dsh-model-sync

中文 | English

One-click / auto model sync for DeepSeek Harness providers. The plugin queries each provider's /v1/models endpoint and replaces the stored model list with what the endpoint advertises (models the endpoint no longer lists are removed; other provider fields are kept).

Features

• Manual sync — one-click sync of every configured provider from the Model Sync panel in Settings. Sync runs in the browser through the official settings / llm APIs — no custom Host RPC.
• Auto-sync — re-syncs whenever a conversation page opens (Host listens on agent/session-start, 10s debounce). The toggle persists in the plugin's model-sync settings namespace + localStorage.
• Replace semantics — the discovered list IS the final list; provider models that were removed or added are reflected correctly.
• Failure isolation — a failing provider does not affect the others; failed providers keep their previous config.

Installation

npx (recommended — nothing to install first)

# One command installs the plugin into the current DSH web profile
npx github:zhangyqjiaoshou-oss/dsh-model-sync install

After the package is published to npm, this becomes npx @dsh-external/dsh-model-sync install.

dsh plugin add (bundle-capable environments)

dsh plugin --profile web add @dsh-external/dsh-model-sync

Manual injection (via dsh-super-injector)

# Build from source
bash scripts/build.sh
# Inject into the running DSH
dev_inject_plugin C:/Users/niclas/Desktop/dsh/model-sync

Usage

1. Open Settings → Model Sync.
2. Click Sync All Models to sync immediately, or turn on Auto-sync on conversation open to refresh the model list every time you open a conversation page.
3. Each provider reports ✅ N total (+added -removed); failures show ❌ with the reason.

Configuration

Field	Type	Default	Description
autoSync	boolean	false	Auto-sync model lists whenever a conversation page opens
debounceMs	number	10000	Minimum milliseconds between two automatic syncs

The auto-sync switch can also be toggled in the Settings UI; changes are written to the model-sync namespace.

How it works

Conversation page opens → agent/session-start event (Host)
  → re-query each provider's /v1/models
  → settings.update writes the new model list
  → settings/document-updated fires
  → browser model selector refreshes automatically

Manual sync uses only official wire APIs (llm.discoverModels + settings.mutate) — no custom Host RPC.

Development

bash scripts/build.sh          # auto-detects a DSH checkout or npm global install
npx tsc -p tsconfig.json       # typecheck
npx tsdown                     # client bundle (lib/client.js)

License

MIT