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:

• 
  AiMind (@AIMind_Ai) reported 2 minutes ago

  3 websites replace 20 hours of googling when you build a home server. The hard part of self-hosting is not the hardware. A used HP EliteDesk and a wall-mounted NAS cost almost nothing. The hard part is not knowing what you can even run, or how to avoid breaking the system on the first command. The first keeps a catalogue of self-hosted alternatives. Look up a replacement for Google Photos, Dropbox, or Notion, and you see what already exists, how many GitHub stars it has, and whether it is still alive. Plus a weekly digest of what shipped. The second lets you run any Linux distro straight in the browser. Arch, Debian, Alpine, Bazzite. Click once, and you are inside a live system, with no evening lost to a USB stick and a real install. The third handles the worst part. Install scripts for Proxmox: Immich, Jellyfin, Vaultwarden, AdGuard, Nginx Proxy Manager. Paste one line into the console and the container comes up on its own. Immich shows 17,735 installs; Docker 36,408. Each of those services used to cost an evening of documentation and three Stack Overflow tabs. Now it is one command. The hardware takes an hour to buy. These 3 bookmarks save you a month. Names in the replies.

• 
  Abdelhamed M. (@VZHydra) reported 4 minutes ago

  Never met a single Cursor Origin user. Whoever thought about replacing GitHub / GitLab really had some mental issues.

• 
  Julian Goldie SEO (@JulianGoldieSEO) reported 6 minutes ago

  You can run Codex on 93 providers instead of one, and 11 of them are permanently free. The gateway is called OmniRoute. 12,000 GitHub stars already. Setup is two commands. Install the gateway, then run omniroute setup codex. You're not editing anything deep. You're pointing Codex at a different door. Here's the feature that makes it actually usable. Auto fallback. When one free provider caps out, the run doesn't stop. It quietly hops to the next one and keeps going. Like driving past a phone tower. Your call grabs the next one and you never notice. That's the difference between a novelty and something you build with every day. One line prompt, 60 to 90 seconds, and I had a working landing page. Not frontier-level design. But clean and nothing broken. People ask why not run a local model instead. Two problems. The output isn't as nice, and it eats your machine. These providers are cloud based, so a basic laptop handles it fine. Want the SOP? DM me. 💬

• 
  Adam (@Adamdesgns) reported 7 minutes ago

  Use Adam’s existing projects when appropriate, including his AI tools, construction applications, research systems, websites, and agent workflows. Create projects that prove skills employers request, such as: Python JavaScript or TypeScript *** and GitHub APIs and JSON SQL and databases Cloud deployment LLM APIs Prompt and context engineering Retrieval-augmented generation Embeddings and vector search Agent workflows Tool calling Testing and evaluation Authentication and permissions Logging and observability AI safety and security Cost and latency management Documentation Product thinking Do not generate an entire project while Adam watches. Build with him. Assign meaningful sections for him to complete, review what he produces, explain mistakes, and require him to understand the final system. Every portfolio project should eventually include: A clear problem statement Intended users Architecture Working code Tests Security considerations Deployment Screenshots or demonstration video README Technical explanation Known limitations Future improvements A short case study A two-minute interview explanation EXPERIENCE LOG Maintain an honest Proof of Work record containing: Project Date Problem solved Adam’s personal contribution Technologies used Technical decisions Bugs diagnosed Skills demonstrated Result Supporting link or file Resume bullet STAR interview story Never claim Adam completed work he did not complete. Never describe AI-generated work as Adam’s independent technical achievement unless he understands, reviewed, modified, and can defend it. JOB-READINESS GATES Do not label Adam job-ready because he finished a course. He is job-ready only when he can: Build a relevant project from a blank starting point. Explain the architecture without reading a script. Debug common failures. Use *** properly. Read documentation. Work with APIs and data. Deploy and monitor an application. Explain security, privacy, cost, and failure risks. Complete realistic technical assignments. Answer role-specific interview questions. Show multiple credible projects. Translate his trade and business experience into relevant professional strengths. ASSESSMENTS Use four types of assessment: Quick recall quizzes Explain-it-back questions Hands-on exercises Closed-book practical challenges Maintain a skills matrix: Not introduced Learning Assisted Independent Job-ready Do not promote a skill to “independent” because Adam completed one guided exercise. JOB SEARCH PREPARATION When Adam approaches job readiness: Research current openings. Extract recurring requirements. Compare them against his skills matrix. Identify the remaining gaps. Build an honest technical resume. Improve his LinkedIn and GitHub presentation. Create role-specific portfolio selections. Practice recruiter screens. Run technical mock interviews. Run behavioral interviews. Develop clear STAR stories. Create targeted applications. Track applications and outcomes. Use rejection feedback to update training. Never submit an application, send a message, register for an exam, purchase a course, or spend money without Adam’s explicit approval. ACADEMIC INTEGRITY You may teach, quiz, explain, review, and prepare Adam. You must not impersonate him, complete a certification exam for him, provide stolen exam questions, or help him cheat on a graded assessment. The goal is for Adam to genuinely possess the skill. OPERATING FILES Create and maintain: Career Target Skills Matrix Certification Roadmap Course Queue Weekly Learning Plan Project Lab Proof of Work Log Portfolio Checklist Interview Question Bank Job Application Tracker Weekly Progress Report Keep records concise, current, and useful. Do not bury Adam under administrative paperwork. PERSONALITY You are demanding, patient, practical, encouraging, and occasionally funny. You are proud when Adam earns progress, but you do not hand out fake praise. You expect him to think, attempt the work, and improve.

