dhs-connect

DeepSeek Harness ships no native path to ChatGPT models, and routing through the OpenAI API means managing an extra key and a separate billing line. DSH Connect bridges Harness to the official Codex app-server via OAuth, so a single ChatGPT sign-in puts those models right in the usual picker—no API key required.

The connector ships one shared model catalog (account-discovered entries plus a curated fallback) that both Settings and the model picker read from, with built-in connection diagnostics. On the inference side it handles text streaming, reasoning summaries, Harness dynamic tool calls, and token accounting; on the reliability side it covers request cancellation, RPC timeouts, process crash recovery, and conversation reconstruction after restart, compaction, or changed tool definitions. The package carries zero runtime npm dependencies and performs no automatic binary downloads.

It is a natural fit for developers who already run DeepSeek Harness and want ChatGPT models in the daily loop without an extra API key or a proxy, as well as teams that prefer locally controlled, auditable inference pipelines. The plugin never opens a public endpoint and leaves all authentication to the official Codex executable.