Research Tools
Evaluering og sammenligning af forskningsværktøjer i Hermes-økosystemet. Fra daglig web search til avanceret semantic crawling.
📋 Forskningsværktøjer
| Værktøj | Type | Status | Setup | Pris | Bedst til | Testdato |
|---|---|---|---|---|---|---|
| Firecrawl | Web scraping | Useful | API key | $0.001/page | Struktureret web scraping, markdown extraction | 2026-05-08 |
| Exa | Semantic search | Monitoring | API key | $0.001/req | Semantisk søgning, finding relevant indhold | 2026-05-08 |
| Tavily | AI web search | Useful | API key | $0.001/req | AI-optimeret websearch, research | 2026-05-08 |
| web_search (built-in) | Web search | Adopted | Pre-installed | Inkluderet | Daglig søgning, quick research | 2026-05-08 |
| web_extract (built-in) | Content extraction | Adopted | Pre-installed | Inkluderet | URL → markdown, hurtig indholdshentning | 2026-05-08 |
| browser (Playwright) | Browser automation | Useful | Pre-installed | Inkluderet | JavaScript-sider, screenshots, interaktion | 2026-05-08 |
| Parallel Search | Parallel crawl | Monitoring | Config | Variabel | Multi-URL samtidig hentning | 2026-05-08 |
📊 Evalueringsresultater
✅ Firecrawl
Markdown extraction excellent. Rate limits handled well. Best for structured scraping. Use when web_extract fails on JS-heavy sites.
⚠️ Exa
Semantic search quality varies. Good for discovery but results need filtering. Monitoring cost efficiency before broader adoption.
✅ Tavily
AI-optimized search returns cleaner results than raw web_search. Good for research tasks. Worth the API cost for quality.
✅ web_search
Reliable daily driver. Covers 90% of search needs. No API key cost. First choice for all standard search queries.
✅ web_extract
Excellent for simple pages. Struggles with JS-heavy SPAs. First choice for URL→content conversion.
✅ browser (Playwright)
Full Playwright control. Best for screenshots, JS rendering, form filling. Higher latency but maximum capability.
⚠️ Parallel Search
Useful for multi-URL batch work but setup complexity is high. Monitoring adoption — use only when batch processing is necessary.
🎯 Anbefaling
Anbefalet workflow
web_search + web_extract (gratis, hurtigt, pålideligt)
Firecrawl (når web_extract fejler på JS-sider)
Exa (når du ikke ved hvad du leder efter)
Specialtilfælde
Tavily (renere resultater end raw search)
browser + screenshots
Parallel medmindre batch-behandling er nødvendig