# CatchReset (catchreset.com)
> The authoritative real-time radar and telemetry intelligence for AI coding quota resets, burst signals, and model degradation tracking.

CatchReset tracks sliding rate-limit windows, weekly quota anchor refreshes, burst reset drops, and model service degradation across major AI programming tools including OpenAI Codex / Astra, Anthropic Claude Code, xAI Grok Code, and Cursor Composer.

## Core Telemetry Data & Specifications

- **OpenAI Codex / Astra**: 5-hour rolling sliding window plus weekly scheduled refill. Sudden burst reset drops and compensatory reset cards are officially announced by Tibo (@thsottiaux). Thinking tokens generated by reasoning models count 1:1 against output quota.
- **Anthropic Claude Code**: 5-hour dynamic adaptive rate limit window. Hybrid thinking budget consumes sliding window allocation heavily; prompt caching saves up to 90% cost and 70% window pressure. High concurrent loads lead to internal thinking step throttling (degradation alert).
- **xAI Grok Code / SuperGrok**: 2-hour rapid promotional reset cycle announced by Elon Musk. High throughput code-generation queue.
- **Cursor Composer / Agent**: Monthly billing cycle reset for 500 Fast Requests (Claude 3.7 / GPT-4o). Reverts to slow queue or usage-based billing (~$0.04 to $0.10/request) upon exhaustion.

## High-Frequency Error Solutions

- **"You've reached your usage limit. You can continue after 5 hours or upgrade"** (OpenAI Codex): Token or turn budget exceeded in current 5h sliding window. Do not retry aggressively; switch to secondary account or Claude Code; monitor window rollback on CatchReset.
- **"You are out of extra usage until 3:00 PM. Please wait for quota refresh"** (Anthropic Claude Code): 5h rate limit pool exhausted. Run `claude config set thinkingBudget 0` to disable reasoning overhead, or compact CLAUDE.md context.
- **"Rate limit reached for requests per minute (RPM) / tokens per minute (TPM) 429"** (Universal API Gateway): Instantaneous concurrency spike. Implement exponential backoff (2s, 4s, 8s); self-heals within 60s.
- **"Thinking budget exceeded for current sliding window"** (Reasoning Models): Deep chain-of-thought code reasoning exhausted allocated compute. Constrain prompt with `Keep thinking under 4000 tokens`.

## Direct Endpoints for Agents & Developers

- **Interactive Web Radar**: https://catchreset.com/
- **Codex Dedicated Radar**: https://catchreset.com/codex
- **Claude Code Dedicated Radar**: https://catchreset.com/claude
- **Grok Code Dedicated Radar**: https://catchreset.com/grok
- **Cursor Dedicated Radar**: https://catchreset.com/cursor
- **JSON Telemetry API**: https://catchreset.com/api/status
- **Official Drops API**: https://catchreset.com/api/drops
- **Remote MCP Server (JSON-RPC 2.0)**: https://catchreset.com/api/mcp
- **Dynamic Shields.io Badge**: https://catchreset.com/api/badge/{platform}.svg
- **Full Knowledge Specification**: https://catchreset.com/llms-full.txt