Answers · agent payments
Since GitHub Copilot moved to AI credits, how do I stop the cloud agent, agent mode or Copilot CLI from running up charges beyond what I've budgeted, for me or for my whole team?
Last updated
TL;DR: Set a budget that actually blocks usage, because by default most GitHub spending limits only send email. Since June 1, 2026, Copilot Chat, agent mode, Copilot CLI, the cloud agent, Spaces, Spark and third-party coding agents are billed in GitHub AI Credits (1 credit = $0.01) based on the tokens each model uses, so a long agent session on an expensive model costs far more than a quick question. Code completions and next edit suggestions don't use credits on paid plans. On an individual plan (Pro, Pro+ or Max), GitHub's advice for avoiding charges beyond the plan is to create a Bundled AI credits budget of $0 USD and turn on "Stop usage when budget limit is reached"; set it higher if you want to allow some extra spend. For an organization or enterprise, set a universal user-level budget (it always hard-stops each person, including use of the included pool), then turn on "Stop usage when budget limit is reached" on every organization, cost center and enterprise budget. GitHub says that setting is off by default, and without it charges keep accruing past the limit. Remember that AI credits aren't the only meter: cloud agent sessions and Copilot code review also use GitHub Actions or cloud sandbox compute, which needs its own budget.
What should I check, step by step?
1. Individual plan: go to github.com/settings/billing, open Budgets and alerts, and create (or edit) a Bundled AI credits budget. Set $0 USD to block anything beyond your plan's included credits, or a small amount if you're happy to pay a little extra, and tick "Stop usage when budget limit is reached".
- Don't delete a budget to "turn it off". GitHub's legacy billing docs warn that deleting a budget is not a substitute for a $0 budget with stop usage enabled and can remove spending limits.
- Organization or enterprise: create a universal user-level budget before rolling Copilot out. It caps each user's AI credit consumption per billing cycle, including credits drawn from the included pool, and always blocks at the limit. A $0 user-level budget blocks credit-consuming features immediately.
- Give heavy agent users an individual user-level budget instead of raising everyone's. Individual budgets can expire at the end of the billing cycle or on a date, so a one-sprint increase cleans itself up.
- On every organization, cost center and enterprise budget, turn on "Stop usage when budget limit is reached". Without it, GitHub says, those budgets are alerts only and metered charges continue without a cap.
- Check where Copilot cloud agent and automations run. Sessions on GitHub Actions or in cloud sandboxes draw on your account's Actions and compute allowance, separate from AI credits, so put a stop-usage budget on Actions too. Self-hosted runners avoid Actions minutes but not AI credits.
- Decide whether Copilot code review runs automatically on every pull request. Each review uses AI credits (GitHub estimates about $0.25 at Lite effort, $1.50 at Balanced and $10 at Max per review, with wide ranges) plus Actions minutes on private repositories.
- Turn on budget threshold alerts (75%, 90% and 100%) and review the AI usage dashboard at least monthly, looking for users blocked early or metered charges above plan.
- If an agent or script can call GitHub's billing or budgets API, don't let it raise a budget on its own. Require a named person to approve the exact change.
How does GitHub Copilot bill agent usage in AI credits since June 2026?
GitHub replaced premium requests with GitHub AI Credits for most Copilot plans. Each interaction is priced by the model's per-token rates for input, output and cached tokens, and the total is converted at 1 AI credit = $0.01 USD. Only Pro and Pro+ subscribers who stayed on an existing annual plan remain on the older premium-request billing.
- Individual plans include base credits plus a flex allotment: Copilot Pro ($10 a month) includes 1,500 credits in total, Pro+ ($39) 7,000 and Max ($100) 20,000. Copilot Business ($19 per user per month) includes 1,900 credits per user and Copilot Enterprise ($39) 3,900, pooled across the billing entity. Included credits reset at 00:00 UTC on the first of each month and don't carry over.
- GitHub's own docs say agentic features can involve many model calls in one task, and that a complex agentic session across a large codebase uses significantly more than a quick chat question. Model choice matters too: more capable models cost more per token.
Which GitHub Copilot budgets block usage and which only send alerts?
User-level budgets (universal, per cost center, or individual) always enforce a hard stop and have no setting to turn that off. When more than one applies, the most specific wins: individual over cost center over universal.
- Organization, cost center and enterprise budgets only limit metered charges (on Business and Enterprise, the usage after the shared pool runs out), and they only block if "Stop usage when budget limit is reached" is on. GitHub calls that setting off by default.
