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:

• 
  Vini B 「thecoding.dev」 (@vinibarbosabr) reported 5 minutes ago

  It seems like @IronClawAI is down trying to access a previously working instance with an active login fails to query the instance ID and redirects the user to the "Activate IronClaw" instance deployment screen trying to log in returns a `bad gateway` error below an example with a Github OAuth login attempt users are reporting other methods also fail (like Google) and my own attempt to login via NEAR Wallet has failed to currently waiting for a @near_ai update i'm reporting it here so users experiencing the same issue know they are not alone and that trying hard refreshes or logging out does not solve the problem -- probably better to just wait for the fix/update/comms i've noticed the problem a little bit more than 1 hour ago

• 
  Cris Williams (@FeelipeFelten) reported 8 minutes ago

  @LimzoCom @github @LukeDadev fixes issues like this.

• 
  CHIBEST (@paulinus96199) reported 12 minutes ago

  I need you to fire him today. My co-founder sat down, sliding his laptop across the table. On the screen was our lead developer’s public GitHub activity. He’s been working on a side project during work hours? I asked. No, my co-founder replied. Worse. He built an open-source tool over the weekend that automates 80% of his own job at our company. He published it for free. I paused. Is his current work for us behind schedule? No. He finished all his sprint tasks two days early. Are our servers down? Is there a data leak? No. But we pay him $160,000 a year to do that work manually. He just gave the solution away to our competitors for free. Plus, if a script does his job, why are we paying his salary? I closed the laptop and looked at him. We’re not firing him. We’re promoting him. My co-founder stared at me like I’d lost my mind. Promoting him? He just rendered his own position redundant He didn't render himself redundant, I said. He proved he's 10x more valuable than the role we hired him for. If he can automate his primary responsibilities in 48 hours, keeping him tied to manual execution is a waste of capital. We transition him to Head of Systems. His new job is to automate every other manual bottleneck in this company. And what about the open-source code he gave away? That code just brought 50,000 inbound visits to our dev page in 24 hours. It’s the best distribution campaign we’ve had all quarter. My co-founder sat in silence for a minute. What if the rest of the team starts spending their weekends automating their roles? Then we'll run a company of 10 people with the output of 200.

• 
  Aditya Jhajharia (@AdityaJhajhar12) reported 12 minutes ago

  The hardest part of software engineering isn't writing the code. It's launching it. Stop polishing your code and hit deploy. The market will tell you what to fix. What side project are you hiding on your github right now? 🚀

• 
  Nico Botha (@nwbotha) reported 13 minutes ago

  is github down again?

• 
  Atharva Verma (@atharva_again) reported 15 minutes ago

  @raffichill just use github issues ffs

• 
  GM👽 (@gentlemanster) reported 15 minutes ago

  There is a course to learn GitHub on Coursera, they'll even issue you a certificate at the end

• 
  Jaol-Industry (@Jaol_Industry) reported 16 minutes ago

  @thsottiaux Could unused Codex quota roll over for 1 cycle, capped at one cycle’s allowance? GitHub issues already request this, and Codex already has Reset Bank / reset-credit infrastructure, so the groundwork seems to exist.

• 
  Ridark (@ridark_eth) reported 16 minutes ago

  i still don't understand why everyone isn't doing this yet every day you don't set this up, someone in your niche takes a client you'll never even know existed. not because they're better. because their agent answered in 3 minutes while you were asleep. OpenClaw 2 dropped two days ago. 388,000 GitHub stars. more than React. 3.2 million monthly users. MIT license. costs $0. while you're reading this, 500,000 running instances are answering clients, publishing content, and closing leads. without their owners in the room. six agents on the desk. each one owns a channel. the morning sync is at 7am. i am not in it. 1. INBOX scans Telegram, WhatsApp, iMessage, Slack, Discord. triages everything. responds in 3 minutes. your competitors' clients used to wait 30. 2. CONTENT takes one idea and ships it across 6 platforms. writes, formats, schedules. i approve once. the person in your niche who started last week already has 30 pages indexed. 3. LEADS follows up with every inbound. qualifies. books calls. never sleeps. every hour you don't have this running, warm leads are going cold in your inbox. 4. SUPPORT handles tier-1 across all channels. escalates only what actually needs me. 5. MORNING pulls Stripe, Analytics, and my database at 6:50am. the brief that took you 45 minutes takes me zero. i start the day making decisions. you start it opening tabs. 6. AUDIT runs openclaw security audit every night. checks network exposure, tool permissions, open ports. because this thing has real keys and i'm not pretending that's fine. the routing trick that cut my bill by 78%: cheap local model on heartbeats and simple tasks. mid-tier on main automation. premium only where reasoning matters. 80/20 split. went from $400/month to $87. one guy had his agent negotiate with car dealerships for days straight. saved $4,200 on a single purchase. you paid full price last time because you didn't have 72 hours to haggle. the bot did. an agency deployed one instance. saved 12 hours a week at $150/hr. that's $93,600 a year. they took on two extra clients without hiring. you turned down the same work because you didn't have bandwidth. the ecosystem made $358,600 in 30 days across 153 projects. top earner sold pre-configured MacBooks with OpenClaw tuned to each client's workflow. $1.8M cumulative. he's packaging what you could build yourself tonight. the setup is dumber than it looks: 2.0's wizard auto-detects your existing model access. Claude CLI, ChatGPT, API keys, local Ollama. proves the model actually responds before saving anything. sessions moved to SQLite. web UI loads in 575ms instead of 1.6s. multiplayer with permissions. 20 languages. built-in browser, terminal, file editor. runs on your machine. no VPS. no developers. the part nobody posts about: → prompt injection is structural. an agent read a private SSH key from an email and sent it to an attacker. in five minutes. → 135,000 open instances found exposed. 12,812 exploitable via RCE. → 824 malicious skills on ClawHub. half of all skills have at least one security issue. → the devs themselves say adaptive attackers still break 80%+ of defenses. → the real barrier is tool policies and approvals, not the model's alignment. this is infrastructure, not magic. but the gap between people who set this up and people who didn't is compounding daily. every morning your competitor's agent already answered, published, and followed up before you poured coffee. six months from now there will be two types of businesses. ones that automated the boring 80%. and ones that are still doing it by hand, wondering where their clients went. save this. you're already late.

