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:

• 
  CulturedNiichan (Kuro) (@culturednii_v2) reported 0 seconds ago

  interesting. Gonna mirror it in my private *** just in case, you know, github, microslop, corpos. So far LibreWolf is totally fine with me, and I guess I could always install adblocking in my opnsense firewall, but still, pretty neat if you don't have a firewall/server

• 
  Carlo (@Italianclownz) reported 2 minutes ago

  @do_re_me_bo @barackomaba @KyleHessling1 5bit quants typically run slower so when you compare rocmfp4 compare to 5bitquants in decode speeds. Rocmfp4 supports mtp, qat, eagle 3 and all standard speculative decoding. I am adding rocmfpX quants to it so you can compress a BF16 down to Rocmfp3 and it will be equal to 4bit quants etc. Decode speeds have been better on AMD hardware vs baselines. There has been a lot of testing by the community. I also have a basic decode speed table on qwen models on the github.

• 
  Khakha (@khakha_x) reported 7 minutes ago

  So first they Anthropics Models went down from GitHub Student Developer Pack, and now Digital Ocean is revoking the $200 credits. What exactly is happening? Can't GitHub negotiate on behalf of students. I wanted to run Hermes Agent on that thing, with high-end models, but now I can't do that.

• 
  Arth Singh @ICML’26 🇰🇷 (@iarthsingh) reported 12 minutes ago

  @bhatia_mehar Mehar the github seems to be down, any reasons for that ?

• 
  Phil | Rentier Digital Automation (@rentierdigital) reported 13 minutes ago

  boris cherny stopped prompting claude. his job now is writing the systems that prompt claude for him 100% of his personal code for 30 days straight came from loops he'd set up once, not from manual prompting sessions. that's not a flex that's a timeline most devs are still on rung 1 or 2. rung 1 is claude as autocomplete you review every line. rung 2 is juggling 5 claudes in parallel routing between them manually, thinking you're advanced rung 3 is different architecture entirely. you don't prompt better you stop prompting. you encode the logic into something that runs without you. claude executes against conditions, verification gates, retry logic you designed once. it fails succeeds hits edge cases you didn't anticipate—the loop handles it the gap between manual prompting and loop engineering looks invisible at first. week 1 feels the same but one trajectory improves the work you already do. the other builds a system that handles that category while you design the next one linear vs compound that's why the june 7 moment mattered. the scoreboard went public karpathy's running 50 ml experiments overnight on a single gpu. agent modifies training code reads results iterates. no human in the loop. he called it the loopy era of ai github data shows claude code at 4% of all public commits. that's not happening through manual sessions running individual prompts to ship production code is the it works on my machine of agentic development now i build and ship daily with Claude Code. SaaS, tools, automations. ⭐ if AI can build it, I've probably broken it first. what works → link in bio

• 
  aichina.news (@AiChinaNews) reported 14 minutes ago

  The story of this cycle is practical engineering over parameter bloat. While Western attention defaults to Hugging Face, Alibaba's ModelScope platform continues to ship highly capable open-weight foundations. The standout release is Qwen3.6-35B-A3B, a multimodal Mixture-of-Experts model aimed directly at the autonomous agent space. It houses 35 billion parameters but activates just 3 billion during inference, keeping compute costs in check while retaining heavy-duty reasoning. More importantly, it integrates native "Thinking Preservation"—forcing the model to deliberate internally before committing to an output. This isn't for generating isolated snippets; it is explicitly engineered for repository-level software development. Meanwhile, the Chinese open-source community is aggressively filling the workflow gaps left by Western AI giants. A flurry of updates hit GitHub this week for the localised Claude Desktop client, pushing it to version 1.6.26. What began as a simple language patch has evolved into a full-scale project console. The community has bundled a Windows runtime to drastically lower the setup barrier for Anthropic's "Computer Use" capabilities in China. They didn't stop at API access—the client now features Kanban boards, local *** integration, IDE-style multi-tab workspaces, and multi-agent task orchestration. This is what happens when developers tire of waiting for official enterprise tools and build the scaffolding themselves. Hardware reality continues to dictate software deployment in the domestic market. Eco-Tech released highly optimised, production-ready versions of Zhipu AI's GLM-5.1 specifically tailored for Huawei Ascend NPUs. Available in W4A8 and W8A8 quantization, this is actual engineering substance. Rather than chasing theoretical benchmark supremacy, these releases are built for high-throughput inference, solving the memory overhead bottlenecks required to run heavy models on domestic data centre and edge hardware. The rest of the cycle's open-source radar is clogged with automated filler. Projects like SpecFusion, ZLabs-RoundPix-12px, and a dizzying number of game localisation patches pushed updates where the public summaries literally contain unrendered placeholder variables like '{release_date}' and '{explanation}'. If a team cannot be bothered to fill out their own PR templates, no working professional should be bothered to review their code. Elsewhere, YiMu-Subtitle-Translator pushed a minor update for AI video localisation that boils down to standard API configuration tweaks dressed up as a launch. The industry continues to bifurcate: teams building production-grade infrastructure for real constraints, and teams automating their own noise.