- GitHub checks a request in order: the user's budget first, then whether pooled credits remain and paid usage is allowed, then the cost center, organization or enterprise budget. Whichever budget has the least headroom blocks first. A blocked user stays blocked until the next billing cycle, an admin raises the budget, or an admin approves the user's request for more.
- Two gaps to plan for. An enterprise budget doesn't cap license fees, so a $5,000 enterprise budget on 400 Copilot Business seats still allows $7,600 of seat charges on top. And usage billed to an organization from a GitHub App installation counts against the organization budget, which only stops that spending if stop usage is enabled.
Does a Copilot AI credits budget also cap GitHub Actions minutes used by the cloud agent and code review?
Copilot cloud agent sessions run either on GitHub Actions or in a cloud sandbox, and GitHub says both draw on the same monthly account compute allowance, separate from AI credits. A session on GitHub Actions has a maximum run time of 59 minutes.
- Copilot code review is billed twice: tokens in AI credits, and the agentic infrastructure in GitHub Actions minutes. GitHub doesn't disclose which model a review uses, so per-review cost varies; its estimates grow with pull request size and custom instructions.
- A budget on AI credits won't stop Actions minutes, and an Actions budget won't stop AI credits. Set both if you want a real ceiling on what an agent can cost you in a month.
How do I ask Rightshold about this (free, no key)?
Add the MCP server https://www.rightshold.com/mcp to Claude, Cursor, ChatGPT, Grok or VS Code, then ask: “My ops agent wants to raise one developer's Copilot user-level budget from $20 to $100 until the end of this billing cycle. Ask me to approve that exact change before it calls GitHub's budgets API.”
{
"tool": "request_human_approval",
"arguments": {
"action": {
"provider": "github",
"api": "budgets",
"budget_type": "individual_user_level",
"user": "<github login>",
"from_usd": 20,
"to_usd": 100,
"expires": "end_of_billing_cycle"
},
"summary": "Raise one developer's Copilot user-level budget from $20 to $100 until the end of the cycle",
"credential_id": "crd_YOUR_AGENT_CREDENTIAL",
"expires_in_minutes": 60
}
}- The action object is hashed, so the approval covers exactly that user, amount and expiry. Before the API call, check the request against the approved hash with verify_human_approval (free) and refuse if anything changed.
- Only the owner's account key can decide (decide_human_approval). Delegated agent keys are refused, so the agent can't approve its own budget increase.
- Rightshold doesn't connect to GitHub, read your Copilot usage or enforce GitHub budgets. Those limits live in GitHub's billing settings.
Rightshold records approval requests and signed decisions for actions an agent proposes. It doesn't read or change your GitHub billing, see Copilot usage, or stop GitHub charges; GitHub's budgets do that. Plan names, prices and credit amounts are as GitHub documented them in November 2026 and may change. Not financial advice.
What are the sources?
- GitHub Docs: Usage-based billing for individuals (AI credits, base credits and flex allotment)
- GitHub Docs: Models and pricing for GitHub Copilot (1 AI credit = $0.01, $0 Bundled AI credits budget, code review costs)
- GitHub Docs: Budgets for usage-based billing (user-level, cost center, organization and enterprise budgets)
- GitHub Docs: Getting started with budget controls
- GitHub Docs: Setting up budgets to control spending on metered products
- GitHub Docs: GitHub Copilot on GitHub.com (cloud agent environments, compute allowance, 59-minute limit)
- GitHub Docs: Overview of request-based billing (legacy)
Related
- Stopping an AI agent from looping forever and burning tokens
- Giving each AI agent its own budget for paid APIs
- Alerting when an AI agent approaches its spend cap before the hard deny
- Stopping an AI agent from running up a cloud bill by launching GPU instances
Not legal advice. JSON · Markdown · All answers
Get a free API key in one call
No card and no email needed. 100 free credits; the license and policy checks stay free without a key.
curl -s -X POST https://www.rightshold.com/v1/accounts/signup -H "Content-Type: application/json" -d '{"accept_terms":true}'
export RIGHTSHOLD_API_KEY=rs_... # the api_key from the reply (shown once)
curl -s https://www.rightshold.com/v1/licenses/check -H "Authorization: Bearer $RIGHTSHOLD_API_KEY" \
-H "Content-Type: application/json" -d '{"license":"MIT","intended_use":{"commercial_use":true}}'