dsh-plugin-tokenmarket

Wiring a new model provider into DeepSeek Harness usually means configuring a base URL, selecting a protocol, injecting credentials, and writing adapter glue code for every family you want to serve. dsh-plugin-tokenmarket collapses all of that into a single bundle: it registers Token Market as an OpenAI-compatible tokenmarket provider, mounts the official pi-ai adapter, and lets DeepSeek, Kimi, and GLM model families share one API key. The starter catalog is intentionally small, so each bundled model can be verified and kept current.

In practice the bundle targets everyday inference workflows. Streaming responses keep long-form generation responsive; tool calling works out of the box through the pi-ai adapter with no custom glue; and the DeepSeek-compatible thinking format surfaces reasoning steps in a structured way that is straightforward to debug and render. Live model discovery via GET /v1/models means catalog changes never require a config edit, and credentials stay outside the repository and the configuration file through the TOKENMARKET_API_KEY environment variable.

If you are running the DeepSeek Harness developer preview and want to compare, fall back, or A/B test across multiple model families inside a single profile, this bundle lets you do it with one console and one key, so your attention stays on prompt engineering and application logic rather than provider plumbing.