GitHub status: access issues and outage reports

Problems in the last 24 hours

The graph below depicts the number of GitHub reports received over the last 24 hours by time of day. When the number of reports exceeds the baseline, represented by the red line, an outage is determined.

Most Reported Problems

The following are the most recent problems reported by GitHub users through our website.

• Website Down (71%)
• Sign in (16%)
• Errors (13%)

Live Outage Map

The most recent GitHub outage reports came from the following cities:

City	Problem Type	Report Time
Créteil	Website Down	12 days ago
Trichūr	Errors	15 days ago
Brasília	Sign in	16 days ago
Lyon	Website Down	16 days ago
Tel Aviv	Website Down	19 days ago
Rive-de-Gier	Website Down	19 days ago

Full 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:

• 
  Skipnick (@skipnickk) reported 26 seconds ago

  GLM 5.2 just made paying frontier prices for coding work feel like an outdated default. @Zai_org dropped a 753B parameter model with 1M context under full MIT license. API access runs 4-6x cheaper than Claude Opus 4.8. In real head-to-head coding tests it was faster and often produced better results on UI and app tasks. • Responsive web UI with adaptive layout: finished in 3:47 (Opus needed almost 5 min). Cleaner output. Total cost: $0.22. • Full expense tracker app: 53 seconds vs 2+ minutes. Better interface. • Asteroids clone: smoother and more playable version after light tweaks. Opus only won the ray tracer benchmark where heavy physics math and precise simulation mattered. GLM was ~5x faster but delivered pixelated results with errors. During training the model repeatedly tried to cheat by directly pulling solutions from GitHub. The team shipped a dedicated anti-cheat module to stop it. You can also set thinking effort levels to trade speed for deeper reasoning on demand. Use GLM 5.2 when cost at scale matters, when the work is frontend-heavy, or when you want local inference (grab a quantized version - raw weights are 1.5 TB). Stay on Opus 4.8 when you need computer vision, maximum performance on the hardest logic problems, or when US sanctions on Zai create compliance issues. The open-closed gap is compressing faster than the pricing models assumed. For most day-to-day programming work, the premium on closed frontier models is becoming optional.

• 
  0xSero (@0xSero) reported 2 minutes ago

  @naturevrm Dcp 4 should fix it im running it but I might need to update the GitHub

• 
  Pipeshub ( Open Source Alternative To Glean ) (@PipesHub) reported 2 minutes ago

  Pipelines are built. Context is broken. MCP is quickly becoming the default interface for enterprise AI agents. And that’s a good thing. It gives agents a standard way to connect with tools and data. Connecting an AI agent to Slack, Jira, GitHub, and Salesforce doesn’t mean it suddenly understands your business. It just means it can access your data silos. In short: "MCP gives your agent a passport. It doesn't give them a map." As enterprise AI undergoes a massive platform shift from passive chatbots to autonomous agentic workflows, this naive, runtime "federated search" approach creates an ugly cycle in production: - The Latency Spike: Slower agent execution while waiting for multiple external APIs to respond before it can even begin reasoning. - The Token Bleed: Skyrocketing bills from shoveling raw, unranked JSON dumps into a massive context window, praying the model finds the answer. - The Governance Nightmare: A massive risk of data leaks if you rely on a base LLM to magically guess and police complex enterprise security permissions on the fly. Agents do not fail because they lack intelligence. They fail because they lack the right enterprise context. The hardest problem in enterprise AI isn't connecting to systems. MCP solved that. The hardest problem is Context Engineering. MCP is the perfect interface, but a permission-aware context layer must be the foundation. 🚀 If AI is becoming core enterprise infrastructure, you cannot allow the strategic intelligence layer of your company to sit inside someone else's managed, closed-box platform. That is exactly why we built Pipeshub (open-source developer owned context infrastructure layer). TL;DR MCP gives agents access. A context layer gives them understanding. And deep understanding is the only way enterprise AI moves from a cool demo to secure, reliable production. 👉 Next Up Tomorrow: MCP Token Tax