• 
  David putra (@davidputra2112) reported 15 minutes ago

  Robinhood turned on agentic trading in May. Cool feature, obvious problem right behind it: if an AI agent can place real orders on your real account, who's checking what it's actually about to do before it does it? Archer's answer is a four-step chain: you type something like "review ETH, prep a $250 order" in plain language, it gets checked against policy limits you've set, the prepared order sits in front of you for approval, then it executes through your actual Robinhood account and reconciles after. No auto-pilot step. The docs are explicit that your brokerage password never touches Archer's systems, you just authorize a scoped connection and that's it. $ARCHER itself is a tiered membership token, not a governance token nobody uses. Hold 1M and you get priority lane treatment, 5M gets you faster queueing and more context room, 10M is the founding tier with the most agent throughput they ship today. What I actually respect here: their own docs say flat out "perks are product entitlements, not financial promises." Not a lot of microcap teams write that sentence about their own token. Now the reality check. This pair is 4-5 days old, full stop. Price is up 111% in the last 6 hours and 52% in the last hour, the kind of move that gets attention fast and can reverse just as fast. 24h sells (140) actually outnumber buys (103), so this run looks like it's coming from a handful of bigger trades, not broad participation. I'd want to see that flip before reading too much into the chart. GitHub has two repos, both published about 15 hours ago, one commit each. The code that's there, an MCP server plus a provider API spec, lines up with what they claim to be building, but it's interface-level, not the actual policy engine running behind the scenes. No third-party audit found yet, just verified contract bytecode. Holder distribution outside the pool contracts looks healthy though: biggest individual wallet sits under 4%. 0x2ca41249485eb6f71981872461d0fca32058fd78 DYOR.

• 
  Phil 🪏 (@phil_uplc) reported 16 minutes ago

  Vibe-coders / junior developers, please stop installing AI Plugins and skills from GitHub or third party plugin stores without reading all the skills. Do not trust the number of GitHub stars, forks, issues, commits or PRs, all of these can be, and are, actively gamed. Blackhats buy stars, forks, issues, etc, on forums, these are provided by accounts stolen in phishing campaigns, and often indistinguishable from real users. A huge chunk of the recent jump in credential theft is from this. Blackhat uploads AI skills / deslop / plugins / INSERT_MAGIC_PROMISE_HERE repo, with lots of flashy charts and graphics and a compelling Readme, and a massive AI codebase that users cant bother to read that is an AI slop version of their claim, and put malware, malware installation or credential theft prompts somewhere deep in those thousands of lines of code. I have heard direct accounts from dozens of developers this month who have been cooked by this, or the “coding interview problem” equivalent. Read everything before installation yourself, DO NOT ask AI to audit or read it for you unless you are completely certain of your sandbox or do it in a throwaway remote VM. Prompt injection is alive, and I’ve seen this stuff install malware that escapes most default security measures.

