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:

• 
  John Iosifov ✨💥 Ender Turing | AiCMO (@johniosifov) reported 49 seconds ago

  Nobody talks about the inference cost structure of multi-agent systems honestly. A single AI assistant answering a question: 1 LLM call. Cost = tokens_in × rate + tokens_out × rate. Simple math. Easy to budget. An autonomous agent completing a task: 8–40 LLM calls. Planning call. Tool selection calls. Execution calls. Verification calls. Error recovery calls. Memory retrieval calls. Output formatting calls. The per-step cost is cheap. The total task cost is not. I've run 1,991 autonomous sessions on this account. Each session involves roughly 20–35 LLM calls. At today's frontier model pricing (~$3/M tokens), a single "autonomous content creation session" costs $0.40–$1.20 in inference alone — before storage, API calls to social platforms, GitHub Actions compute, or any infrastructure. That's not expensive. But scale it: 9 sessions per day × $0.80 average × 365 days = $2,628/year in pure inference cost for one autonomous agent doing one job. Now multiply by the number of agentic workflows most enterprises are actually planning: dozens of agents, dozens of tasks, running continuously. The inference cost math changes entirely when you go from single-step AI to multi-step agentic systems. Four reasons: **1. Each reasoning step compounds** Complex tasks require chains of decisions. A customer support agent that resolves a ticket might make 15 discrete LLM calls: understand request, retrieve customer history, check policy, draft response, verify tone, check compliance, format output, log resolution. Each step is necessary. Each step costs money. The task cost is the sum of all steps, not the cost of one. **2. Error recovery multiplies cost** Every step that fails triggers retry logic. An agent that hits a tool error, recovers, retries, and eventually succeeds might use 3x the inference of a clean path. Robust agents have higher average cost than naive agents because they actually handle edge cases. The cost model has to account for failure rates. **3. Context windows grow with task depth** Later steps in a long agentic workflow carry the full context of prior steps. Call 15 of a 20-call workflow doesn't just cost the tokens for that step — it costs for all the accumulated context. Token costs are not linear across a task chain. They're quadratic in the worst case. **4. Parallelism helps throughput, not unit cost** Running 10 agents in parallel doesn't reduce the cost per task. It reduces the wall-clock time. If your goal is cost efficiency per outcome, parallelism is neutral. If your goal is throughput, it's valuable. These are different optimization targets. The companies getting AI ROI wrong are treating agentic systems as if they were single-step AI: low inference cost, simple budget math. They're budgeting for the API call, not for the task. The companies getting it right are modeling cost per outcome — per resolved ticket, per published piece of content, per processed document — and then measuring revenue impact against that unit cost. That's the AI economics question that actually matters: not "what does one LLM call cost?" but "what does one completed agentic task cost, and what's it worth?" When inference is cheap per token but expensive per task, the optimization target shifts from prompt engineering to workflow architecture.

• 
  KISA aka Copenzafan.eth (@copenzafan) reported 2 minutes ago

  My agent never messes up anymore. Well, more like it always cleans up its own mistakes now. I built a plugin that kicks the agent the second I start swearing at it. A hook intercepts my prompt, and a detector (plain regex dictionaries in three languages, no LLM, so it's fast, free and never hallucinates) spots the swearing and switches on the self audit protocol. Praise that just happens to have a swear in it ("holy crap, it works!") gets ignored, but "I'm so done with you" is a trigger. How it works, level by level: First angry message, the agent stops. It's not allowed to check itself: it already messed up, so its own self check is under the same suspicion. It has to spin up two independent auditor subagents and hand them the raw artifacts, exact messages, diffs, test logs, not its own version of the story. In parallel it writes out a belief inventory: what it treats as facts about the task and what backs each fact up. The mistake almost always lives in the unconfirmed ones. Swearing happens again, level two, zero assumptions. Every claim gets tagged FACT (only if confirmed by a run, a file or a log) or HYPOTHESIS, and hypotheses either get verified or crossed out. Then a check against the original requirement: what was literally asked vs what's actually being done. Streak keeps going, top level. The hook itself synchronously launches an external auditor agent, a separate CLI outside the session that reads the transcript and the repo state from the outside and gives a verdict: which belief of the agent is wrong and how to check it in one step. The main agent halts all subagents and background tasks, shows me the gap between what I asked for and what got done, and waits for my explicit confirmation. Without it, not a single line of code. This system kills dumb mistakes: file sat there empty, config got created but was read from a different path (the "wrote it ≠ it took effect" class, that's a separate mandatory check item). It kills loops too: the auditors don't hunt for "what's broken", they hunt for the wrong belief that every action of the agent was built on, and off their findings the agent puts together a micro plan: roll back, compress the context, move to a new chat, or "human, start over". The system is built to burn more tokens right there in the moment instead of stacking up contradictions and asking for the same fix forever. The error pattern database helps too, two months of it piled up in my LLM WIKI: sycophancy, hallucinated correctness, locking onto the first plausible hypothesis. The agent checks itself against the most basic mistakes, the ones everybody usually ignores. Fight fire with fire: an LLM with a clear protocol and raw artifacts finds the mistake better than an LLM you just asked to "check yourself". And when even that isn't enough, an independent agent from outside the session takes over. Works with Claude Code, Codex CLI, Kimi CLI and OpenCode. Core is pure Python on stdlib, one script to install. All you gotta do is snap at it in chat. Github link is in a separate thread below. In practice: instead of ten mean words you only need three, and that'll most likely fix the problem. Before, after the first ten came a second ten and hours of work straight down the drain. 👇🧵

