dsh-model-router: A Cost-Aware Routing Workbench for DeepSeek Harness
alice-marx/dsh-model-router
A cost-aware model-routing workbench for DeepSeek Harness that decomposes tasks, assigns models or official CLIs by dependency, previews costs, and controls execution permissions.
dsh-model-router is a model routing and cost planning plugin for DeepSeek Harness that puts the two hardest parts of multi-model work, spend and ordering, onto a single workbench. It decomposes a task locally into work packages and compares the estimated cost of each configured route without making any model call during planning; once confirmed, packages run in dependency order through model APIs or official vendor CLIs. It also offers quality evidence, joint model and effort selection, historical P90 resource estimates, and an editable boundary between the sandbox and vendor environments, with routing profiles kept in a local state file.
How to Install
dsh plugin --profile web add @ljwei-stak/dsh-model-router- Category
- Models & Reasoning
- Platform
- DSH Plugin
- Author
- alice-marx
- Distribution
- Plugin
dsh-model-router Key Features
dsh-model-router Repo Summary
What Does It Do?
Model Router is a DSH plugin for DeepSeek Harness that turns model selection into a cost-aware routing workbench. It solves the problem of unpredictable spend and tangled dependencies when several models collaborate: you decompose a task locally, compare the estimated cost of each configured route, then run the dependent work packages through model APIs or supported official vendor CLIs. Planning makes no model calls, so you can inspect permissions and budgets before any charge is incurred.
Core Features
- Cost-aware routing: algorithm recommendations or manual assignment of each decomposed package to an exact model, with refreshed cost previews and retry targets.
- Dependency-aware execution: packages run sequentially with dependency context preserved, so ordering is never guessed.
- Quality and resource evidence: task-specific quality evidence, joint model/effort selection, and historical P90 resource estimates.
- Explicit execution permissions: editable teams choose either the Harness sandbox or the vendor CLI's normal environment, making the boundary visible.
- Local state persistence: routing profiles and preferences live in a local state file, and API keys are never copied into model profiles or terminal logs.
How to Use This Plugin?
After enabling it in DSH, configure providers, models and credentials on the Models page, then set up routing profiles and preferences in the plugin settings. In daily use, hand a task to the workbench, review the recommended routes and cost preview, and execute the packages in dependency order. Manual mixed execution is read-only and tools fall back to their configured default model. Because planning is local and free, you can compare routes freely before paying for a real run. The plugin is maintained by Alice-Marx under the MIT license and was last updated on 2026-10-20.
This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: alice-marx/dsh-model-router. The plugin is third-party code that runs on your machine once installed; inclusion does not imply endorsement — please review the source before installing.