dsh-provider-rate-limit

English | 简体中文

Per-provider & per-model rate limiting for DeepSeek Harness LLM traffic, plus gateway identity rules (client spoofing) for restricted free-tier gateways.

适用于 DeepSeek Harness 的按供应商/模型粒度 LLM 限速插件，附带网关身份规则（客户端伪装）能力。

Features

• Token-bucket rate limiting per (provider, model) route — smooth refill with burst support, idle-time recovery
• Two modes when the bucket is empty:
  • wait — hold the request up to maxWaitMs, then let it through (transparent queueing)
  • reject — short-circuit immediately with a synthetic RATE_LIMIT response carrying providerRetryAfterMs
• Strict FIFO — reservation-based design guarantees same-order admission without polling
• Gateway identity rules — rewrite User-Agent / inject static headers for URLs matching a pattern (e.g. gateways that validate client identity), with a one-click OpenCode Zen preset
• Master switch — flip enabled off to pass all traffic instantly, no listener re-registration
• Settings UI card — full configuration from the Harness settings page, zh/en localized
• Live stats line — compact readout in the composer dock (under the chat input), auto-refreshes every 5s; hover to see per-route provider·model breakdown
• Stats HTTP API — GET /api/provider-rate-limit.stats returns aggregate and per-route counters as JSON (used by the dock line; also available for external tooling)
• Cross-plugin stats service — provider-rate-limit/stats service for in-process consumers (getStats, getAllStats, getAggregateStats, resetStats)
• O(1) route lookup — pre-built Map for rule matching instead of linear scan
• Standard ULID — 26-char Crockford base-32 IDs (48-bit big-endian time + 80-bit random)

Install

DSH plugin manager (recommended)

dsh plugin --profile web add github:jyao-SUSE-power-group/dsh-provider-rate-limit

Then restart DeepSeek Harness. The plugin registers itself into the llm service via its cordis patch.

Manual

git clone https://github.com/jyao-SUSE-power-group/dsh-provider-rate-limit.git ~/.dsh/plugins/dsh-provider-rate-limit
cd ~/.dsh/plugins/dsh-provider-rate-limit && pnpm install --prod

Configuration

Open Settings → 插件 → Provider Rate Limit. All options hot-reload — no restart needed.

…