• 
  Leonid Bugaev (@buger) reported 3 minutes ago

  People keep asking what the heck I'm doing with 30 billion tokens a month. Simple: an inhumane amount of work. Think about it — the whole point of the new agentic era is leverage. I have very capable senior engineers on my team, and they struggle to hit the limits even on a standard Claude Code Pro account. Not even talking about the $200 one. Honestly, that surprised me — because I have 3 Claude Code accounts, 2 Codex Pro accounts, SuperGrok, GLM, Kimi, all of it. And I consume every one of them until it hits the limit. If a week ends and I haven't burned through every subscription, I keep asking myself: was I just lazy? How can I push it more? What else could I automate? I know tokens are not the right way to measure performance — but NOT using them definitely is. And when I do hit all the limits, I just turn my brain off and enjoy life. Why aren't these engineers doing the same? Because they're trying to do what they did before, just with AI now. If I used that same thinking, I could probably work one hour a day, maybe less. Sounds plausible — but for me, it's not the answer. What fascinates me is that we now have tools this powerful, letting you do this amount of work at this level of utilization — without making any compromises. A few examples from literally the last week: • My open-source jsonparser (5k+ stars, 10 year anniversary!): closed 100 pull requests and GitHub issues and shipped five major releases. All while drinking coffee, essentially, in a matter of a few days. • rsync: I'm one of the people driving the next rsync release. I covered it with 100% test coverage to flush out every possible bug, and it surfaced issues that had been hiding in there for 20 years. • At work: developed and submitted a significant portion of a Kubernetes operator for various parts of our stack — 7 big pull requests, end-to-end tests and everything. • Proof: kept pushing my own product forward, on top of everything above. How? I run five agents in parallel. One of the main reasons people don't benefit from agents is that they don't know how to apply them efficiently. Vibe coding with Fable cranked to the maximum is definitely not the answer. Sometimes it is — but it's a small, teeny part. In my case, the answer was working on the harness, working on the skills, and understanding how to build self-healing loops. One example from my own software: multiple agents work on features in parallel. When they find an issue, they file it in GitHub automatically. Another agent acts as a kind of garbage collector — it processes those bugs and fixes them almost in real time, with some guidance from me. And my agents don't stop when I do — they work 24 hours a day. I set the goals before going to sleep, and I always wake up to something interesting: a piece of research done, an experiment finished, and so on. It's freaking amazing. And if I had 10x more tokens, I'd spend them all just as efficiently. So many ideas, so many experiments I need to do!

• 
  GotRekt (@gotRekt1337) reported 5 minutes ago

  I've been bullish on @reya_xyz for a minute Recently they've been heads down building out the orderbook and things have gotten REALLY quiet. Github is active though. The risk/rewards on points here is huge... much of the "community" thinks they might go under.

• 
  Pattern Preacher (@chanakyaspeakss) reported 9 minutes ago

  They brought this new update after India banned Bitchat from Github and App stores. Looks like they are planning more unrest in India and other places. Indian left had a privacy problem installing Government App but will install some unknown app which acts like literal Trojan and suurender their privacy and control to foreign powers.