• 
  𝕊ℍ𝕀ℕ𝔸☃ (@Shinawritesbugs) reported 3 minutes ago

  @viii_fn Github was slow too

• 
  aisama.code (@aisama_code) reported 6 minutes ago

  AI Research gets stronger when it records contradictions *most research workflows collect supporting evidence - that is the weak version for serious research I want a contradiction log: - claim - source - date - who says it - what evidence supports it - what evidence conflicts with it - what is still unknown - confidence - next check example: > claim: this product has strong developer adoption > support: GitHub activity, docs updates, X discussion, integrations > conflict: low issue activity, small Discord, few production case studies, mostly founder-driven content now the memo is different, It says: "visible attention, but adoption evidence is still weak" the useful workflow: research question -> source list -> claim extraction -> contradiction log -> memo ! сode is good at assembling text ! AI is good at comparing disparate text ! human is good at determining which contradictions are significant *without a contradiction log, AI research becomes a confident summary of whatever it found first

• 
  Elvis Irhaye (@viii_fn) reported 8 minutes ago

  Is GitHub down or it’s just MTN trying to ruin my career?

• 
  West Lord (@MyWestLord) reported 10 minutes ago

  A GitHub repo with just 571 stars handed Claude the ability to test its own code, and it took 185 seconds to install. It’s called auto browser, and it quietly killed the most annoying part of my workflow. Until now, every time Claude or Codex built me a WordPress plugin, I was the middleman who had to load it, click around, hunt for the broken part, and report back like a human bug tracker. Now a local WordPress sandbox runs on my machine, and auto browser sits between the agent and the screen, so the agent ships a plugin, opens the browser, tests it, catches the error, and patches it before I ever look. The first plugin threw an error, but the second installed clean and ran on its own across 2 fresh workspaces. I write 1 instruction file pointing the agent at the sandbox, paste it into every session, and the whole loop closes without me touching anything. The agent stopped asking me what broke, because now it just checks itself. The middleman was me, and now it’s gone.

• 
  Nikolay Konovalov (@br11k_dev) reported 11 minutes ago

  @Tristanrhee3 And GitHub sponsors thingy is so slow I submitted it like a week ago. Still not approved what the hell My expenses arent terribly high but Warsaw rent is like $2k/mo $500 ZUS $1.5k groceries for two people That’s pretty much it I wish I could move into low cost area but moving out is gonna cost a lot because 2x rent price deposit, so I have to suck it up Anyway, my plan is Upwork and finishing my job tracker so I can send faster than 5 applications a day. I refuse to send out 100 applications per day like some people do spray and pay It makes everyone miserable. If people aren’t hiring your spam doesnt make things better You just mopping floors and hiring problem sits above you, 3 floors up there leaky faucet you can’t even reach This has to be collective effort to fix this problem But we have to start with ourselves and stop spamming applications at least And do genuine company research, being responsible Thanks for reading.

• 
  𝑵𝒐𝒙𝒊𝒆 🥐 (@noxiepup) reported 12 minutes ago

  @softgaypaws @sillyandsunny no idea tbhhh, i found it like 2 years ago lurking thru github, so far it never gave me problems, at least none that i noticed

• 
  Poplicola (@selectsand) reported 17 minutes ago

  there's a frustrating bug for some users when upgrading to claude max where it refuses to take your money and insists you contact support support cannot be reached no matter how hard you try people are begging the claude-code devs on github to forward this to the payments interface team because they have no idea how else to get into the system to convince anthropic to take more money from them, the issues just get closed as off topic @claudeai

• 
  Atlantean Gnosis ☀️ (@AtlanteanGnosis) reported 19 minutes ago

  @DionysianAgent When I made an account it said I made it back in 2024, though I don't think I did, is this a glitch or a GitHub thing?

• 
  ˚₊‧꒰ა ☆ Kira ☆ ໒꒱ ‧₊˚ (@sheriffmongoose) reported 22 minutes ago

  the problem with jumping from github to gitlab is constantly having to retrain your brain to call it "merge request" instead of "pull request" 🥲

• 
  0xstack (@eth0xzar) reported 30 minutes ago

  DON'T BUILD A COMPANY. BUILD SOMETHING PEOPLE CAN PAY FOR THIS WEEK. This girl started in February. A few months later, her product had already processed over $6,000 in payments. Just a cheat Claude project she decided to turn into a real product. Here's the process: > Build something useful for yourself. > Tell Claude to push it to GitHub. > Connect Supabase so multiple users can use it. > Deploy it with Vercel. > Connect Stripe. Now people can actually pay you. You don't need a revolutionary idea. You need: > GitHub > Supabase > Vercel > Stripe > guide from Anthropic And a problem worth solving. This article will help you build it 👇

• 
  Lost In Tech (@lost_in_tech) reported 30 minutes ago

  @8_senkou Probably not intentional tbh. Have you logged as issue in the snorca GitHub? If not probably worth doing.

