Luck9Star/dsh-gateway-provider
Lets you use all the models behind your own LLM gateway — NewAPI, LiteLLM, Higress or any OpenAI-compatible endpoint — directly inside DeepSeek Harness. The README documents a Settings → Gateway Models surface where you declare gateways from templates (or a fully custom one with per-protocol URLs), a configuration reference with fields such as flavor (template label only) and sortModelsByRelease (newest first), and how model metadata — context window, output cap, reasoning levels, release date — is used with config defaults only filling gaps. It also documents key storage, troubleshooting, and a security section; the plugin adds a single provider route, so installation is the standard bundle flow.
Install
dsh plugin --profile web add dsh-gateway-providerREADME install, quoted verbatim: dsh plugin --profile web add dsh-gateway-provider (the README notes dsh plugin add runs pnpm add under the hood). Then store your key by one of the documented routes — writing it to the profile config or exporting it in the shell that launches dsh (e.g. export NEWAPI_API_KEY=sk-REPLACE_WITH_YOUR_KEY) — restart, and open Settings → Gateway Models. The plugin pins its own pi-ai version so the gateway model list does not depend on the harness-bundled pi-ai. npm dsh-gateway-provider 1.0.12 (registry-verified 2026-09-12; repository field → github.com/Luck9Star/dsh-gateway-provider).
Compatibility
DeepSeek Harness web profile. The README lists gateway templates for NewAPI, LiteLLM, Higress, generic OpenAI-compatible and fully custom per-protocol URLs; it pins its own pi-ai so gateway models are decoupled from the harness-bundled version.
Details
- Repo: Luck9Star/dsh-gateway-provider
- Category: Coding & Development
- Stars: 0
- Version: npm dsh-gateway-provider 1.0.12 (registry-verified 2026-09-12)
- Last push: 2026-08-15
- First seen: 2026-08-13
Recent updates
The README documents configuration behaviour rather than a release timeline: the flavor field is a template label, sortModelsByRelease defaults to true, and per-model metadata (context window, output cap, reasoning levels, release date) drives the picker with config defaults only filling gaps. It includes a development section explaining why the plugin pins its own pi-ai (decoupling the gateway model list from the harness version) and a references/credits section.
FAQ
- How do I install dsh-gateway-provider?
- dsh plugin --profile web add dsh-gateway-provider, then store your gateway key (e.g. export NEWAPI_API_KEY=...), restart and open Settings → Gateway Models.
- Which gateways are supported?
- NewAPI, LiteLLM, Higress, any OpenAI-compatible endpoint, or a fully custom gateway with per-protocol URLs — the README lists these as templates.
- Does it change which pi-ai version the harness uses?
- No — the README states the plugin pins its own pi-ai so the gateway model list is decoupled from the pi-ai bundled with the harness.
Alternatives
nexu-io/open-design · dsh-market/dsh-market · xiufengsun/TokenTracker