent session as a single roll of the dice.

• 
  observer (@rznstn) reported 29 minutes ago

  @theo @shadcn Do some kanban or a tasks queue, so it feels like a teammate, and then let user move GitHub issues into those kanbans

• 
  Rene Faurskov (@renefaurskov) reported 31 minutes ago

  @SimonHoiberg I use GitHub and I like it; but if remote is down I just push via local.

• 
  Offensive Lab (@OffensiveLab) reported 31 minutes ago

  North Korean state hackers are no longer content with simply typing commands into public chat bots. One of the country's major intelligence groups has begun running artificial intelligence (AI) offline on its own servers, connecting document search tools to the files it possesses, and starting to collect the software components needed to build AI into its malware. South Korean security firm Genians says it uncovered this setup after months of tracking and log analysis of infrastructure linked to a hacking unit, Kimsuky, under the North Korean Reconnaissance General Bureau. Genians found no evidence that the group had trained its own AI model, and the firm does not present this as a certainty. Instead, it describes the actor as being in a "research and knowledge acquisition" phase, collecting and testing existing tools rather than creating new models, with the clear goal of incorporating AI into operations, from writing malware to analyzing data. For an intelligence unit that has spent years targeting government, research, and other strategic entities, this suggests a shift towards attacks that are more sophisticated and difficult to detect. There's nothing here that's particularly novel, the weight falls on the defenders. Once the AI starts writing, it reveals a vulnerability they once relied on: machine translation, clumsy formatting, and spelling errors. What the machine inserts becomes a tell. Genians' report tells defenders to look for LNK execution, PowerShell, hidden scheduled tasks, GitHub traffic, and payload activity, primarily judging how "flashy" it appears, rather than relying on more subtle indicators.