Lesson 4 — Cloud vs Self-Hosted: How to Choose?
Before deciding between Firecrawl Cloud and Self-Hosted, you need to weigh your options across three dimensions: features, performance, and cost.
4.1 Feature Comparison Matrix
| Feature | Cloud | Self-Hosted |
|---|---|---|
| Basic Scraping (Scrape) | ✅ Fully Supported | ✅ Fully Supported |
| Search Engine (Search) | ✅ Built-in | ❌ Requires external Search API |
| Research Agent (Agent) | ✅ Exclusive Feature | ❌ Not Supported |
| Residential Proxy Pool | ✅ Built-in 3 levels | ❌ Must provide your own |
| Structured Extraction (LLM) | ✅ Ready to use | ⚠️ Requires own LLM API Key |
| Browser Interaction | ✅ Ready to use | ⚠️ Requires Playwright service |
| Maintenance Cost | Zero Maintenance | Requires server management |
4.2 Billing Model and Cost
Cloud Billing (Credits)
Firecrawl Cloud uses a Credit system. In most cases, 1 credit corresponds to 1 page scraped.
- Scrape/Crawl/Map: 1 credit / page.
- Search: 2 credits / 10 results.
- Interact: 2 credits / browser minute.
Cost Analysis
- Cloud: The Hobby plan ($19/mo) includes 3,000 pages, suitable for small to medium-sized projects or those needing rapid deployment.
- Self-Hosted: You bear the server costs (approx. $20-$100/mo) and proxy IP costs. Self-hosting usually becomes economically advantageous only when monthly volume exceeds 500,000 pages.
4.3 Decision Tree
Use this logic to quickly decide:
- Do you need Search functionality?
- Yes → Cloud is preferred (Self-hosted lacks a search index).
- Is your monthly volume less than 100,000 pages?
- Yes → Cloud is preferred (Save time on maintenance).
- Does your data have strict privacy or compliance requirements?
- Yes → Self-hosting is mandatory.
- Do you have strong DevOps skills and seek the lowest possible cost?
- Yes → Self-hosting.
4.4 Cost-Saving Tips (Cloud)
- Enable Caching: Use
storeInCache: true. Repeated requests for the same URL within the cache period do not consume credits. - Targeted Scraping: Use Map to discover URLs first, then use
includeTagsto fetch only the core content. This reduces downstream token consumption (saves AI costs). - Remove Images: Setting
removeBase64Images: truecan significantly speed up response times.