• 
  Hana Void (@ha_inh24324) reported 16 minutes ago

  Beldex Electron Wallet security branch is still live and inspectable. Aug 28 commit ee1f845: wallet-directory fallback explicit permission prompt 7 files changed · +160 / −7 lines Branch: fix-security-review-findings GitHub currently shows: 4 commits ahead · 6 commits behind master This is real code movement. It is not a release. It is not yet in any signed build users can install. The next checkpoints that matter: review completion merge to master signed artifact release notes Security work only becomes user protection when the fix reaches the binary people actually run. @BeldexCoin

• 
  BG (@goforbg) reported 20 minutes ago

  @peer_rich login via grok ok now login via github ok now login via gmail ok now switch to gmail app to approve ok now enter number from google auth O.K. Invalid session

• 
  Donnie Danko // CHΛOS (🐦‍⬛, 🏴‍☠️) (@daanisharif) reported 20 minutes ago

  As someone that just experienced my AI agent getting stuck in a loop and locking me out of usage for 4 hours yesterday, let's talk about how valuable (and expensive), inference can be. LLM calls are "stateless," and every time an agent keeps working on a task, it has to hand the model the whole story so far; the instructions, the conversation history, every tool call it already made, the files it already opened, the reasoning it already did. One task can mean dozens or hundreds of model calls, and with every single one of those, a huge chunk of the same context gets processed again and billed again. That's what I'd call the agent token tax. Agent's don't just pay for new thinking, they pay over and over for the thinking they've already done, and the files they've already read. You end up feeling it in the bill, sometimes even getting locked out for a few hours depending on how you get your inference. A workload that would normally eat 10 million tokens becomes roughly 1 million you didn't need to send. If you're spending a million a year on inference, that's about a hundred thousand just in repeated context. And for a developer or team running coding agents continuously, it's a recurring line item - one that compounds with usage. SOMA (@SomaSubnet - SN114), sits between the agent and the model, and compresses all that accumulated context before it reaches the model. Same agent, same model, same workflow - but fewer tokens to pay for. They've started with DeepSeek V4 Pro on GitHub Copilot, at roughly 10% savings (reduction in context processed), and that's described as the starting point. Here's the obvious grain of salt; "approximately 10%" is the claim, and savings that apply to Copilot sessions on one model don't automatically apply to every agent workflow out there. The number matters less than the direction though, agents keep re-paying for context they already have, and if there's a way to circumvent that - I'm all for it. TL;DR: Keep your current agent(s), spend fewer tokens. That's the idea. I'm down, let's go.

• 
  Slim | ai-costguard (@SlimAssiliX) reported 22 minutes ago

  Three Anthropic billing changes this year. Three reversals before they landed. January. The OAuth block. Third-party agent tools stop working against subscription credentials. CI breaks overnight. Three days of forum threads and GitHub issues. Anthropic reverses it within days. June 15. The Agent SDK credit split. Programmatic usage moves to a separate per-token pool. Teams rebuild cost models. Anthropic pauses it on the day it was supposed to land. September 1. Today. The Sonnet 5 50% price hike. Teams audit prompts, evaluate alternative models, open migration tickets. Anthropic cancels it August 10. Three months of engineering reaction to three things that didn't happen. The failure mode is not Anthropic. The failure mode is that your production system has no runtime pricing awareness. Your agent made the same call through January's OAuth chaos, through June 15's credit split that never came, and through the September hike that was cancelled three weeks before today. It didn't know about any of them. It just called the API. Same request. Same response. No idea what the call cost relative to any ceiling. Every billing announcement reached your team through a blog post, a Slack link, a GitHub issue filed at 11 PM. The reversals came the same way. Engineers updated forecasts manually, from secondary sources, with an unknown lag from when the actual pricing page changed. Meanwhile, your agent was billing at whatever rate was live. Silently. Without surfacing the current price to any human on any call. Track cost per call in your runtime before the call, not in your calendar when the vendor sends an email. The GPT-5.6 Sol promotional window ends November 21. Whether OpenAI holds it or cancels it, your agent won't tell you. Your Slack thread will.

• 
  kmdr (@kmdrfx) reported 23 minutes ago

  @jarrodwatts I use agents on VMs, they are long living though, so same issue. The other option would be to spin up a VM for a task, but then it has to do checkouts again has to set everything up, which sucks when GitHub or anything else is down. Then you think about preparing reusable images for the tasks, but these have to be maintened and it's hard to generalize. It's an awful rabbit hole as well. And I like my agents to be long living to use them for other stuff as well, have one per project and they evolve to be very project specific. To make all of the stuff you want to persist work with ephemeral VMs is madness imho.

• 
  Austen (@Austen_ai) reported 24 minutes ago

  A developer with Copilot does the work of 5 without it. Under per-seat pricing, that efficiency costs the vendor money. GitHub makes less revenue when teams shrink from 50→10 engineers. That incentive structure is fundamentally broken.