fastCRW is an open-source Firecrawl and Tavily alternative — a single 8 MB Rust binary that gives you scrape, crawl, search, and extract APIs behind one drop-in /v1 endpoint. Built for RAG pipelines, LLM agents, research assistants, and any product that needs to read the live web and turn it into clean, model-ready markdown.

The /v1 surface is Firecrawl-compatible, so existing Python and Node SDKs swap in with a base URL change. Scrape returns LLM-ready markdown with optional screenshots, HTML, and link graphs. Crawl walks full sites with sitemap support, robots.txt, dedup, and streaming results. Extract pulls structured data via JSON schema or natural-language prompt. Map returns the URL graph without fetching every page.

The same binary exposes a Tavily-compatible AI search API. Each result already comes back as full clean markdown with relevance scoring — one round trip, no second scrape pass, no second vendor bill. Perfect for LangChain, LlamaIndex, CrewAI, n8n, Dify, and any agent loop where latency kills UX.

Performance is the headline. On a public 1,000-URL benchmark, fastCRW hits 710 ms p50 and 833 ms average — roughly 17× faster than hosted Firecrawl on the same workload. Idle RAM is 6.6 MB, the Docker image is 8 MB, and cold start is sub-second. A $5 VM runs a real production scraping and search API.

Dual-mode by design: managed cloud with 500 free credits per month and linear paid tiers, or self-host the entire binary under AGPL-3.0 with zero per-request fees. Credits work across scrape, crawl, search, and extract — one bill, not a stack of vendor invoices.

Open-source web scraping API and AI search API in one Rust binary. Firecrawl alternative, Tavily alternative, self-hostable, and built for the LLM era. Start free at fastcrw.com or clone the GitHub repo and run it yourself.