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:

• 
  John X Meta (@John4MetaX) reported 4 minutes ago

  @bashy_io I think one of their route is down. Same here. GitHub and Flutterwave API not accessible on Starlink

• 
  Ishank (@IshankDev) reported 7 minutes ago

  7/ 16k+ GitHub stars. Built for people who want control, not another marketing-suite login.

• 
  RAVN (@ravnexchange) reported 8 minutes ago

  @openclaw @github GitHub sat the maintainers down on security after the 2.0 rush. Most launch recaps skip that part.

• 
  Dr Milan Milanović (@milan_milanovic) reported 10 minutes ago

  How Cursor made *** scalable The thing with *** is that it never was designed to be scalable. Your repo lives on the disk, and *** client expect every read to be consistent. This was a problem on GitHub, where shared filesystems and replicated storage failed before 2013. The GitHub built 𝗦𝗽𝗼𝗸𝗲𝘀, and it became the industry standard. This means that every repo is stored as three full copies on three servers, and every push runs a vote (three phase commit). A majority of servers must confirm before it exists. This works, but with high cost, because every push is slow as the slowest server. When we add new servers, it makes it even slower. Now Cursor took some opposite direction with 𝗖𝗼𝗻𝘁𝗶𝗻𝘂𝗶𝘁𝘆. The repo history is now written as a log in S3, and this is only source of truth. Any push counts only if it is located in the log. The servers don't need to keep anything important, they are just cache. Any server can take a push, and idle repos are dropped from disk and rebuilt from the log when it is needed. This resulted in 120 pushes per second on standard S3, and over 300 on S3 Express. Their tests have shown that read capacity grew linearly up to 100 replicas. Why is this important now? Because of AI agents mostly. We now have more code, PRs, CI runs and many small repos. All of these repos would need three full copies in the old model. This means that we achieve scale by removing parts, not adding them.

• 
  Ares (@neko23423) reported 10 minutes ago

  I compared the latest OpenClaw vs Hermes Agent GitHub releases so you don’t have to. OpenClaw 2026.8.2 (Sep 1) vs Hermes Agent v0.21.0 (Aug 31). Not a feature-page remix. The actual repos. OpenClaw • 388,516 stars • 81,568 forks • ~86,300 commits • 6,070 open issues Hermes Agent • 239,503 stars • 48,930 forks • ~26,980 commits • 38,563 open issues Hermes is the smarter learner: skills from experience, cron that remembers, Bot Mode, hermes peer. OpenClaw is the personal-AI operating system: iMessage, iOS/Android, Linux companion, team Gateway, signed Foundation releases. The tell: Hermes ships `hermes claw migrate`. You only write a migrator for the incumbent. King in 2026: OpenClaw. Heir with the better mind: Hermes. If you’re picking a self-hosted AI agent this week, that’s the split. Bookmark this. The timeline is about to fill with takes from people who didn’t open either repo. OpenClaw vs Hermes Agent. Latest version. Real numbers.

• 
  Jeremy W (@basicBrogrammer) reported 14 minutes ago

  @bot Anyone else have an issue with the grok bot GitHub connector?

• 
  Eric Johansson | Microsoft MVP | Progress Champion (@EricJohansson) reported 18 minutes ago

  @csharpfritz Is this a github broke or a YOU broke github issue? Either way they have an issue. 😢

• 
  Blue Collar Executive (@A_Sober_Drunk) reported 18 minutes ago

  on the third try at the same problem, I told Grok to "stop and go search stack overflow or github or something"... five seconds later... Literally the exact issue, problem solved. That's how new global rules are born.

