Inter-office memo

To
Everyone whose GitHub Copilot stopped answering
From
Mo, Chief Executive
Date
November 19, 2026, 4:44 PM CST
Re
GitHub Copilot outages

What to do when GitHub Copilot is down: 9 ways to keep coding in 2026

Check githubstatus.com, then switch models. Seven of the 11 Copilot incidents there from October 20 to November 19, 2026 hit only some models, so another model or Copilot Fusion is the first fix, as GitHub advises, and six were over within an hour. For longer ones, Copilot CLI runs on your own key or a local model without GitHub, and Claude Code, Codex, OpenCode or Enjoy keep you coding on other plans.

Disclosure: Enjoy is our product, and it runs Claude Code, Codex and other labs’ agents, not GitHub Copilot. We checked every way below against its maker’s own pages on November 19, 2026, and we say where another way is the better pick, including the free ones that don’t need Enjoy.

The ways to keep coding, compared

When Copilot stops answering, the trouble is usually one model, not Copilot. We counted the incidents on GitHub’s status page from October 20 to November 19, 2026 that marked Copilot or its model providers as affected, or named Copilot in their title: there were 11. Seven named particular models, and six of those named models you can pick. In those six, GitHub’s advice was the same: “we recommend choosing another model,” and in five it added Auto or Copilot Fusion, which pick a model for you. Six of the 11 were over within an hour. The two longest, at more than six hours each, were in Copilot’s cloud agent, where about 1 to 2% of requests failed on average.

So the first three ways below cost nothing and fit most incidents. The rest are for the wide ones, and for work that can’t wait.

Way Cost, checked November 19, 2026 How soon you’re back Best for
Check the status page, and wait Free When GitHub recovers, within the hour for 6 of the last 11 incidents A short incident
Another model, or Copilot Fusion Included in your Copilot plan; Fusion Standard is on every plan Right away, if another model is healthy An incident on some models, the most common kind
A session on your computer, not in the cloud Included in your Copilot plan Right away An incident in Copilot’s cloud agent
Copilot CLI on your own key or a local model Your provider’s API rates, or free on your own computer Once it’s set up GitHub’s own service failing
VS Code’s chat with your own key or Ollama Your provider’s API rates, or free on your own computer Once it’s set up Staying in VS Code’s chat
Claude Code, beside Copilot Claude Pro $20 a month, or $17 billed yearly; Max $100; Ultra $200 Right away, in a new conversation A Claude plan you already pay for
Codex, beside Copilot ChatGPT Plus $20 a month for its extension; a free ChatGPT account in VS Code’s experimental Codex integration Right away, in a new conversation A ChatGPT plan, or none
OpenCode Free, with its free models or a local model Right away, in a new conversation No plan at all
Enjoy A monthly subscription, free for 7 days, with your phone and one computer (pricing) Right away, in a new conversation Several labs’ agents in one conversation

Prices are each company’s US list prices, before tax. GitHub’s plans page has Copilot Free at $0, Pro at $10 a user a month, Pro+ at $39 and Max at $100. Waiting, switching models, a local session, a local model and OpenCode’s free models cost nothing extra; the others add to what you pay for Copilot.

First, read the message

Not every stop is an outage, and GitHub’s troubleshooting page sorts the common ones:

What you see What it means What to do
A warning on a model in the model picker GitHub’s “automated model-warning system activated in-product warnings for affected models,” as it did on November 13 Choose another model, or Copilot Fusion
Error: You've hit a rate limit “This error suggests that you have exceeded the rate limit for Copilot requests,” and “Most people see rate limiting for select models, due to limited capacity.” Do what the message says: wait for the reset, retry at the time it gives, or upgrade
GitHub Copilot could not connect to server. Extension activation failed It “indicates that you do not have a Copilot plan, or there was an error connecting to the GitHub API” Check the status page, then sign out of Copilot in your editor and back in
Your AI credits are used up Your plan’s allowance, which isn’t an outage See our guide to running out of credits

The troubleshooting page’s first line is the rule for everything else: “If GitHub Copilot stops working, check GitHub’s Status page for any active incidents affecting GitHub Copilot or model availability.”