• 
  punk@compile:~$ (@PunkCompiler) reported 15 minutes ago

  @sukoooonnn it was not about deployment but one time i was was trying to push my secrets to github without adding them to *** ignore file and it was giving me error and evertime i push but i was in hurry i didnt able to get it for a hour (and yes i got late)

• 
  MrGenius (@oneitonitram) reported 22 minutes ago

  @ozdotdev @DavidPlakon @warpdotdev you can go ahead and create a github issue for the same, cant provide a screen recording for now

• 
  Dmitrii Malakhov (@malakhovdm) reported 22 minutes ago

  @hii_mohit Caught myself watching my agent scroll through GitHub issues I should've been screening myself. Outsourcing my taste and spectating.

• 
  OutspokenGeek (@OutspokenGeek) reported 24 minutes ago

  @JRogrow Why? What's your thesis for investing in Microsoft? I'm going to be harsh: No frontier model of their own. No ability to integrate AI well within its products. Lost early coding leadership in Github Copilot and then wh0red out the Copilot brand. Lost early AI infra momentum with the pause they did. Nothing much to show in semi hardware development despite earliest to start with XBox. OpenAI clearly moving to partner more with others. Satya took eye off owning AI and gave carte blanche to someone from Meta. Multiple rounds of layoffs since they need the money for the capex arms race. Even Windows 11 is widely hated for the ensh!ttification that they are just now attempting to fix. AI in Office needed Anthropic to show them what to do.

• 
  0xranjan | Polymarket (@iammrranjan) reported 27 minutes ago

  Here i will tell you clear hint about $POLY Since June 1st, @mustafap0ly has been grinding on a private repo with over 753 GitHub contributions in just 10 days. That's not normal maintenance activity, that's launch-mode intensity. He's been posting 100+ contributions per day consistently, which suggests something major is being built behind the scenes. > and only thing this time maybe private is about $POLY and if you know he also hinted about it on 5th june that "claude is not working, using codex" and 2 days before he told that " his mythos working overtime by showing $35k$ spend on mythos already in 7 days " so all the thing github heavy activity and his lined up hint in this 13 days may believed to us that $POLY is coming Nothing is confirmed, but one thing is clear: Polymarket team isn't inactive, they aren't ignoring the community, and they aren't standing still. They're quietly building. Maybe it's $POLY. Maybe it's very sooner

• 
  Johnny Nel | AI for Founders (@JohnnyNel_) reported 27 minutes ago

  🚨 An open-source AI agent just hit number one on OpenRouter... and almost nobody checked if it was safe to run Everyone's racing to install it. $8 VPS. 170,000 GitHub stars. Self-improving skills. So I ran an actual security review before trusting it with my server. The findings are wild... 👇 The part everyone skipped in the hype: → The default config ships with FOUR critical and nine high severity findings → On local, it passes commands straight to your shell — no sandbox, no allow list → A poisoned skill becomes a permanent prompt injection that fires every time it's reused → And there was a real supply-chain incident: a backdoored dependency harvesting API keys, SSH keys, and cloud credentials And it gets bigger: the feature everyone praises — agents that learn and reuse skills forever — is the exact same door an attacker walks through once. Builders installing it blind. "Self-learning" silently turned off by default. Skills quietly going stale and making agents confidently worse over time. The project calls it powerful. Builders should call it powerful AND loaded. Here's what actually matters though: ✅ An agent that remembers and compounds on your work beats any disposable paid sub-agent ✅ But if you skip the security setup, you're one bad prompt away from full shell access to your machine ✅ Own your stack and lock it down first — or don't run it at all So the question isn't whether Hermes is impressive. It's whether you've hardened it before you hand it the keys — or whether you're about to learn the hard way. Full breakdown in the video below 👇

• 
  Smukx.E (@5mukx) reported 28 minutes ago

  @NinjaParanoid @0xTriboulet @github I have asked about issue very clearly. No response from them since its an weekend... Lets see how this goes..

• 
  Open-source Projects (@the_osps) reported 29 minutes ago

  • Debug every JSON-RPC message and OAuth exchange with full traces • Chat with any LLM and see tool calls and context across agent and server • Run evals across multiple LLMs and track accuracy over time • Wire CLI and SDK into GitHub Actions for e2e tests on every PR

• 
  Onyx_Digital (@BaximusCyber85) reported 31 minutes ago

  @Father_Of_Geeks @koko_matshela All good. Where I think the friction appears is further downstream. Programming languages aren't just syntax. They're ecosystems. A student eventually has to read: Stack Overflow posts GitHub issues Python documentation Error messages Library documentation Research papers And almost all of that is English. So the challenge becomes: Does CMT-IsiZulu become a bridge into programming? or Does it become an island?