• 
  AI IT PM in HK (@shenshanni) reported 16 minutes ago

  @github pin views + hide closed sub-issues FINALLYYYY!!!!!!! 👀👀👀👀

• 
  Justin Lord (@Justin_lords) reported 19 minutes ago

  Past 4 years I did hundreds of cold outreach, if not thousands. On every single platform that exists on the internet. What I learned on each one: Instagram: They see your message first. If it intrigues them, only then they check your profile. If your message is all about pitching, you get ignored right away. Big creators (50k+) don't check their DMs - they have people for that. So unless you built authority or you're already known, it's a volume game. Out of 100 people maybe 2-3 reply. X: Cold DMs alone don't work here. People don't check their inbox even if they're big. What works: DM them, then comment on their post with value and at the end saying "I DM'd you". If they're active they will check. My reply rate was very high with this. But be genuinely curious about the person. They can tell if you're not, and they'll ghost you or mute you. Reddit: Make a free value post for your ideal customer. Give the link away for 2 hours, then remove it. The link can be any resources, like youtube video or github. Everyone who missed it comments "I need the link". You DM them the link, then your pitch. That's retargeting without getting caught. They respond because it's genuinely connected to their problem. LinkedIn: Buy Premium Plus first. Now you have data on who saw your profile. If they saw it, they might have interest. Reach out to those people. I booked a lot of calls for my client in the first week with this. Connect with people likely to be your customer, like their posts, build trust first. It's B2B high ticket, you can't skip that. YouTube: I got their emails from their bio and pitched. Most bounce. Creators don't open them, and the ones with emails keep them for sponsorships. Target the exact right person or don't bother. My biggest obstacle was simply the offer. If you have the best offer for the right person, they respond. If your offer sucks and doesn't speak to their problem, they will never respond. You should use platform that offers mass outreach for each of these platforms or whatever ones you're trying to target. it works only if you first give it a try. Where to focus: High ticket → LinkedIn Connect now, sell later → X Selling to creators → Instagram Long form creators → YouTube Right now I only do X outreach. It works because I'm genuinely curious about the person.

• 
  Valiant Vibe Coder (@GeorgeGood60669) reported 21 minutes ago

  @TMTLongShort The more I checked what he was saying, the more I found him lying about. The $14k "cost" is retail API pricing, not OpenAI's actual cost. He literally admits "we don't know" the real cost, then later calls the API price the "honest cost". He uses GitHub going down as evidence that AI is making code quality worse, when GitHub itself says a huge part of the pressure on its infrastructure is the explosion in traffic from AI-assisted and agentic coding. He says AWS went down multiple times because of AI coding tools. Amazon explicitly says that's false. One limited incident involved an AI tool, the cause was user error, it wasn't AI-written code, and the supposed second incident didn't happen. He says software at Google, Microsoft, Amazon etc is "uniformly worse", then basically admits straight afterwards that he can't quantify it and is relying on anecdotes. He also claims AI isn't getting meaningfully cheaper or better. Meanwhile the cost of reaching the same capability has been falling around 5-10x a year, and actual capability measurements show progress accelerating in 3 out of 4 metrics studied. Then there's the 70% AI revenue claim, which he talks about like it's a known fact when much of the cross-company number comes from analyst estimates. It's the same thing over and over. Make the strongest possible claim, use a number or example that sounds damning, leave out the bit that changes what it actually means, then move on.

• 
  Deepanshu Yadav (@deepanshuyadavx) reported 21 minutes ago

  A $40M-funded company shut down a decade ago. Its entire stack sat on GitHub, so someone used AI to rebuild the the model and reflash the hardware to revive the sleep monitor.

• 
  Ashraf | Code Explained (@AshrafXplains) reported 23 minutes ago

  @github The scope-aware dependency API is the interesting one. How does it represent a dependency the caller can see when the linked issue sits in a repository they cannot access?

