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:

• 
  narcotic_nik 🍉 (@narcotic_nik) reported 40 seconds ago

  @skyfall_ggs @paytondev Cosmic is basically in pre alpha and they don't even look at the issue tracker on GitHub

• 
  Fusion Fix (@fusionfix10) reported 3 minutes ago

  @MMaRsu An update was just released, try to redownload the fix and see if it works. If there's still a problem, open a GitHub issue with attached crash dump for further assist.

• 
  Commander Zolfaghari (@DirectorOfNATO) reported 4 minutes ago

  @bruvimtired Looks interesting, much faster than Github itself lol I wonder how much it cost you, because it does have some issues - No favicon - No logo - Avatar not showing up after upload - 404 page got no styling - When I change my username, it still has my old username cached either client or serverside Ok ok i'll stop bitching I like the idea

• 
  adamalix 🇩🇴 (@adamalix) reported 4 minutes ago

  I would assume GitHub is the tip of the iceberg for the onslaught of agent-driven traffic. Given that, I wonder: what is the next software product line that will have huge reliability issues as agents start reaching into them at scale?

• 
  Non-Linear (@nonlinear_james) reported 6 minutes ago

  @davidfowl @ksemenenk0 Hence why it’s completely unready for production. Everyone is focused on AI and your core services have major issues. Here a simple list: Azure app services: 1. Voluemount errors constantly on deploys 2. Volume mount errors don’t cause a job run to fail, just hang 3. GitHub actions for deploys don’t detect failure at all and thus leave your entire system down routinely because it doesn’t verify success. 4. UI for jobs can’t cancel failed/hung so you have to go to the cli. 5. Jobs logs has that antiquated mess of logging reporting from 10+ years ago and is awful. Azure k8s: 1. Azure storage mounts are 95% slower than they should be. None of the drivers solve it. 2. NFS is this complex mess of VPN like stuff and other crud and if you set it up, your k8s gets throttled by azure sql and other stuff outside the networking crap even if it’s internal and then your perf goes to hell. All of these are well documented on multiple GitHub repos. Someone senior should take an interest in making azure app container services an actually good product to go with aspire. (And then go and fix k8s later although you’ll probably fix k8s by fixing your managed k8s)

• 
  Ai or Die Now (@aiordienow) reported 7 minutes ago

  GitHub Copilot just paused new signups and cut usage limits. Classic playbook. Here's why this keeps happening and what to do instead. 🧵 The pattern: 1. Launch with subsidized pricing to get devs hooked 2. Usage grows faster than revenue 3. Quietly raise limits, cut model access, pause signups 4. Existing users get a worse product than what they signed up for This isn't a GitHub problem. It's the subscription model problem. From the official blog post: "It's now common for a handful of requests to incur costs that exceed the plan price." Translation: we gave you a deal we couldn't afford. Now we're taking it back. Every SaaS subscription is a temporary subsidy. They buy your loyalty with below-cost pricing. Once you're dependent, they adjust the terms. You stay because switching is painful. This is not a bug. It's how the model works. The HN thread is full of devs right now saying: "I hit my limit and I'm looking at running Qwen locally." That was always the answer. You just needed a rug pull to make it obvious. The self-hosted alternative for AI coding: - Ollama (local model runner) - free - Qwen 3.6 or Llama 3.3 - free weights - OpenCode + local model = zero monthly cost Steeper setup curve. Zero dependency on GitHub's business decisions. Same story plays out in WordPress plugins every year. Akismet, Yoast, WPForms - they all started cheap. Now they're $100+/year each. The answer is the same: own your tools. One-time purchase. No subscription hostage situation. The lesson from Copilot applies everywhere.

• 
  Cvewhen? (@cvewhen) reported 13 minutes ago

  why is github so slow omfg

• 
  Rajdeep Singh (@rajdeeptwts) reported 14 minutes ago

  hey @github @GithubProjects fix the @GitHubCopilot plans asap

• 
  Igor Gouveia 🦞 (@igorgouveiaf) reported 15 minutes ago

  @steipete Peter, in OpenClaw I have a problem where my agent can't send me a file in the Telegram chat to download. I always have to send it to GitHub to sync with Obsidian so I can download files. Is there a solution for this?

• 
  yung_gunzzz (@Michaelgunzzz) reported 15 minutes ago

  First,Ai doesn't understand problems the way humans do.Tool like ChatGPT or GitHub or copilots can generate codes but they rely on you to: Define what needs to be Built. Decide the architecture. Spot mistakes or security issues.

• 
  BP (@everyonebpup) reported 16 minutes ago

  🚨do you know what's happening with GitHub CLI? it went out in v2.91.0 a few hours ago. every gh command you run now sends a ping to github: command name, flags, OS, version, a device_id, and a field called "agent" that last field is the whole point. claude code, copilot CLI, cursor, and codex are all driving gh usage now, and github had no way to measure it. from their own issue (12522, january): "we do not have client side telemetry, it's not possible to say this with certainty" opt-out is three commands, any one works: export GH_TELEMETRY=false export DO_NOT_TRACK=true gh config set telemetry disabled small weirdness: the config command prints "unknown key" but actually writes telemetry: disabled to ~/.config/gh/config.yml. it works, the warning is a bug

• 
  Sergio Pereira (@sergiopreira) reported 17 minutes ago

  A CLAUDE.md file is the #1 trending repo on GitHub - 68k stars. But this is a v0 workaround. Text files describing the codebase going stale the moment code moves. The actual unsolved problem is 'can AI keep understanding your code as it drifts'.

• 
  Risto Anton (@blogtheristo) reported 18 minutes ago

  GitHub MCP in Claude Code is maintained by Anthropic, built into the platform, no public version number. It is distinct from GitHub’s own open-source github/github-mcp-server (Go, first-party, versioned) and from the deprecated @modelcontextprotocol/server-github (npm, archived). @ClaudeDevs @github this makes no sense.

• 
  David Fernandez (@dxvidfernxndez) reported 19 minutes ago

  I think the Gemini 3 Hackathon judging is broken. And I have proof 👇🏻 The rules clearly said: • Public working product link • AND GitHub repo (or AI Studio link) So judges can test + verify the code. But some winning projects don’t even have a working product link. Just a GitHub repo. And it gets worse… One of the winning submissions literally contains what looks like a prompt injection in the description. Not even cleaned up. Meanwhile, I built Cortex Protocol: A full MCP memory system Cross-platform AI memory Working product + demo + infra Didn’t even get honorable mention. I’m not saying my project deserved to win. But something about this judging doesn’t add up. Now I’m honestly questioning everything. I built a full working AI memory system… with real infra, real product, real demo. Didn’t even get honorable mention. Meanwhile… some winners didn’t even have a working product link. Am I missing something here?

• 
  George Fahmy (@kajogo777) reported 21 minutes ago

  Now people do cold outbound through open source github issues and PRs...this is ridiculous @github