dsh-quota-limits: simulate vendor subscription quotas for DeepSeek model calls

[ttzevol]

A third-party dynamic Cordis plugin that applies other vendors' subscription-plan quotas (OpenAI / Anthropic / Google) to DeepSeek model calls — since DeepSeek has no consumer subscription tier.

Repo: https://github.com/ttzevol/dsh-quota-limits (MIT)

What it does

• Vendor plan presets (calibrated to a 2026-07 snapshot, each with source/approximation notes): OpenAI Plus/Pro, Claude Pro/Max 5x/Max 20x, Google AI Pro/AI Ultra
• Dual limits: request rate (time window) + daily input/output token budgets
• Hard blocking: over quota, no real request is sent — returns a synthetic 429, same as real vendor rate limiting
• Persistent usage in <DSH home>/quota-limits.json (survives plugin stop / process restart)
• Settings page via the settings.section slot; read-only quota_status tool for the model

Install

Paste src/host.js / src/client.js into cordis_define and cordis_run in a Web GUI session (see the repo README, bilingual zh/en).

Notes

• Unofficial, AI-generated (reviewed by its author); values are approximations, not official vendor data
• Built against pre-release harness APIs: llm/stream waterfall, dshHomePath, settings.section, harness tool registration

Feedback welcome — especially on the quota-interception approach and preset calibration.