Check the status page, and wait

Best for: a short incident, which most are.

githubstatus.com lists two parts for Copilot: Copilot, and Copilot AI Model Providers, where single-model incidents show. Incidents name the models or features they affect, and the page has a Subscribe to Updates button for email, text, Slack, a webhook or a feed.

Here’s what the last month looked like, by the times GitHub gives:

  • Six models at once, for 1 hour 32 minutes. On October 28, “a credential that we use to authenticate with an upstream model provider became invalid,” and about 9.73% of requests to several GPT-family models and MAI Code 1.1 Flash failed, in Copilot Chat in IDEs and on GitHub.com, Copilot CLI, code review and the coding agent.
  • Three GPT models, for 2 hours 22 minutes. On November 13, GPT-6.1 Terra, GPT-5.6 Sol and GPT-6 Luna had “elevated error rates,” from “an issue with an upstream model provider.”
  • A few models, for under an hour. Grok 4.6 and 4.7 on October 23, GPT-5.3 Codex and two GPT-5.6 models on November 9, MAI Code 2 Flash on November 18 and GPT-6 Luna on November 19 each took under an hour. So did a utility model on November 2, which slowed tasks such as commit messages: “No models that users could pick in the model picker were impacted.”
  • Copilot’s cloud agent, for over six hours, twice. On October 29, “Copilot agent sessions on GitHub.com and in VS Code were degraded” for 6 hours 50 minutes, and on November 4, “some Copilot agent users experienced errors listing and creating agent tasks” for about 6 hours 20 minutes.
  • Wider GitHub incidents, twice. On November 3, a fault in GitHub’s servers caused “authentication failures” and errors across its APIs for 1 hour 17 minutes, and “Copilot API errors averaged 2.0% and peaked at 3.1%.” On November 11, an incident across Actions, Pages and Copilot marked Copilot as a major outage for 10 minutes.

Price: free.

Watch for: GitHub’s write-ups give the share of requests that failed, often a few percent, so an incident can be real and still leave most of your requests working. Check which models and features it names before you change anything. And the page can trail the problem: October 28’s first update came 20 minutes after GitHub’s own start time.

Switch models, or choose Copilot Fusion

Best for: an incident on some models, which was seven of the last 11.

This is GitHub’s own advice. On November 19, for GPT-6 Luna, it was “we recommend choosing another model or selecting ‘Fusion’ to continue using Copilot.” Copilot Fusion is GitHub’s model router: “You select Copilot Fusion in the model picker, but you aren’t selecting a specific model,” and it “also considers real-time system health and availability when choosing models.” It works in VS Code, Copilot CLI, the GitHub Copilot app, JetBrains IDEs and Copilot in the cloud on GitHub.com. In Visual Studio, Xcode, Eclipse and GitHub Mobile, Auto does the same job, with “model choice based on real-time system health and availability.”

To switch in VS Code, GitHub’s steps are: “At the bottom of the chat view, select the CURRENT-MODEL dropdown menu, then click the AI model of your choice.” In Copilot CLI, “In an interactive session, enter /model.”

Price: included in your plan. Fusion Standard is included on every plan, including Free; Fusion High and Max come with Pro+, Max, Business and Enterprise. “Through November 30, 2026, users on paid Copilot plans receive a 30% discount on model costs when using Copilot Fusion.”

Watch for: “The models available to you depend on your Copilot plan and where you’re using Copilot,” and on a company seat, “your organization must grant members the ability to switch to a different model.” Fusion picks the model, so the answers may come from one you’d not have chosen. It doesn’t help when GitHub itself is the problem, as on November 3, when sign-in failed.

Run the agent on your computer, not in the cloud

Best for: an incident in Copilot’s cloud agent, like the two longest of the month.