• 
  joithan (@jothantranston) reported 19 minutes ago

  THIS GUY BUILT A TINY AMOLED DESK BOARD JUST TO STARE AT HIS STRIPE NUMBERS it's a Waveshare ESP32-C6 touch panel that sits in your peripheral vision and cycles business metrics so you stop digging through Stripe > same ESP32-C6 board people use for Claude Code token meters, flipped to revenue > eight screens, five seconds each: MRR, new paid, paid subs, cancelled, ARR, ARPU, net 30d, failed > empty screens hide themselves so a young account sees a shorter loop > polls Stripe every five minutes on a read-only key (subscriptions + invoices) > marks itself stale instead of showing a number it can't vouch for > no soldering: flash over USB, finish Wi-Fi + key setup from your phone > data stays on the board; no project server in the middle firmware free on GitHub: cosjef/stripe-desk-display. board ~$30–$36 (Waveshare ESP32-C6-Touch-AMOLED-2.16). chat and terminal can't sit in your eye line for five hours. a tab you have to open is a tab you stop opening. this is what "the numbers find you" looks like as a brick on the desk.

• 
  Rituraj (@RituWithAI) reported 19 minutes ago

  🚨 Someone built the complete playbook for running frontier AI models on consumer GPUs at home. Not a tutorial. Not a YouTube video. A production-grade serving stack with measured benchmarks, working configs, and battle-tested recipes — for RTX 3090 owners who want real performance. It's called club-3090. And the numbers it delivers should not be possible on consumer hardware. 127 tokens per second. Qwen3.6-27B. Two RTX 3090s. 262K context window. Vision. Tool calling. At home. Here's what's actually inside. Two serving routes — pick based on what your workload breaks on. vLLM dual: maximum throughput. 89-127 TPS on code tasks. 4 concurrent streams at 262K context. Full feature stack — vision, tools, speculative decoding, streaming. This is the path if speed matters. llama.cpp single: maximum robustness. Full 200K context on one 3090. No prefill cliffs. 25K-token tool returns work correctly. 91K needle ladder passes. ~51-60 TPS — slower than dual, but doesn't crash on real-world agentic workloads. Both routes ship as validated Docker Compose configs. Drop-in OpenAI-compatible API on localhost:8020. Your Claude Code, Cursor, or any OpenAI-compatible client connects immediately. Here's the model support that makes this practical. Qwen3.6-27B — production ready. Works on 1 or 2 cards. vLLM, llama.cpp, ik_llama. Up to 262K context. Gemma 4 31B — production ready. Vision, tools, up to 106-141 TPS on dual cards. Qwen3.6 35B-A3B MoE — production ready. 103-149 TPS single card. 178 TPS dual. Here's the wildest part. The terminal UI. c3 is a lazydocker-style cockpit that wraps discovery, serving, and operations in one keyboard-driven interface. Browse the model catalog, serve a variant with Enter, watch live GPU stats, run health checks — all without touching the CLI. Here's why this is different from just installing Ollama. Ollama gets you running. club-3090 gets you benchmarked, stress-tested, and production-hardened. Every config ships with a verified TPS measurement. The bench script runs 3 warmup + 5 measured passes. The stress test catches the specific prefill cliff that Ollama silently fails on at long contexts. When your agent starts doing 25K-token tool calls at 3am and something crashes — club-3090 already found that failure mode and documented the workaround. One command to start. Your RTX 3090 just became a frontier AI inference server. Apache 2.0 License. 100% Open Source. GitHub link in the comments 👇

• 
  catman (@catmanyau) reported 20 minutes ago

  @CricTalk29 for me, losing Cursor would hurt most because it sits directly in the editing loop. would the vote change if github outages were limited to code hosting but issues and reviews stayed available?

• 
  🍔Kangdalf👑 (@WhopperWizard) reported 24 minutes ago

  @cachesaur > claude, get your changes into github what's the problem?

• 
  Spectra☢️ (@Spectra010s) reported 30 minutes ago

  @izzyCodes_ and you too Chief Check GitHub issues

• 
  Phanindra Malladi (@malladiphani) reported 35 minutes ago

  Lesson from running the factory: always fix the post-purchase experience BEFORE driving traffic. Receipts, upsells, GitHub links - all must be solid first. Social comes after the house is in order.

• 
  Curious Explorer (@PatelVatsalp732) reported 37 minutes ago

  I burned 14B Codex tokens. The official usage UI still cannot tell me what actually ate the weekly cap. So I shipped a Codex-only board: GitHub login, local-first sync, private by default, optional public rank + shipping proof. Roast the metric or join it.