opencode2dsh

DSH users who want free OpenCode Zen models have always had to take detours: register an account, grab an API key, spin up a sidecar or a local proxy. Every one of those steps is ongoing maintenance. opencode2dsh removes all of that. It registers a native LlmAdapter inside DSH that streams straight from OpenCode Zen's anonymous free lane. Install, restart, and a new provider group called opencode2dsh appears in your model picker, ready to use like any local model.

The engineering details make it genuinely low-maintenance. A single npm package, no child process, no binary, no local port. Every request carries the same user agent and session-correlation headers as the OpenCode CLI itself. The model catalog follows a three-tier fallback chain: live upstream list, pricing-metadata filter, and a verified static list. Startup fetch races the network? It retries on a short cadence before settling into periodic refreshes, and a health snapshot is written after every round. Upstream failures, rate limits, timeouts, transport errors, surface in DSH as classified finish reasons and retry ownership stays with DSH.

It is built for people already running DSH who refuse to add another credential or another process to their stack: students, solo developers, anyone for whom every extra moving part feels like one more thing to break. Zero secrets, zero storage, zero telemetry; requests go directly from your machine to opencode.ai and models.dev with nothing in between.