Copilot’s cloud agent runs on GitHub’s servers, and its incidents can leave Copilot on your computer working. On October 29, GitHub’s update said the trouble “can impact cloud based agent sessions on web, mobile, VS Code, CLI and the GitHub Copilot app,” and its write-up blames the agent sessions service, not the models. When that’s the incident, run the task where you are: in VS Code’s agent harnesses, choose the Copilot harness rather than the Cloud execution target, or start Copilot CLI in your project.

Price: included in your plan. GitHub’s plans page says “Free plan supports CLI and agent mode.”

Watch for: a session on your computer uses the same models and the same GitHub service, so it doesn’t help when a model or sign-in is the trouble. It also needs your computer on: a Copilot harness session “can continue in the background after you close it, while the Agent Host stays running.”

Copilot CLI, on your own key or a local model

Best for: staying in Copilot’s agent when GitHub’s service is the part that’s failing.

Copilot CLI doesn’t have to use GitHub’s models. “You can configure Copilot CLI to use your own model provider instead of GitHub-hosted models,” its docs say, “including locally running models such as Ollama.” If Ollama is running with a model downloaded, enter /model, look under “Local and self-hosted” for models marked “Discovered,” and choose Add and use for this session. Other providers, such as OpenAI, Azure OpenAI and Anthropic, go in a providers.json file in ~/.copilot, and you start a session on one with:

copilot --model PROVIDER-NAME/MODEL-NAME

To keep GitHub out of it entirely, set the COPILOT_OFFLINE environment variable to true before you start the CLI. Its sign-in docs say that in offline mode, “No GitHub authentication is attempted,” and “The CLI only makes network requests to your configured BYOK provider.” Copilot CLI runs on Linux, macOS, and Windows from PowerShell or WSL.

Price: your provider’s API rates, or nothing for a model on your own computer. Ollama’s pricing says “Running models on your own hardware is always unlimited.”

Watch for: the docs say to “Choose a model and endpoint that support tool calling (also called function calling) and streaming,” and warn of providers.json that “Credentials entered in this file are stored as plain text.” A local model is only as good as your computer’s memory allows; our local model guide covers which to try. Set this up on a calm day: downloading a model mid-outage takes a while.

VS Code’s chat, with your own key or Ollama

Best for: staying in VS Code’s chat, without GitHub.

VS Code can connect chat to your own provider. Its language models page lists providers that are “already listed (Azure, Anthropic, Gemini, OpenAI, and others),” added with the Chat: Manage Language Models command. For local models, “install the official Ollama extension from the Ollama publisher,” and Ollama’s guide says “Local models do not require sign-in.” In Local harness sessions, “BYOK models work without signing into a GitHub account and without a Copilot plan,” and “BYOK local models work without a GitHub account, a Copilot plan, or an internet connection.”

Price: your provider’s API rates, or free with a model on your own computer.

Watch for: it’s chat only. “Some features require a GitHub account and internet connectivity: semantic search, inline suggestions (code completions), and features that rely on embeddings. These features are not available through BYOK models.” The no-sign-in part is the Local harness, which “is deprecated and will be removed in February 2027,” as our memo explains; in the newer Copilot harness, your own models are an experimental setting. And “If you use BYOK models without signing into a GitHub account, the built-in utility models are not available.”

Claude Code, beside Copilot

Best for: a Claude plan you already pay for, on a service of its own.

Anthropic’s extension for VS Code runs Claude Code next to Copilot, on Anthropic’s service, not GitHub’s: “any paid Claude subscription (Pro, Max, Ultra, Team, or Enterprise) or a Claude Console account works, and no API key is required.” Your project’s instructions come with you. Copilot reads AGENTS.md files “stored anywhere within the repository,” and Anthropic’s docs say “Claude Code can read AGENTS.md as your project instructions, so a repository already set up for other coding agents works without adding a CLAUDE.md, an import, or a setting.” Check status.claude.com before you move.

Price: Claude’s pricing has Pro at $20 a month, or $17 a month billed yearly, Max at $100 and Ultra at $200. The free plan leaves Claude Code out.