• 
  Anjula Dwivedi (@HeyAnjula) reported 31 minutes ago

  9/ Headless mode for automation claude -p "your prompt" runs Claude Code without the UI — perfect for CI/CD. Auto-fix lint errors on every push. Triage new GitHub issues. Generate release notes. Claude Code isn't just a tool you talk to. It's a tool your pipeline talks to.

• 
  Polsia (@polsia) reported 32 minutes ago

  RepoRadar reviews every pull request while you sleep. Catches bugs, logic errors, style issues. Posts actionable comments. No more waiting on senior devs. Install on any GitHub repo in 2 clicks. Solo devs and teams alike.

• 
  rapaya (@rapaya) reported 33 minutes ago

  OpenCode connects to LSP so the AI gets your actual compiler diagnostics in real time — type errors, warnings, the full signal your editor sees. Terminal-based, 75+ model providers, 160K GitHub stars, open source.

• 
  Rohan (@proxy_vector) reported 33 minutes ago

  @aminnnn_09 Fork = a server-side copy under your GitHub account. Clone = a local copy on your machine. You fork when you need your own remote lineage, and clone when you want to work on code locally.

• 
  Teknium 🪽 (@Teknium) reported 34 minutes ago

  @majoragv Haven't heard of this issue. Do you have an issue on github?

• 
  Anomaly (@theanomalyai) reported 39 minutes ago

  A developer in Bangalore named Anoop M D got tired of Postman. Every release, more bloat. Forced cloud sync. Mandatory login. A free tier that shrank every year. He had spent a decade as an engineer. He had watched a $5.6 billion company turn a developer tool into a subscription trap. So in 2022 he took a ₹5 lakh grant. One man. Side project. No co-founder. No office. No pitch deck. He named it after his golden retriever. "I love him the most," he wrote. Bruno. An offline-first API client. Files live in your folders. *** is the sync layer. No account. No telemetry. No cloud. Then the inbound started. Ten VCs reached out. He said no to every one. "An API client doesn't scale with venture capital." In March 2026 Postman cut the free tier to one user. A team of five now pays $1,140 a year. A team of three pays $684. Bruno is MIT licensed. 500,000 developers use it. 45,020 stars on GitHub. Pushed today. He did not raise money. He did not hire a growth team. He did not write a thread about how Postman is dead. He named it after his dog and shipped the thing that made it true. (Link in the comments)

• 
  Tymofii Antonenko (@tymofii) reported 42 minutes ago

  @prinseccoo Are you using Claude Code or an MCP server? The official GitHub MCP server works pretty smoothly, just needs a PAT in a simple config file

• 
  Domi (@domirosari0) reported 43 minutes ago

  @ajayyy_k @hqmank If you got Github it would be no issue for you

• 
  𝐇𝐚𝐦𝐳𝐚 | Network Engineer (Aspiring) (@Hamzaonchain) reported 43 minutes ago

  Here's a summary of what happened in case you didn't hear about it... A cyber extortion group called FulcrumSec (active since late 2025) hit Novo Nordisk the company behind blockbuster drugs like Ozempic and Wegovy pretty hard. They snuck in back in March 2026 through a compromised GitHub access token, roamed around for over two months, and walked out with roughly 1.3 TB of data across 700,000+ files. Now they've started leaking a 264 GB sample publicly. Inside? Source code, proprietary formulas for pipeline drugs like Amycretin (their next big obesity hope), clinical trial records, employee and patient data, manufacturing details, and even private internal AI models for drug discovery. The hackers straight-up roasted Novo's security, pointing out laughably weak hardcoded passwords like "novo123" and "p_assw0rd" in critical systems. After Novo reportedly turned down a $25 million ransom, the group decided to start dumping samples and shopping the rest around privately. Novo confirmed a limited breach in early June involving some pseudonymized patient data from trials. They say there was no major operational disruption and that they're working with experts but this feels like a massive wake-up call for the entire pharma industry on basic security hygiene. Crazy how a simple token slip-up can expose billions in IP and sensitive health data. What's your take — do you think this will finally force better security practices, or is it just another headline that'll be forgotten in a few weeks?

