TonyWu2333/dsh-mercury-diffusion

Watch a diffusion LLM think: real Inception Mercury denoising frames stream into the answer bubble, from noise to finished text.

Every frame it plays is a real denoising state returned by the Mercury API, replayed at a readable pace instead of being faked as an animation. Intermediate frames travel on a side channel, so only the converged text is written to the session log and shown to the model. Attachments, auxiliary calls, unwatched sessions, routes without a credential and gateways that reject the field all fall back to the original adapter untouched. A small floating button toggles diffusion per model, and the choice is stored in the plugin config key for that provider and model, so it survives restarts.

Models & Providers ★ 2
View on GitHub ↗

Install

dsh plugin --profile web add dsh-mercury-diffusion

The npm line from the repo's English README (dsh-mercury-diffusion 0.2.1 on the registry). The README states three prerequisites before it: dsh between 0.2.1-rc.2 and 0.3.0, Node 22.19 or newer (or 24+), and pnpm on your PATH, because dsh plugin uses pnpm to manage profile plugins.

Compatibility

Requires dsh >= 0.2.1-rc.2 < 0.3.0 and Node ^22.19 || >=24. The plugin registers no provider and no settings page of its own; it adds a diffusion layer underneath a route you configure on the Models page, and whether a route honours the diffusion field is probed with a request that is rejected before any token is generated, so the capability check costs nothing.

Details

Alternatives

GooDAnDReaDY/dsh-key-rotation · btspoony/dsh-llm-fallbacks · zeng6125-rgb/dsh-llm-retry-settings

More plugins in Models & Providers

Browse more in Models & Providers

Guides for Models & Providers plugins