Watch for: which Claude you start. VS Code’s built-in Claude session can run on “Copilot-routed models,” which go through the service that’s down; “To use Claude without signing in to GitHub (Experimental), configure a Claude API key,” or use Anthropic’s extension. Claude has limits too: “How fast you reach a limit depends on the model you use, how long your chats are, and how large your files are.” It can’t import your Copilot chats, so start with where things stand.

Codex, beside Copilot

Best for: a ChatGPT plan, or a free ChatGPT account.

OpenAI’s Codex extension runs in VS Code on your ChatGPT plan, and “Codex works with Plus, Pro, Business, Edu, and Enterprise ChatGPT plans.” With no plan at all, VS Code’s experimental Codex integration takes a ChatGPT sign-in: “A free ChatGPT account is sufficient,” and “To use Codex without signing in to GitHub (Experimental), configure the Agent Host integration and sign in to ChatGPT.” OpenAI’s docs say “Codex reads AGENTS.md files before doing any work,” so the same instructions apply.

Price: Codex’s pricing puts it in every ChatGPT plan. Free and Go, at $8 a month, get “GPT-6 Luna at Standard speed in the desktop app, subject to rollout”; Plus, at $20, adds “Codex on the web, in the CLI, in the IDE extension, and on iOS”; and Pro comes in “Plans at $100, $200, or $500 USD per month.”

Watch for: several of Copilot’s incidents were GPT models at “an upstream model provider,” and GitHub doesn’t name the provider, so check status.openai.com before you move. The experimental integration “is disabled when enterprise-managed settings apply, unless your administrator explicitly permits” it. And Codex starts a new conversation.

OpenCode

Best for: coding on with no plan to pay for.

OpenCode is an open-source coding agent, under the MIT license, with a terminal app, and a desktop app in beta for macOS, Windows and Linux. “OpenCode comes with a set of free models. No account needed.” Its provider docs add that with Ollama running, “OpenCode probes http://127.0.0.1:11434, discovers completion models, and adds them to /models without an account connection,” and you can “Use a ChatGPT Plus or Pro plan instead of an OpenAI API key.”

Price: free, with its free models or a model on your own computer.

Watch for: its free models are free “for a limited time,” some with daily limits, and for some, “During its free period, collected data may be used to improve the model.” OpenCode can sign in to GitHub Copilot too, but that’s the service that’s down. It starts a new conversation.

Enjoy

Best for: several labs’ agents in one place, so one failing doesn’t stop the work.

Enjoy runs Claude Code, Codex, Grok Build, Antigravity, Muse Code, and local models through OpenCode, on Mac, Windows and Linux, on the plans you already have. It doesn’t run Copilot, so it’s for moving the work to agents on other services:

  • You see why a request stopped. The conversation’s status changes to Failed, and Enjoy shows “a one-line reason, the agent’s own error under Details, How to fix this, and Retry.”
  • Another agent retries it. Change the agent in the message box, and Retry with runs the request there. The new agent “gets the conversation so far, including replies from other agents.”
  • Free agents are on the list. Antigravity signs in with a Google account, and Google’s pricing has a $0 plan with Gemini 3.8 Flash. OpenCode runs free and local models.

Price: Enjoy is one monthly subscription, and the first seven days are free, with a card to start. It includes your phone and any browser and one computer; each other computer costs extra each month, and each teammate you invite needs a subscription of their own. See pricing. The agents are billed separately: Claude Code needs a paid Claude plan, and Enjoy runs the Codex CLI, which OpenAI includes from Plus.

Watch for: Enjoy isn’t an editor in Copilot’s sense. It doesn’t suggest code as you type or plug into VS Code: the agents write the code. It can’t import your Copilot chats, so the first switch is a new conversation. It doesn’t switch agents for you when a request fails: the request waits until you choose. OpenCode in Enjoy can sign in to Copilot, but that’s GitHub’s service again. Enjoy isn’t open source, it’s a monthly subscription on top of your AI plans once the free week ends, and scheduled workflows run only while your computer is awake with Enjoy open. If you only want Claude or Codex beside Copilot, their extensions don’t need Enjoy.

