altra/web-search-advanced • LM Studio HubWeb Search Advanced Plugin
A streamlined web search plugin designed for minimal context usage while delivering relevant, up-to-date answers.
🎯 Design Philosophy
This plugin was redesigned to solve the "context bloat" problem:
| Original | Optimized |
|---|
| 21 tools | 7 tools |
| ~10KB system prompt | ~500 bytes |
| Verbose outputs | Compact JSON |
| Multiple similar tools | Merged tools |
| Tool | Purpose | Parameters |
|---|
search | General search + read pages | q, n (1-6), time |
deep | Multi-angle parallel research | topic, angles[] |
verify | Fact-check with verdict | claim |
Web Search Advanced Plugin
A streamlined web search plugin designed for minimal context usage while delivering relevant, up-to-date answers.
🎯 Design Philosophy
This plugin was redesigned to solve the "context bloat" problem:
| Original | Optimized |
|---|
| 21 tools | 7 tools |
| ~10KB system prompt | ~500 bytes |
| Verbose outputs | Compact JSON |
| Multiple similar tools | Merged tools |
| Tool | Purpose | Parameters |
|---|
search | General search + read pages | q, n (1-6), time |
deep | Multi-angle parallel research | topic, angles[] |
verify | Fact-check with verdict | claim |
fetch | Read specific URL | url, max |
news | Recent news search | q, time, n |
academic | Search papers | q, source, year |
cite | Generate citation | url, format |
Before: ~10,000 characters (21 tool descriptions + rules)
After: ~500 characters (concise rules + tool summary)
Before: Each tool 200-500 chars description
After: Each tool 50-100 chars description
// Before (verbose)
{
"query": "...",
"total_results_found": 10,
"pages_read": 3,
"independent_publishers_read": 2,
"pages_read_content": [...],
"additional_snippets": [...],
"instruction": "Synthesize across ALL angles..."
}
// After (compact)
{
"q": "...",
"found": 10,
"read": 3,
"publishers": 2,
"results": [{ "t": "Title", "u": "url", "c": "high", "text": "..." }],
"tip": "Cross-reference claims"
}
| Setting | Default | Description |
|---|
| Max Results | 8 | Results per query |
| Fetch Timeout | 8000ms | Page load timeout |
| Time Window | year | Default recency |
| Cache TTL | 30 min | Results cache time |
| Cache Size | 50 | Max cached entries |
{ "q": "latest AI regulations 2024", "n": 3, "time": "month" }
{ "topic": "quantum computing applications", "angles": ["healthcare", "finance", "security"] }
{ "claim": "90% of startups fail in first year" }
{ "url": "https://arxiv.org/abs/2401.xxxxx", "format": "apa" }
| Format | Output |
|---|
| APA | Author (Year). Title. Domain. URL |
| MLA | "Title." Domain, Year. URL |
| Chicago | Author. "Title." Year. URL |
| IEEE | [Author, "Title," Domain, Year. [Online]. Available: URL |
web-search-advanced-plugin/
├── src/
│ ├── index.ts # Entry point
│ ├── config.ts # Configuration
│ ├── toolsProvider.ts # All tools
│ ├── promptPreprocessor.ts # System prompt
│ └── donations.ts # Supporter system
├── scripts/
│ └── add-supporter.js # Add supporters CLI
├── manifest.json
├── package.json
├── tsconfig.json
└── DONATE.md # How to support
cd ~/.lmstudio/plugins/
git clone <repo> web-search-advanced-plugin
cd web-search-advanced-plugin
npm install && npm run build
- Merged Tools:
search_recent → search with time param
- Compact Outputs: Short keys (
t, u, c instead of title, url, credibility)
- Minimal System Prompt: Rules in ~500 chars instead of ~10KB
- Smart Caching: Avoids redundant searches
- Inline Credibility: No separate credibility assessment tool