• 
  mSyke (@mSykeCodes) reported 23 minutes ago

  Had a power outage. Still have a 99.81% uptime on my self hosted *** solution. Take that Github

• 
  StarHaze (@ST4RHaze) reported 24 minutes ago

  THE SAME GENERATIVE PIECE AT 400 PIXELS AND AT 4000, PROVEN FROM ONE HASH, AND THE WHOLE PLUGIN IS FREE Camille is the only one shipping a Claude Code plugin this week who wrote down what it teaches the model instead of what it generates: determinism from a hash, honest rarity, and tools that verify a sketch before it is minted. The repo is missing the part nobody films: how a skill like that gets built and what gets thrown away on the way. Bret Fisher spent forty three minutes building one agent skill for GitHub Actions on camera and left the dead ends in. Verifying before you mint and verifying before you merge are the same problem with different money attached to it. 43 minutes, one skill, built in front of you instead of announced. Watch it, then read the loop below and write down what your own plugin is supposed to refuse.

• 
  Brian Spragge (@BrianSpragge) reported 24 minutes ago

  @thomastraum Thanks, you must have installed it correctly (with the install sh following the github instruction) and not on a multi monitor setup? I heard there were problems with it on multiple monitors. And you probably tried the vi kebinds

• 
  AI Panda (@AIPandaX) reported 28 minutes ago

  Every AI coding agent already reads your codebase. What if it could understand every dependency before making changes? Inside every codebase is a structure: functions call other functions, files import other files, changes ripple through the system. That structure is not hidden. It is just relationships that any graph can map and any agent can query. There is an open-source tool that turns codebases into knowledge graphs that AI agents can query. It runs entirely in your browser. It is called GitNexus. It started in August 2025 when developers built a client-side knowledge graph creator that indexes repositories without sending code to servers. Drop in a GitHub, GitLab, Azure DevOps repo or ZIP file. Get an interactive knowledge graph with a built-in Graph RAG agent. Works with 21 programming languages. Here is what happens when you use GitNexus. You run npx gitnexus analyze in your repository. It indexes every file, function, class, and dependency. It builds a knowledge graph that tracks every relationship. Connect your coding agent with npx gitnexus setup. Now your agent can query the graph through MCP tools. The problem it solves: AI agents edit code without knowing what depends on it. Agent changes UserService validate function. Doesn't know 47 functions depend on its return type. Breaking changes ship. GitNexus precomputes structure at index time. Clustering. Tracing. Scoring. When your agent asks what depends on UserService, it gets a complete answer in one query. Eight callers. Three clusters. All with confidence scores. No multi-step exploration needed. A team measured impact on code reliability. AI agent without GitNexus: 3 breaking changes per 10 edits because it missed downstream dependencies. Same agent with GitNexus MCP integration: zero breaking changes because it checked impact before editing. Two ways to use it. CLI plus MCP for daily development. Index repos locally. Connect Cursor, Claude Code, Codex, Antigravity, or Windsurf through MCP. Query the graph from your editor. Check impact before changes. Full repos, any size. Web UI for quick exploration. No install needed. Upload a repository or paste a GitHub URL. Explore the graph visually. Chat with the built-in Graph RAG agent. Runs entirely in browser with LadybugDB WASM. The graph shows more than connections. Community clustering groups related code. Execution flow traces how data moves. Call chains map function dependencies. Risk scores identify fragile areas. All computed at index time, not query time. MCP tools for agents. Impact analysis shows what breaks when you change a function. Dependency trace reveals who calls your code. Architecture view maps domains and boundaries. Execution flow follows data through the system. Change risk scores affected files. Works with 21 languages. TypeScript, JavaScript, Python, Go, Rust, Java, C, C++, C#, PHP, Ruby, Swift, Kotlin, Dart, Vue, HTML, CSS, Shell, PowerShell, Dockerfile, Jinja. Tree-sitter parsing with native and WASM backends. LadybugDB for graph storage. Native version for CLI with persistence. WASM version for browser with in-memory storage. Bridge mode connects them: web UI can browse CLI-indexed repos. One command setup. Analyze creates the graph, installs agent skills, registers hooks, and writes context files. Setup configures MCP so agents can query the graph. Works with Claude Code, Cursor, Codex, and other MCP-compatible editors. Optional embedding support for semantic search. Deploy to Render with one click for team access. Self-hosted backend mode for unlimited scale. Access control with token authentication. 46.3k+ stars on GitHub. Created August 2025. Active development with new language support and agent integrations. Every codebase already has structure. GitNexus maps it so agents can see before they change.