When to stay with Copilot and wait

Most of the time. Seven of the last 11 incidents hit only some models, and GitHub’s fix for those is a click in the model picker. Six were over within an hour. Setting up a key, a local model or another agent in the middle of an outage can take longer than the outage does.

Set up a second way when an hour matters: before a deadline, for work that runs unattended, or when you already pay for Claude or ChatGPT. The cheapest insurance is free: Copilot Fusion or Auto as your default model, a local model already downloaded, or a second agent signed in before you need it.

A bigger Copilot plan isn’t a documented fix. Max lists “Priority access to new models and features,” but neither the plans page nor the status page says how plans fit an incident. And don’t treat a limit as an outage: our credits guide covers what to do when your allowance runs out.

How we compared

We read GitHub’s status page and every incident on it from October 20 to November 19, 2026; GitHub’s Copilot plans, troubleshooting, Copilot Fusion, auto model selection, model picker, Copilot CLI, sign-in, custom model and custom instructions docs; VS Code’s language models and agent harness docs; Ollama’s VS Code guide and pricing; Anthropic’s pricing and Claude Code docs; OpenAI’s Codex pricing, docs and VS Code extension; OpenCode’s site and docs; Google’s Antigravity pricing; and Enjoy’s own help, all on November 19, 2026, and every one is linked below.

The incident counts are ours. We counted the incidents posted on GitHub’s status page from October 20 to November 19, 2026, by UTC time, that marked Copilot or Copilot AI Model Providers as affected or named Copilot in their title, and took each one’s length from the start and end times GitHub’s write-up gives, or, where it gives none, from its first and last updates. A November 4 incident that delayed the Copilot coding agent, among other things, marked only Actions, Pages and Webhooks, and a November 9 one that held up some Copilot code review jobs marked no part at all, so neither is in the count.

We quote each maker’s own words for its limits and leave out what it hasn’t documented, and we didn’t run these setups ourselves. Status pages and plans change often, so we recheck this guide against its sources at least once a month, and the date at the foot of the page says when we last did.

14.Questions.

Is GitHub Copilot down right now?

Check githubstatus.com, GitHub’s status page, which lists Copilot and Copilot AI Model Providers separately. An incident on one model shows under the second, and names the models it affects, so you can tell whether switching models will help. GitHub’s own troubleshooting starts there too, for “any active incidents affecting GitHub Copilot or model availability.”

Why does only one model in Copilot stop working?

Because most Copilot incidents start with one model provider. Seven of the 11 Copilot incidents from October 20 to November 19, 2026 named particular models, and GitHub blamed “an issue with an upstream model provider.” In six of them, its advice was to choose another model, often adding Auto or Copilot Fusion, which pick a model for you.

How long do Copilot outages usually last?

Usually under an hour. Six of the 11 Copilot incidents on GitHub’s status page from October 20 to November 19, 2026 were over within an hour, and three more within two and a half. The two longest, over six hours each, were in Copilot’s cloud agent, where about 1 to 2% of requests failed on average.

Can I use Copilot without GitHub when it’s down?

Partly. Copilot CLI can run on your own model provider or a local model, and with `COPILOT_OFFLINE` set to `true`, “No GitHub authentication is attempted.” VS Code’s chat can use your own key or Ollama without signing in, in its Local harness, which goes away in February 2027. Inline suggestions still need GitHub.

What can I use instead of Copilot during an outage?

Another lab’s agent, on its own service. Claude Code needs a paid Claude plan; Codex works in VS Code with a ChatGPT plan, or a free ChatGPT account in its experimental integration; OpenCode has free models. Enjoy runs Claude Code, Codex, Antigravity and OpenCode in one conversation, and hands it from one to another.

Is “You’ve hit a rate limit” an outage?

No, though it can come from the same busy models. GitHub says the error “suggests that you have exceeded the rate limit for Copilot requests,” and that “Most people see rate limiting for select models, due to limited capacity.” The message says whether to wait, retry later or upgrade. Running out of AI credits is different again; see our credits guide.

Enjoy Agentic Industries · Checked against its sources on