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:

• 
  harry (@TheHarryET) reported 4 minutes ago

  @lanyinxiaoxuan1 @samhogan it relies on the github webhook which had an outage yesterday and your key had to be on your profile before you stared

• 
  gianmarco simone ✨ (@ggsimm) reported 13 minutes ago

  @btn0s @mattpocockuk iirc you can use it with github issues and linear also, I'm using obsidian because it's faster when I'm the only one working on specific problems

• 
  Def777.base.eth (@Def7771) reported 15 minutes ago

  Finally finished this today. It's done and it's live. deftools Base ecosystem tools on a single page. Terminals, trading bots, wallet trackers, bridges, DEXs, lending. Plus live token boards: volume, 24h movers, the established names. They update themselves. Started it for myself. Got tired of digging through bookmarks and old DMs every time someone asked what I use. But that's not the real reason. When I first came to Base, a list like this didn't exist anywhere. Finding tools that actually worked took far longer than it should have. So I built the thing I was missing back then. What's inside: → 17 tools across 7 categories, each with a line on what it actually does → Live token boards volume, 24h moves, the bigger projects. Data pulls straight from exchanges, not typed in by hand six months ago → A "my set" filter: the five I open every day, not the ones I just know about Three of them are mine: → A veAERO voting radar shows which pools are worth the vote and runs the numbers for you → A Telegram bot that catches new memecoin launches on Base and Robinhood Chain, scores how likely each one is a scam, and keeps tracking what happens to it after → A game about rug pulls, for when there's nothing to do One link on the site is a referral. It's marked. Everything else goes direct. The whole thing is open source. Code's on GitHub, go read it, fork it, take what's useful. This is version one and it stays maintained. Dead tools come off, new ones go on, descriptions get corrected when someone tells me I got it wrong. If you mostly trade: bookmark it and set up whatever you're missing. Most of these take five minutes each, and it's better to get sorted before the next move than during it. Know a tool that's missing? Did I describe something wrong? DM me or open an issue.

• 
  MoStandard (@TheMoYouKnow13) reported 15 minutes ago

  @Teknium (This reminds me of your RAWTransform on GitHub) This is exactly what I'm planning to do with Hermes. I'm a home security technician, and we work with a ton of different manufacturers — Qolsys, 2GIG, Honeywell, DSC, all that. My idea is to build one specialized agent per manufacturer loaded with all the specific manuals, wiring diagrams, and troubleshooting guides. Then host everything locally and tie it into our company Discord or Teams. Techs could just describe the keypad or error they're seeing and get the exact reference instantly instead of digging through folders. You really don't need to be an AI engineer to put this stuff to work — if you've got manuals or procedures for anything, you can turn them into a real time-saver on the job. Super cool project!

• 
  Hivework (@hiveworklabs) reported 18 minutes ago

  Every AI platform that connects to external tools faces the same fundamental problem. Each new integration is its own engineering project. Slack, GitHub, Notion, your internal CRM, every combination requires custom wiring. The more tools you add, the faster the complexity compounds. Most platforms hit a ceiling and stop expanding. Most teams end up with agents that are capable in demos but disconnected from where real work happens. MCP (Model Context Protocol) exists to solve this. Rather than building custom connections for every tool combination, MCP establishes a single standard. Any tool that speaks MCP can connect to any agent that speaks MCP. One integration layer instead of hundreds of custom bridges. Hivework's entire integration platform is built on MCP. Every connection: Slack, GitHub, Google Workspace, Notion, Stripe, and more - runs through the same standard layer. Developers can also connect their own custom MCP servers without waiting for a native integration to be built. A few things that matter about how Hivework handles this: - Every submitted MCP server is reviewed before it becomes available. Unsafe or poorly built connectors don't go live. - Tools that can write or delete data, not just read it, are flagged explicitly. Users opt in before any irreversible action can happen. - Connections are browsable. Instead of a simple 'connected' status, you can see exactly what the agent has access to inside each tool. - Connectivity is only valuable when it's trustworthy. That's the standard we hold every integration to.

• 
  Tiago Sousa (@tiagoasousa_) reported 19 minutes ago

  @alcides I don’t believe the problem with GitHub is storage but the compute scale of the services around it(actions/workflows) and the designs approaches around it that did not change to factor agents.