| Metric | Before | After | Improvement |
|---|
| System Prompt | ~10KB | ~0.5KB | 95% smaller |
| Tool Count | 21 | 7 | 67% fewer |
| Avg Output | ~2KB | ~0.8KB | 60% smaller |
| Context Window Usage | High | Low | Much more room for reasoning |
This plugin is free and open source. If you find it useful, consider supporting development:
| Tier | Amount | Perks |
|---|
| ☕ Coffee | $3 | Name in credits |
| 🍔 Lunch | $10 | + Priority support |
| 🍕 Pizza | $25 | + Name in README |
| 🏆 Founding | $50 | + Lifetime support, consultation |
Every coffee helps keep this project alive! ☕
fetch | Read specific URL | url, max |
news | Recent news search | q, time, n |
academic | Search papers | q, source, year |
cite | Generate citation | url, format |
Before: ~10,000 characters (21 tool descriptions + rules)
After: ~500 characters (concise rules + tool summary)
Before: Each tool 200-500 chars description
After: Each tool 50-100 chars description
// Before (verbose)
{
"query": "...",
"total_results_found": 10,
"pages_read": 3,
"independent_publishers_read": 2,
"pages_read_content": [...],
"additional_snippets": [...],
"instruction": "Synthesize across ALL angles..."
}
// After (compact)
{
"q": "...",
"found": 10,
"read": 3,
"publishers": 2,
"results": [{ "t": "Title", "u": "url", "c": "high", "text": "..." }],
"tip": "Cross-reference claims"
}
| Setting | Default | Description |
|---|
| Max Results | 8 | Results per query |
| Fetch Timeout | 8000ms | Page load timeout |
| Time Window | year | Default recency |
| Cache TTL | 30 min | Results cache time |
| Cache Size | 50 | Max cached entries |
{ "q": "latest AI regulations 2024", "n": 3, "time": "month" }
{ "topic": "quantum computing applications", "angles": ["healthcare", "finance", "security"] }
{ "claim": "90% of startups fail in first year" }
{ "url": "https://arxiv.org/abs/2401.xxxxx", "format": "apa" }
| Format | Output |
|---|
| APA | Author (Year). Title. Domain. URL |
| MLA | "Title." Domain, Year. URL |
| Chicago | Author. "Title." Year. URL |
| IEEE | [Author, "Title," Domain, Year. [Online]. Available: URL |
web-search-advanced-plugin/
├── src/
│ ├── index.ts # Entry point
│ ├── config.ts # Configuration
│ ├── toolsProvider.ts # All tools
│ ├── promptPreprocessor.ts # System prompt
│ └── donations.ts # Supporter system
├── scripts/
│ └── add-supporter.js # Add supporters CLI
├── manifest.json
├── package.json
├── tsconfig.json
└── DONATE.md # How to support
cd ~/.lmstudio/plugins/
git clone <repo> web-search-advanced-plugin
cd web-search-advanced-plugin
npm install && npm run build
- Merged Tools:
search_recent → search with time param
- Compact Outputs: Short keys (
t, u, c instead of title, url, credibility)
- Minimal System Prompt: Rules in ~500 chars instead of ~10KB
- Smart Caching: Avoids redundant searches
- Inline Credibility: No separate credibility assessment tool
| Metric | Before | After | Improvement |
|---|
| System Prompt | ~10KB | ~0.5KB | 95% smaller |
| Tool Count | 21 | 7 | 67% fewer |
| Avg Output | ~2KB | ~0.8KB | 60% smaller |
| Context Window Usage | High | Low | Much more room for reasoning |
This plugin is free and open source. If you find it useful, consider supporting development:
| Tier | Amount | Perks |
|---|
| ☕ Coffee | $3 | Name in credits |
| 🍔 Lunch | $10 | + Priority support |
| 🍕 Pizza | $25 | + Name in README |
| 🏆 Founding | $50 | + Lifetime support, consultation |
Every coffee helps keep this project alive! ☕
