Is Firecrawl open source?
Yes: Firecrawl is open source under the AGPL-3.0 license, so the code costs nothing to run on your own servers. The self-hosted stack does not include Fire-engine, Firecrawl's advanced anti-bot service, nor screenshots and page actions, and you pay for the servers, proxies and any AI model yourself. Firecrawl Cloud gives 1,000 free credits a month and prices 100,000 JavaScript pages at $114.00 a month (Standard plan), at a 5% failure rate.
Self-hosting facts from Firecrawl's own README and guides; cloud prices worked out from Firecrawl's pricing page, read on the day its vendor page shows.| Self-hosted (open source) | Firecrawl Cloud | |
|---|---|---|
| Scrape, crawl, map and search | Included in the default stack | Included and managed |
| Advanced anti-bot (Fire-engine) | Not included; a separate service to run and configure | Managed where the Cloud product supports it |
| Screenshots and page actions | Not in the default stack; they need Fire-engine | Screenshots are an output format |
| AI extraction (JSON) | Connect your own OpenAI-compatible model or Ollama, and pay for it | +4 credits per page (on top of 1 credit per scrape) |
| Agent, Browser and dashboard | Not in the default stack | Included by product and plan |
| Security, storage, uptime and upgrades | Yours to run | Run by Firecrawl |
| What you pay | Servers, storage, proxies and the time to run them | 1,000 free credits a month; 100,000 JavaScript pages: $114.00 a month (Standard plan) |
The AGPL lets you run and change the code; if you let others use a modified version over a network, it asks you to share your changes with them. Self-hosting pays off when the code or the data must stay on your own machines and you can run the queue, database and browsers yourself; for a quick start or for sites that block bots, the cloud is the shorter road.
What the open-source code gives you
Firecrawl’s repository says it “is open source under the AGPL-3.0 license”, with the SDKs and some interface components under MIT (Firecrawl on GitHub, checked Oct 12, 2026). The self-hosting guide starts the API, its queue and its database with Docker Compose, and the default stack runs the core scrape, crawl, map and search routes with plain fetching and Playwright (self-hosting guide, checked Oct 12, 2026).
What it leaves out
Three things a buyer comparing it with the cloud should know, all from Firecrawl’s own guides (checked Oct 12, 2026):
- Fire-engine, the service behind Firecrawl’s advanced anti-bot behaviour, “is not included”; it has to be run and configured separately. Sites that block bots are where self-hosting gets hard.
- Screenshots and page actions are “not available in the default stack”, because both need Fire-engine.
- Agent, Browser, Interact, the dashboard and enterprise controls are delivered through the cloud (open source or cloud).
The guide is also plain about the work: “You own upgrades, secrets, storage, monitoring, recovery, and incident response.” AI extraction on a self-hosted stack runs on a model you connect and pay for, an OpenAI-compatible provider or Ollama.
Which costs less
At a few thousand pages a month the cloud’s free credits cost nothing and the self-hosted stack still needs a server. At large volumes on sites that do not block bots, self-hosting can be cheaper than credits, if your team already runs containers and databases. On protected sites the comparison is with the cloud plus Fire-engine, which the open-source stack does not have. The table above prices the cloud for 100,000 JavaScript pages from Firecrawl’s pricing page; the Firecrawl page prices any volume, and the alternatives page lists APIs that cost less for the same job.