GitHub Copilot token billing June 2026 causes 10x to 50x cost shock for agentic users with no real-time spend kill-switch at the IDE network layer
GitHub Copilot switched to token-based billing on June 1 2026, replacing flat-rate PRUs with GitHub AI Credits priced at API token rates. The official GitHub discussion thread accumulated 400 comments and 900 downvotes. Reports of bills jumping from $39 to $750 per month and from $50 to $3000 per month spread across Reddit and X. The root cause: AI coding agents running in agentic mode consume tokens silently in the background, with no per-session visibility or kill-switch at the developer's machine. Existing solutions (MCP budget proxies, backend spend dashboards) operate at the agent orchestration layer -- they do not intercept the IDE extension's own network calls to GitHub/Anthropic/OpenAI APIs. The gap: a CLI-resident local proxy that sits at the OS network layer, intercepts AI IDE extension traffic (Copilot, Cursor, Claude Code) matching known API endpoints, tracks live token spend with a terminal dashboard, and enforces a hard per-session or per-day budget that terminates the connection when the cap is hit.
A web app that attributes and hard-caps AI coding assistant spend across seats, credit pools, and agent runs for engineering orgs
85.3k ▲Score Breakdown
Social Proof 2 sources
Gap Assessment
MCP server budget proxies in the pipeline target agent orchestration calls. This targets the IDE extension's direct HTTP traffic -- a different interception point that works even when the IDE is not using MCP. The GitHub Copilot billing change is fresh (June 1 2026) and the 900-downvote community reaction shows immediate pain. No existing tool intercepts AI IDE network calls for real-time cost enforcement. The closest thing is mitmproxy used manually -- a productized version with pre-built rules for Copilot/Cursor/Claude Code endpoints is the gap.