dsh-plugin-ai-gateway

If you hold subscriptions to both Google Antigravity and OpenAI Codex only to wrestle with incompatible APIs, opaque rate limits, and scattered quota dashboards, dsh-plugin-ai-gateway pulls those fragmented AI channels into a single standard OpenAI-compatible endpoint that plugs directly into your DeepSeek Harness workspace. Gemini, Claude, and the GPT family stop being three separate logins and become peers you switch between in one panel.

The plugin goes well beyond a simple reverse proxy. A built-in multi-account rotation pool handles five-hour rate limits and transient 429 errors with millisecond-level failover, so inference tasks never stall. Real-time quota dashboards mirror the official rolling windows and weekly budgets with countdown-to-reset timers and color-coded saturation capsules, all rendered in zero milliseconds thanks to a Stale-While-Revalidate cache. An adaptive egress proxy sniffs local ports automatically, eliminating TCP handshake timeouts that plague TUN-mode virtual NICs on large request bodies. On the context side, a 262k agile-switch track and a 1M long-context track run in parallel, so daily coding and heavy legacy-code scanning each get the throughput they need.

Built for developers who maintain both Google and OpenAI subscriptions and run multi-model workflows in DSH day to day, and for users who want zero-config access to Zhipu ZCode Coding Plan credits to invoke GLM-series coding models locally. All traffic is strictly bound to 127.0.0.1, data-plane and management-plane keys are generated independently, and credentials are stored in an isolated vault, keeping the security perimeter clean and predictable.