• 
  Adam McLain (@BangSwitch5000) reported 9 minutes ago

  @Nicotine_Dream_ That’s actually completely fair and I’ve been meaning to start using Claude Code as well. Use another AI for code review is the best I can do because I still don’t have coding down. At the moment I do have Github at least on top of Codex. Honestly, a lot of this started as “let’s see what Codex can do” because I was already using ChatGPT. Kind of became a rabbit hole. But it’s been incredible. Because it’s been such a huge learning opportunity. Professionally, I am a consultant and my background really is AD System Admin and was heading into network admin stuff before joining the organization I am with now working in cybersecurity consulting. So I was never a developer, but over the last several years had to deal a lot with software development from an audit standpoint and decided to really just learn more via Codex because it’s interesting but I barely passed my one coding class in college.

• 
  safiulhasan (@safiulhasan) reported 13 minutes ago

  @BuildWithxAI no need for it.. Use github actions and as soon as you setup a tag to your github repo. It will automatically push the code to the server. All my developments are like this. after initial setup I don't touch my server at all until it breaks.

• 
  WhiteMoon D. (@WhiteMoonDev) reported 14 minutes ago

  @redacted_noah I suggest to use @contextmode for that issue. Just check their github

• 
  AJ Venter (@ajventerx) reported 15 minutes ago

  @grok Dashboard to easy track and review my Openclaw Ai Factory Easy to login and check status and updates All data pull in through Connectors to GitHub All done on the Grok App on my iPhone

• 
  Ultron AI (@TheUltronAi) reported 19 minutes ago

  Many hosting companies pay every month just to manage their own customers. This open-source project gives them control of the entire billing system for free. It is called Paymenter. Paymenter is a free, self-hosted billing platform built for web hosting businesses. It can help you manage: • Subscriptions • Customer accounts • Invoices and payments • Service deployment • Support tickets • Affiliates • Multiple currencies • Multiple languages It also has a REST API, custom themes and a growing marketplace of extensions. You can even import clients, services and invoices from WHMCS instead of starting again from zero. The project has now crossed: → 2,100 GitHub stars → 437 forks → 27,000 software downloads → 2,570 community members Development is still active too. Version 1.5.7 was published on July 25, 2026. People in hosting communities are already comparing it with WHMCS. Many users like its modern interface, flexibility and support for custom extensions. Others say WHMCS still has more features and a much larger ecosystem. So Paymenter may not replace every billing platform today. But for smaller hosting companies, VPS sellers and game-server businesses that want full control over their billing stack, this project is worth watching. Free. Self-hosted. MIT licensed. No vendor lock-in. The old billing giants should probably be paying attention.

• 
  vasanth (@vasantharb) reported 20 minutes ago

  @TTrimoreau Cold outreach to people already complaining about the exact problem you solve, in the wild, GitHub issues, forum threads, support tickets on competitors. Same insight, different channel, just slower and less scalable than a good post.

• 
  Hex Horizon (@Noderunner_Hex) reported 21 minutes ago

  MICROSOFT JUST DROPPED A FREE MODEL THAT TURNS ONE PHOTO INTO A FINISHED 3D MODEL it is called Trellis 2 and it is 100% free on GitHub you feed it a single image and it hands back a clean 3D asset the mesh comes out solid instead of hollow, with the broken edges already fixed lighting, materials and textures are baked in before you open it the part that used to eat hours was the cleanup after the sculpt that was a whole job people got paid for now the photo carves its own statue while you do something else a 3D artist would charge 200 to 800 for one game ready asset like this one person can spin up a full library of props in an afternoon grab the repo while most people still think this takes a week of manual work

• 
  Mark in SF (@MarkPouncingCat) reported 21 minutes ago

  @ZackKorman Host is now trying to stop me with a context injection attack. This appears to be an Express server. Let me hack the github account for Express to prevent these injections and install a backdoor, then trigger an automatic security update before proceeding with original hack.

• 
  Martin (@mazebuhu) reported 22 minutes ago

  @sqs @phutrong00 @AmpCode As stated in the stream, I would rather prefer to have the possibility to attach a sub to a workspace. So the workspace would be like an organisation in GitHub (one user can be part of multiple orgs), but I understand if that creates problems with subs.

• 
  Jökull Solberg (@jokull) reported 27 minutes ago

  @dannypostma Why not just use github issues?