Improved Agent tools, steerability, and usage visibility

Source: Cursor Date: 2025-08-06 URL: https://cursor.com/changelog/1-4

Summary

Cursor 1.4 ships interrupt-aware agent steerability (Alt+Enter queues messages, Ctrl+Enter interrupts immediately at optimal points), removes the 2MB file read cap for large codebases, adds per-tab model assignment for individual agents, GitHub PR integration via @Cursor mentions (agent reads prompt, applies fixes, commits), and usage statistics in chat (all-time or >50% quota warning). Enterprise gets AI code tracking API and repository blocklist admin API.

Implications

Agent-IDE feature race. Interrupt-at-optimal-times steerability is a UX innovation over simple “stop” buttons — Cursor is building agent-aware interrupt semantics rather than hard kills. Per-tab model assignment is a power user feature with real workflow implications: use Claude for planning, GPT-4o for implementation, Gemini for context-heavy lookups, all within one session. No competitor surfaces this level of per-agent model granularity.

Pricing model evolution. Usage statistics at >50% quota warning is the first real-time cost signal in the Cursor UI. Combined with the AI code tracking API (commit-level AI attribution) and admin repository blocklists, Cursor is building the enterprise cost governance layer that procurement teams require. Watch for these features to drive enterprise tier differentiation.

Watch: Whether @Cursor GitHub integration expands to Issues (not just PRs) and whether per-tab model assignment generates pricing complexity (different models have different fast-request costs).