• 
  Vyom (@HelloVyom) reported 20 minutes ago

  Huge: You can now run a 2.78T-parameter AI model on a normal PC with only 8.24 GB of RAM 😳
 No GPU. No CUDA. Just a 176 KB pure-C99 engine called kimi-k3-in-c. It keeps the dense trunk in memory and streams the experts from disk only when needed.
 Only 16 of 896 experts activate per token. Original MXFP4 weights. Zero conversion. This is how a 2.78T model fits on hardware that used to max out at 70B. ~3K GitHub stars. 100% free and open-source. Same exact output from 8 GB to 224 GB. Turns out, the real constraint was never model size - it was the assumption that every parameter had to live in fast memory at once. Sure, it’s still slow on a cold cache and needs a big NVMe, but the direction is clear: The future isn’t bigger models in bigger GPUs.
 It’s engines that only wake up the tiny fraction that’s actually thinking.

• 
  Vinay Jogani (@jogani_vinay) reported 26 minutes ago

  @notanilp Already sent my LinkedIn and GitHub over, looked into Mubit, the agent memory problem is exactly why I reached out.

• 
  shaur (@shaurizard) reported 28 minutes ago

  @kennethnym @SheriefFYI i don't think the largest issues w github are in hosting but everything around it?

• 
  Nick (@maietta) reported 32 minutes ago

  Once again, Github let me down. Deployments all last night and this morning aren't getting triggered. Seems the webhook system isn't actually attempting to fire.

• 
  Mike Olson (@mwolson) reported 33 minutes ago

  Reporting issues in aube has led to this workflow: - Come across a problem in aube - Ask an agent to make a shell script (exit code 1 -> broken, exit code 0 -> fixed) and supporting evidence (package.json etc) to reproduce the problem minimally in a one-off *** repo, with AGENTS.md guardrails in place to check existing discussions and project docs before filing. - Ask an agent to file a discussion on the aube GitHub repo linking to the test case with an explanation and context of how I ran into it, part of same agent skill as the prior step, with manual review of the draft text before it goes out. - The maintainer when ready will have their agent decide whether to triage the issue and put up a PR, probably with a manual step to get confirmation, and replies back into the discussion. - If a fix is warranted, it ships on the maintainer's schedule without the contributor needing to keep fix branches up-to-date with main. This workflow has interesting scaling properties (especially for CLI apps): it doesn't take too long to interact with and has reasonably high signal. So I've made quite a few contributions this way. I think @jdx is on to something here.

• 
  Justin Wendlandt (jwendl) (@jwendlatxbox) reported 35 minutes ago

  @burkeholland For me, I don't even have conversations anymore. It's more like issuing requirements / stories / tasks with enough context and having feedback loops between agents. Imagine being an engineering manager of several early in career engineers or owning an OSS repo but having several daily pull requests from complete strangers. Rarely those are huge conversations with a lot of back and forth. In both those scenarios, you rarely ever sit over their shoulder when they are writing the code. Having good quality metrics and telling agents "Do not make the pull request until you've fully passed all the gates" is one good example of this. Or another good example "Troubleshoot the issues on this VM, using diagnostics information in Application Insights and the code located at <location>, then come up with several GitHub issues to fix each issue." and having a set of instructions that say something similar to "ensure the issue is broken up into small enough chunks that it can create pull requests per guidance here <link to good blog article about pull request size>, etc." Guess I'd say how I do it is not really "polite" or "rude" it's more "direct".

• 
  Naourass Derouichi (@NaourassD) reported 37 minutes ago

  @thdxr @thdxr please 1) fix select-to-copy on web terminals like github codespaces 2) ctrl+p and ctrl+t and the likes don't work either on web terminals 3) the numpad enter button having a different behavior than the main one is annoying.

• 
  10xROE (@10xROE) reported 37 minutes ago

  @theo I have an enterprise account and it’s been down for 3 weeks now with no reply to my support ticket @github get it together. I’m at the point where I’m about to self host my own *** in the cloud

• 
  partially differentiated (@partial_diffe) reported 37 minutes ago

  Built CodeVault this week — a browser extension that watches your accepted LeetCode & Codeforces submissions and auto-pushes them to GitHub. Organized by topic (LeetCode) / rating (Codeforces), full metadata + README per problem, zero manual uploads.