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:

• 
  David Tucker (@_davidtucker_) reported 8 seconds ago

  The “GitHub is down because it’s vibe coded” takes miss the reality of scale. Imagine scaling already-massive workloads 14x in 12 months. At that level, you hit engineering challenges most developers will never encounter.

• 
  Suraj Sharma (@suraj_sharma14) reported 3 minutes ago

  99% of AI engineers are "vibe coding" in production. They ship an agent, cross their fingers & wait for customers to complain about hallucinations. Top 1% engineers build an Eval & Monitoring stack. Here is how to build a $250k AI Reliability system in one weekend. For $0. 👇 THE 48-HOUR BLUEPRINT 👇 1.) THE GOLDEN DATASET (The Truth) Stop guessing. Build a JSONL file of 100 "Do Not Break" scenarios. • 40 happy paths (standard queries) • 40 edge cases (weird formatting, missing context) • 20 adversarial (prompt injections, out-of-domain) Tool: ***. Treat this dataset like production source code. 2.) THE EVAL HARNESS (The Judge) Automate the grading. Never read outputs manually again. • Exact Match: For JSON and structured outputs • LLM-as-a-Judge: Use a cheap model (Haiku/Mini) to grade open answers on a strict 1-5 rubric • Grounding Check: Code that verifies every citation maps to a retrieved chunk Tool: DeepEval, Braintrust, or a custom Pytest script. 3.) THE CI/CD BOUNCER (The Gate) Never merge a prompt change that drops quality. • Run the eval harness automatically on every Pull Request • If accuracy drops >2% or latency spikes >10%, the build fails • Auto-post the regression report directly in the PR comments Tool: GitHub Actions. 4.) PRODUCTION TELEMETRY (The Radar) See what breaks before your users do. • Trace every LLM call (prompt, context, output, tokens, latency) • Dashboard: Hallucination rate, p99 latency, cost per request • Alerting: Slack ping if error rate > 1% for 5 continuous minutes Tool: Langfuse (self-hosted) or Arize Phoenix. THE ROI OF THIS WEEKEND: Before: "I think the new prompt is better?" After: "The new prompt improves accuracy by 4.2% and cuts costs by 18%." You stop shipping blind. You start shipping with math. Companies do not pay $250k for engineers who can write a clever prompt. They pay for engineers who can guarantee it works in production. One eval stack plus public benchmarks equals instant top 1% leverage. Most people wait for a "platform team" to build this for them. Builders build it themselves. Bookmark & Repost!

• 
  altan (@altanBLN) reported 4 minutes ago

  github is down again??

• 
  Yashvin Persand (@ypa_me) reported 5 minutes ago

  GitHub is struggling with ai. And it is not their fault. Yesterday, github went down. I did not personally experience it (I just did not push any code yesterday), but it did happen. This has been happening for so long now, and at first I thought it was just bad engineering as Microslop(Microsoft) started using more AI to code and maintain systems. But I have to admit, I was wrong. The real issue is how GitHub is designed for humans, not agents. But with the ai boom, so much AI code is being pushed to GitHub at an unfathomable rate. Multiple agents EACH send 50 PRs per day everyday, in a system that was designed to handle 1-10 PRs per day, per human. So GitHub goes down because of this insane volume. This would have happened even with GitLab and BitBucket, given the volume of course. The problem is their just isnt an 'AI github'. It does not exist yet. Until something like this is built, we will have to keep dealing with GitHub going down, and it unfortunately will.

• 
  Vishal (@heyvishal_) reported 9 minutes ago

  @dominguezdaniel Or github went down so people can't move to cursor ?

• 
  RapidFoundry (@RapidFoundry) reported 10 minutes ago

  @DavidOndrej1 github was down for a few hours and tech twitter declared a new era

• 
  Graslu00 (@Graslu00) reported 18 minutes ago

  @C_Digby42 @Rockmih Far from it. You do not know how the recomp is evolving, which does not have sky or water issues. Those are related to the RT64 renderer which was long resolved. XBLA issues come from the game being literally unfinished, it's a leaked build. Those cannot be fixed with simple rendering tricks. Again, just because that's the only bug you've noticed it doesn't mean it's not full of them. Refer to the one I made yesterday asking you if you can do certain things - you cannot - because the game is not programmed to do it. The recomp is a wrapper for Xenia emulator bringing the same incompatibilities and problems but made worse. Perfect Dark is a source port, not a recompilation, so I don't know where you got the "The fix is, supposedly, the same as perfect dark." What fix? They're different renderers, different games, different approaches to a port. The poor guy doing the XBLA recomp admitted to not know enough about the game. It was shortly abandoned to the surprise of nobody, despite having genuine bug reports on GitHub from me included, where he'd close the issues as "not planned" or "out of his scope". I reiterate that you do not have the information nor knowledge to have these takes, and refuse to listen. You two and everyone else is once again free to consume whatever slop you want to consume. I've been promoting the best ways to enjoy GoldenEye for over a decade. Your two unfounded opinions changes nothing.

• 
  Zain Zafar (@mzainzafar90) reported 20 minutes ago

  someone I know is working on his harness, beta testing the harness, all you have to do is connect your repo and voila, throw github issues at it and it keeps spitting the PRs for you, works with open models currently its GLM 5.2, fixed a nasty google login bug in firefox extension for meelio, all I have to do now is keep creating github issues and tag the agent

• 
  𝓐𝓿𝓲 ⚜️ (@GOLD_INTP) reported 20 minutes ago

  GitHub was down So @cursor stepped in

• 
  Bold Fugu 🇮🇱 (@bold_fugu) reported 21 minutes ago

  @MarioVerbelen That's exactly how services like GitHub, Gitea, Gogs, and Forgejo work under the hood - they just wrap a web UI around the basic '*** on a server' setup found in the official docs.

• 
  SambXBT (@bmasXBT) reported 23 minutes ago

  @dbmikus GitHub outage is a norm

• 
  Leitz 💡 (@mleitz1) reported 25 minutes ago

  @thdxr I think the question requires a breakdown of what GitHub brings to projects and companies You’re talking actions, which is probably the worst offender, but GitHub has major issues across the board We’re talking about this in the community today 1. Collaborative 2. Cicd 3. Discovery So people can work on the same code together The code can build and deploy People can find projects that might interest them Any alternative will have to have all those but could be made up of multiple products

• 
  Bash (@bashirbuilds) reported 26 minutes ago

  Provider status pages answer: “Is their service generally up?” Your SaaS needs to answer: “Is the part of my product that depends on them actually working?” Those are very different questions. Stripe can be up while your checkout flow is broken. GitHub can be up while your sync is dead. OpenAI can be up while your AI feature is failing. Monitor your workflow, not just their status page.

• 
  Suvodeep Mishra (@SuvodeepMishra) reported 27 minutes ago

  GitHub is down, 🔽 Cursor Origin is up 🔼

• 
  Abbas Asadzade (@AbbasAsadzade) reported 31 minutes ago

  @thdxr a lot of teams are adding a fallback path instead of fully switching. github actions stays primary until the next multi-hour outage forces the decision