dsh-llm-api-pool

Juggling several opencode go subscriptions by hand means constantly checking balances, swapping keys one at a time, and eating 429 errors at the worst possible moment. dsh-llm-api-pool collects multiple {baseUrl, apiKey} entries into a single pool, calls the official /usage endpoint to pull exact 5-hour, weekly, and monthly quota percentages with precise reset timestamps, and renders each key as a progress bar with a dollar conversion so the whole pool is visible at a glance.

Routing is driven entirely by official usage pressure: before every model request the plugin automatically selects the subscription with the largest remaining quota. When any key returns a 429, a 5xx error, or times out, it is cooled down for 60 seconds and the next key takes over with zero human intervention. On load the plugin also self-registers as a native DSH provider, making every probed model in the pool immediately selectable with a reasoning-effort picker, and it exposes an OpenAI-compatible endpoint so any SDK can simply point its baseURL at the pool and inherit load-balanced, failover-ready inference.

Built for developers holding multiple opencode go subscriptions who want to eliminate manual key rotation and balance anxiety, and for teams that need to unify several LLM endpoints under one load-balancing and failover layer inside the DSH ecosystem.