dsh-localhost-cost

English | 中文

dsh-localhost-cost adds an estimated session cost view to DeepSeek Harness for people running local models. It shows what a session would have cost under the official DeepSeek pricing tables, and it includes a Settings page for mapping local model ids to those billing tiers. It ships as a DSH bundle, so you can install it directly into a profile. When a model id already matches an official tier name, no mapping is needed.

Screenshots

Estimated cost view:

Settings mapping view:

Install

dsh plugin --profile web add dsh-localhost-cost
dsh --profile web --dump-config

This is the native DSH install path. If you only want the npm package as a dependency, npm install dsh-localhost-cost also works, but that does not add the bundle to a profile by itself.

If your profile is different, replace web with that profile name.

After the bundle is installed into a profile, use the Settings page to map local model ids if needed.

The Settings page reads the model catalog already configured in Harness, so you can usually choose a provider first and then pick one of its models from a dropdown. The page writes the mapping for you and refreshes the current session after save, so the cost view updates without a manual browser reload.

What it calculates

The plugin folds logged usage samples into a CNY estimate using the official DeepSeek rates for deepseek-v4-flash and deepseek-v4-pro.

• totalNanoyuan
• inputCacheHitNanoyuan
• inputCacheMissNanoyuan
• outputNanoyuan

It only prices sessions that already contain usage samples. If a local deployment does not log those samples, the plugin cannot infer billable tokens from the session log alone.

Configuration

type SessionCostConfig = {
  aliases?: Record<string, 'deepseek-v4-flash' | 'deepseek-v4-pro'>
}

aliases maps a local or legacy request model id to one of the official DeepSeek billing tiers. The Settings page writes the same mapping for you through the UI, so most users never need to edit JSON by hand.

Supported billing tiers

The plugin currently supports these official tiers:

• deepseek-v4-flash
• deepseek-v4-pro

If you want to price a local model against one of those tiers, pick the tier in Settings and save the mapping. The plugin will then recompute the current session from the existing log.

How to use it

1. Install dsh-localhost-cost.
2. Open the session cost page to see the estimated cost.
3. Open Settings to map local model ids to an official billing tier.
4. Save once. The UI refreshes the current session automatically.

Model Experience

None, as the plugin only computes a client-facing read model of already-logged session events and touches no prompt, message, schema, stream, or tool result.

KV Cache effect

None; the plugin never assembles or sends provider requests.

Known Limitations and Deferred Work

• It only prices sessions that carry usage samples — when a local deployment does not log assistant/chunk or assistant/message usage, the projection cannot infer billable tokens from the session log alone.
• Only two official tiers are supported today — the plugin currently knows deepseek-v4-flash and deepseek-v4-pro; more billing tiers can be added later.
• Unknown model ids still need a mapping — the UI makes this a dropdown workflow, but a local model must still be mapped to one of the supported official tiers before the plugin can price it.
• Peak/off-peak is decided from the request header timestamp — the pricing bucket is based on when the request started, not when the answer finished.