dsh-model-search

📦 @goodandready/dsh-model-search

---

⚡ Overview & The Problem

As autonomous workflows in DeepSeek Harness scale with dozens of models across diverse local and remote providers (OpenAI, Anthropic, OpenRouter, Ollama, DeepSeek), the default model selector dropdown becomes difficult to navigate. Users often find themselves scrolling through lengthy lists or losing track of exact provider namespaces.

@goodandready/dsh-model-search solves this by injecting an instant, non-invasive, keyboard-friendly live search field directly into the DSH WebUI model picker:

• Instant Filtering: Real-time matching by model name, technical identifier (provider/model-id), or provider group.
• Smart Fuzzy & Acronym Matching: Subsequence detection allows matching queries like dsr1 directly to deepseek-reasoner-1.
• @provider Syntax: Instant filtering of models by specific provider (e.g., @ollama, @openrouter).
• Zero CPU Overhead: Selective MutationObserver guarantees complete zero CPU overhead during chat streaming.
• Full Keyboard Accessibility: Strict arrow-key cycling across filtered items, immediate selection on Enter, and dismiss with Escape.

---

🏗️ Architecture

graph LR
  A[DSH WebUI Dropdown] --> B[lib/client.js]
  B --> C[Search Input Mount]
  C --> D{Query Parser}
  D -->|Text / Acronym| E[Subsequence Fuzzy Matcher]
  D -->|@provider| F[Provider Group Filter]
  E --> G[DOM Visibility Toggler]
  F --> G
  G --> H[Keyboard Navigation & Selection]

---

✨ Core Features

1. Multi-Term & Delimiter-Aware Search

Supports multi-part search terms separated by spaces, commas, slashes, or semicolons:

• Query: openrouter/claude-3-5
• Query: reasoner; 32b
• Query: @ollama llama3

2. Acronym & Subsequence Matching

Matches non-contiguous abbreviations quickly:

• Typing dsr matches deepseek-reasoner
• Typing gpt4o matches openai/gpt-4o-mini

3. Dedicated @provider Syntax

Prefixing a query with @ isolates models belonging to that provider:

• @openai shows only models registered under the OpenAI provider.
• @ollama qwen finds Qwen models hosted on your local Ollama instance.

4. Keyboard Navigation Matrix

Key	Action
ArrowDown	Move active focus to the next visible matched model
ArrowUp	Move active focus to previous visible model; returns to search input at top
Enter	Immediately selects the currently highlighted or top visible model
Escape	Clears active search input; closes dropdown if input is already empty

5. Resilient Core Integration

• Non-invasive DOM manipulation: Only toggles item visibility (display: none / visible); never detaches or rearranges React virtual DOM nodes.
• Resistant to upstream changes: Independent of DSH core CSS module hash changes.
• Theme-Native Styling: Uses official DSH CSS variables to seamlessly adapt to Dark and Light modes.
• Multi-Language Support: English (en) and Chinese (zh) built-in; Russian (ru) dynamically provided when @goodandready/dsh-russian-lang is active.

---

6. One-Click In-App Updater & Settings Card

• Native settings card: Integrates into DSH Settings → Plugins → Plugin Settings (settings.plugin.item) under namespace dsh-model-search.
• In-place updates: Checks the npm registry for new releases, compares versions (including prereleases), and performs one-click updates with loopback-only fail-closed security and single-flight click protection.
• 100% Theme Token Compliance: Fully styled using official DSH design tokens (--dsw-*) with zero hardcoded hex or rgba colors, matching both Dark and Light themes.

📦 Installation

Add to your DeepSeek Harness Web profile:

dsh plugin --profile web add @goodandready/dsh-model-search

Restart DSH or reload the browser interface to activate live search in the model selector.

---

🧪 Verification & Testing

Run unit and syntax checks:

# Verify JavaScript syntax
npm run check

# Run automated test suite
npm test

---

📄 License

MIT © GooDAnDReaDY