Fynth/dsh-nanogpt

Registers NanoGPT as a first-class DSH provider with live model discovery, per-model reasoning levels, three wire protocols and a Settings page.

The plugin discovers the live model catalogue from the NanoGPT models endpoint with a last-known-good disk cache, a background refresh and a manual refresh, keeping exact upstream model IDs instead of normalising them. Each exact model exposes its own reasoning efforts, from none and minimal through xhigh and max to Auto and provider default, with unsupported levels rejected before any HTTP request and none kept distinct from Auto. Three protocols are supported, OpenAI Chat Completions by default, OpenAI Responses and Anthropic Messages, with reasoning, text and tool calls streamed back through the usual harness surfaces.

Models & Providers ★ 0 updated 2026-10-18
View on GitHub ↗

Install

dsh plugin --profile web add github:Fynth/dsh-nanogpt#v0.1.1

Copied verbatim from the README's install line, pinned to the released tag. There is no registry package for this name, so the documented GitHub spec is the install path; tag v0.1.1 exists on the repository (verified 2026-10-18). The plugin installs one Loader row named llm-nanogpt and does not modify DSH core.

Compatibility

Requires DSH 0.2.0-rc.1 or newer, with 0.2.0-rc.2 the verified host, and Node.js 22.19 or newer. It registers the independent nanogpt route rather than reusing a built-in one, so it appears in the native model picker and the native reasoning-level selector like any shipped provider.

Details

Alternatives

ffyuuu/dsh-llm-longcat · wenzetan/dsh-llm-newapi · lament-z/dsh-chanhub

More plugins in Models & Providers

Browse more in Models & Providers

Guides for Models & Providers plugins