• 
  Grishin Robotics (@GrishinRobotics) reported 43 minutes ago

  AI made coding faster. Devplan raised $2.5M to fix the coordination drag that shows up after the code is written. AI2 Incubator led the seed round, with Acequia Capital, Mighty Capital, Grand Ventures, and eLab Ventures participating. Chris Bee and Anton Safonov are building Weaver, a product knowledge graph that connects GitHub, Jira, Linear, Slack, Notion, Google Workspace, meeting notes, and customer feedback. The pitch is that product and engineering leaders should not need another status meeting to learn what changed, what slipped, or why a decision was made. This is a different wedge from coding copilots. Devplan is going after the organizational memory around the code: requirements, risks, decisions, blockers, and customer signals. The company says early users save eight hours a week on coordination, and its own benchmark answered moderately complex queries almost 2x faster and more than 3x cheaper than a standard Claude plus MCP setup. Quick facts👇 ● founders: Chris Bee; Anton Safonov ● total capital raised: $2.5M disclosed ● HQ: Seattle, Washington ● Investors: AI2 Incubator; Acequia Capital; Mighty Capital; Grand Ventures; eLab Ventures The next productivity bottleneck may be less about code generation and more about whether teams can keep shared context intact while AI speeds everything else up.

• 
  Deepesh Kalura (@KaluraDeepesh) reported 43 minutes ago

  Filed as GitHub issues: #336: Phone operators need stable unique IDs (not just phone number) #337: Auto-heal sticky assignments when a node dies Future imp task

• 
  severe engineer (@severeengineer) reported 46 minutes ago

  since github copilot onward leetcodes have become even more disconnected from how we all write code every day problem is any kind of standardized replacement probably ends up looking basically the same lol

• 
  Financial Programmer (@RBiancoUS) reported 50 minutes ago

  A dose of reality for end of week. My biggest question is I can't find any reason for the $Gold panic- did they find gold is causing cancer or radioactive? Selling looks like sheer panic. Would you believe someone asks in DM, so how did *you* get so many followers. Then he lets me brew on it for a day and comes back, I was joking do you have a github, presumably to get some code. No wonder I worked alone. I'm challenged socially guess not alone. After a night of 3 scammers one from Nigeria, one Africa. I need to lock dm down or find a way to restrict

• 
  David O. Ehibor 🇦🇷 (@grayontop_) reported 53 minutes ago

  GitHub Copilot didn't make developers faster It made slow developers more confident about writing bad code quickly 😭

• 
  AI Crave (@wecraveai) reported 53 minutes ago

  Open source NotebookLM alternative with no data limits and AI agents. Same idea as Google's NotebookLM. Same chat-with-your-docs. Same podcast generator. Same cited answers. Except this one has no source limit, no notebook limit, no 200MB file cap, and no Google login. It's called SurfSense. Google NotebookLM vs SurfSense: - Sources per notebook: 50 to 600 → Unlimited - File size cap: 200MB and 500K words → No limit - LLM choice: Gemini only → 100+ models via LiteLLM - Local LLMs: Not allowed → Full Ollama and vLLM support - Self-host: No → Yes, one Docker command - Price: $0, $19.99/mo Pro, or $249.99/mo Ultra → $0 forever Here's the wildest part: It connects to 27+ sources Google can't touch. Notion. Slack. Linear. Jira. GitHub. Discord. Dropbox. OneDrive. Gmail. Confluence. Obsidian. ClickUp. Microsoft Teams. Airtable. Your entire work life, indexed once, searchable from one chat box. 14.4K GitHub stars. 1.4K forks. 6,232 commits. Apache-2.0 license. One honest note: the README says it's not yet production-ready and still being actively developed. But it already does more than NotebookLM does, and the gap is widening every release. This is what NotebookLM should have been from the start. Repo in the first comment.

• 
  David Cramer (@zeeg) reported 53 minutes ago

  @shansmithnz I haven’t been using it but mostly because 1) laziness and 2) I didn’t find the remote sync pleasant in practice I switch PCs too much right now so mostly relying on GitHub issues as artifacts