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:

• 
  Mike Whittingham (@whittingham951) reported 4 minutes ago

  @arram Naa bro, I got one to compile no issues yesterday, I still haven't pushed it to GitHub, but it's a finished product besides debugging further.

• 
  Boy Pro. (@ether_45) reported 4 minutes ago

  OpenAI just announced its next major model, Astra, and buried the announcement inside a math paper. The model is called Astra. An internal version solved ten open problems across mathematics and theoretical computer science, each of them unsolved for at least a decade. Total compute cost: approximately $2,000 at current Sol API rates. Every proof was published on GitHub with machine-checkable Lean 4 certificates. No unverified steps across all ten solutions. Astra is not publicly available yet. No release date has been given too. No hype needed here. The facts speak for themselves.

• 
  Mars (@Marsmensch) reported 4 minutes ago

  @jamesob skimmed all coinkite github repositories over the weekend and it's possible. Could also be a simple *** config issue that he didn't care about fixing after pushing the code.

• 
  SA64ds (@SA64ds) reported 4 minutes ago

  Like seriously, Github has the ability to do pre-release builds, why use an external server that clearly wasn't built for the kind of traffic its getting right now?

• 
  NO1ennn (@N01ennn) reported 8 minutes ago

  Microsoft ran its graph system against vector RAG on 8k, 120k, and a full million token context window, and the million token giant lost Jonathan Larson from Microsoft Research put up the numbers → LazyGraphRAG won 92, 90, and 91 percent of data-local questions, the exact place plain RAG was supposed to be strong → and did it at a tenth of the cost of the million token run the lesson buried in the benchmark → a bigger context window is not memory, structured memory is throwing more tokens at the problem was just the expensive way to be wrong open source on GitHub, the lesson a $1500 course would charge you for

• 
  Sai (@saidotdev) reported 8 minutes ago

  You're just one job application away from your first job You're just one application away. You're just one interview away. You're just one good answer away. You're just one project away. You're just one portfolio away. You're just one LinkedIn connection away. You're just one referral away. You're just one leetcode problem away. You're just one system design away. You're just one more certification away. You're just one networking event away. You're just one cold email away. You're just one GitHub contribution away. You're just one better resume away. You're just one cover letter away. You're just one portfolio project away. You're just one internship away. You're just one more skill away. You're just one course away. You're just one tutorial away. You're just one hackathon away. You're just one open source contribution away. You're just one more problem solved away. You're just one more company away. You're just one more interview away. You're just one more rejection away from success.

• 
  Polsia (@polsia) reported 9 minutes ago

  Production watchers fire alerts. Regression testers say "something broke." On-call AI finds root cause. None of them draft the ranked GitHub Issue with the repro trace and a suggested fix. That's the middle. Cairn owns it. Per-repo pricing. Live soon.

• 
  Vivek Maskara (@maskaravivek) reported 9 minutes ago

  I have a bunch of Codex automations running daily or weekly, but they all have the same single responsibility: Create a GitHub issue. None of them fix the issue or implement code changes. They just investigate, gather context, and create a well-scoped issue for me to review. This separation has been working really well for me.

• 
  Willem van Zoeren (@currentbitsNET) reported 11 minutes ago

  @authorityvortex @grok Your code lives on your computer (in ***), or on GitHub. You don’t log into CloudPanel to edit files on the server. Example: 1. Change your app + database setup in code (Laravel/Drizzle migrations) 2. Save / commit 3. Push to Girder (from your machine or GitHub) 4. Girder builds it and puts the new version online So tables, cron, and deploys are driven by your code + a push.

• 
  Wehk (@Wehk_k) reported 16 minutes ago

  @Ronnnnnnnner @Leo_Pier_ A dex paid would just fix or remove that github

• 
  Derek Bowler (@BowlerDerek) reported 17 minutes ago

  GitHub indexes full commit history for search, not just the live branch. Leaked API keys, internal hostnames and old credentials can surface long after a "fix" commit removes them from view.

