dsh-model-sync

English · 中文

Writes each provider's live model list into DeepSeek Harness settings. The composer ring follows the model this session is using. Plan routes read 5h and 7d windows, metered routes read the balance left since the last top-up. With windows and a top-up pack together, the icon draws the windows and keeps the pack in the hover card.

It takes two settings pages, one for fetching model ids per provider and one for per-model usage. Model Sync (sync-arrow icon) writes new model ids from each provider into settings. Usage lists plan windows, cumulative counts, and API balance. Resets show as countdowns such as 2d 3h left. The 5h and 7d windows each get their own row. UI text follows the DSH language setting, English or Chinese.

Screenshots

	中文	English
Model Sync
Usage
Composer

A model with both limits draws two rings, inner 5h, outer 7d. Progress bar colors follow the remaining quota.

Install

On npm.

dsh plugin --profile web add dsh-model-sync

The GitHub source also ships prebuilt lib/, no allowBuilds.

dsh plugin --profile web add github:jiay98528-dev/dsh-model-sync

Restart dsh web, refresh the page, open Settings → Model Sync. The plugin is listed in the awesome-dsh-plugin market list.

The package ships a cordis.patch.yml bundle patch (dsh.bundle) and declares dsh.client.immediately: true, so it takes effect on install with no manual config.

Usage

Model Sync

1. Settings → Model Sync.
2. Refresh if the list is empty.
3. Each card lists the provider id, baseURL, and model chips. A blue outline marks ids not in settings yet. Apply N new models writes them to llm-pi-ai.providers.<id>.models.
4. The collapsed Enable plugins block toggles this plugin and the optional sub-model-access, writing the disabled flag in cordis.patch.yml. Takes effect live.

MiniMax, Kimi Coding, and OpenAI Codex have no OpenAI /models endpoint. The plugin uses the known catalog ids.

Usage

1. Settings → Usage.
2. Each provider lists its 5-hour and 7-day windows, one row each, with remaining percent, vendor cumulative counts, a reset countdown, and a colored bar.
3. Metered rows show current / baseline. The first sample becomes the baseline. To count spend from before the plugin was enabled, click Use current balance as baseline. The balance row also records when the baseline was set.
4. The page auto-refreshes every 60 seconds. Refresh pulls immediately.

Composer rings

1. Pick a model as usual.
2. The ring is right of the model name and left of the context meter. It follows the session model and switches with it.
3. Hover the ring for a detail card titled provider / model for this session. Each window gets a row with remaining percent and reset countdown, for example 5 hours left 100% · 2h 15m left.
4. Other providers stay on the Usage page.

Quota sources

The DeepSeek metered baseline lives in $DSH_HOME/profiles/<profile>/model-sync.baseline.json. A reading above the stored baseline counts as a top-up and resets the baseline.

Config

- id: model-sync
  name: dsh-model-sync
  config:
    profile: web
    pollMs: 60000

Notes

• Host code lives under node_modules. After changing and rebuilding this package, restart dsh web. Client-only edits need a page refresh.
• Quota calls use credentials already stored in DSH. The plugin keeps no keys of its own.
• Ring placement is bound to DSH 0.1.0-rc.6 composer class names.
• Changelog is CHANGELOG.md.

License

MIT