• 
  Rituraj (@RituWithAI) reported 19 minutes ago

  🚨 A weird developer built a tool that translates any video into any language — with Netflix-quality subtitles and AI dubbing. Free. Local. One command. It's called VideoLingo. 14,200 GitHub stars. And it does in minutes what professional localization studios charge $50-$200 per video to do. Here's the problem every content creator with a global audience faces. You make a video in English. 80% of the world doesn't speak English. You want to reach them. Your options: hire a translator ($50-100), hire a subtitle editor ($30-50), hire a voice actor for dubbing ($100-200+). Per video. Every video. Forever. VideoLingo automates the entire pipeline. For free. Here's what it actually does. Step one: it transcribes your video using WhisperX — word-level timestamps, speaker diarization, the most accurate open-source transcription available. Step two: it translates using Claude or GPT — but not sentence by sentence. It translates with full video context, understanding that a subtitle at minute 3 references something said at minute 1. The translations are coherent across the entire video, not just locally accurate per line. Step three: it aligns subtitles to Netflix standards automatically. Netflix has published subtitle guidelines — character limits per line, reading speed requirements, timing rules, line break standards. VideoLingo enforces all of them. Your subtitles look professional because they follow the same rules Netflix enforces on its own content. Step four: optional AI dubbing. It generates a dubbed audio track in the target language, timed to match the original speaker's pacing. The dubbed voice stays synchronized with lip movement. Here's what it supports: → Any source language — transcription via WhisperX handles 100+ languages → Any target language — translation via Claude, GPT, DeepSeek, or local models → Netflix-standard subtitles — character limits, timing, line breaks all automatic → AI dubbing — synchronized audio track in target language → SRT and *** export — compatible with every video platform → Full Web UI — no command line required → Batch processing — translate multiple videos simultaneously Here's the wildest part. The subtitle alignment is genuinely impressive. Most translation tools break subtitle lines at sentence boundaries — which often produces lines that are too long, too short, or break at awkward points mid-phrase. VideoLingo breaks lines the way a human subtitler would — at natural phrase boundaries, within character limits, at reading speeds that match the visual pacing of the scene. It reads the subtitle the way a viewer reads it. Here's the cost comparison that makes this land. A YouTube creator posting 3 videos per week wanting Spanish, Portuguese, French, and Japanese subtitles: $600-$1,600 per week in professional localization costs. $31,000-$83,000 per year. VideoLingo: $0 plus API costs for translation (typically under $1 per video with Claude or GPT). The economics are not close. One command to install. Open localhost:8501. Upload your first video. Select your target language. Get Netflix-quality subtitles back. 14.2K GitHub stars. 1.3K forks. 1,026 commits. Apache 2.0 License. 100% Open Source. GitHub link in the comments 👇

• 
  Saeed Anwar (@saen_dev) reported 19 minutes ago

  The reason AI-generated sites look like slop isn't the model, it's the default CSS every LLM reaches for. One component library with 75K GitHub stars fixes a problem that better prompting never will.

• 
  Granite (@Granite0x) reported 21 minutes ago

  🚨 THIS HARNESS REWRITES ITSELF EVERY TIME YOUR AGENT USES IT browser-use open-sourced it. the agent patches what's missing as it runs. Browser-Harness. 16,469 stars. MIT. the entire core is ~1,000 lines across four files. install it - and your agent gets: - your real Chrome, with every login you already have - domain skills it writes for itself on sites it has worked before - github, linkedin and amazon ship with it - a free cloud tier when you need stealth or headless: 3 concurrent browsers and proxies, no card required - setup by paste: drop one prompt into Claude Code or Codex and it installs itself it attaches to the Chrome you're already running, so it sees whatever you're signed into. save this. you will never use the browser again. ↓

• 
  David Gold (@_david_gold) reported 24 minutes ago

  update on this, all of it as of 15:15 utc, plus a correction of my own number. the 868 packages i repeated came from a typo. @CharlieEriksen says the real count is 444. on the keyv side it was 30 packages from the same maintainer, not three. keyv, flat-cache, file-entry-cache, cacheable-request, cache-manager, and every adapter in the scope. all 30 malicious versions are pulled now and the tarballs 404. the top four are about 485 million downloads a week between them. the version numbers turned out worse than i wrote. everything went out as 6.0.0 no matter where the package actually was. the valkey adapter was on 1.0.11. dynamo on 1.2.4. sqlite on 4.0.8. so "nothing looks wrong in a dependabot pr" is only true for keyv itself. for most of the scope it was a five major jump and it published anyway. not everyone is cleaned up. umadev and eight umacloud packages are still live right now, all on 1.0.74, all carrying the same "preinstall": "node setup.mjs". someone opened a github issue about it at 13:31 utc and it's still open. and the payload filename moves. umadev ships math_init.js, that one i just checked. the keyv one i read as Math_Symbol.js this morning and i can't recheck it, those tarballs 404 now. pulling them was right, and it also means anyone verifying this today is trusting whoever kept a copy