Platform
One engine, every face.
Webfold is a set of tools that share one core. Use the face that fits where you work: a terminal, your own Rust code, an AI agent, a browser driver, an HTTP call or a sandboxed desktop.
Every way in.
Command line →
One binary for scrape, crawl, search, documents, WARC and diagnostics.
Rust library →
Embed conversion, acquisition and search in your own program.
MCP server →
Fetch, search, crawl, capture and more as tools for any MCP client.
CDP endpoint →
A Playwright and Puppeteer-compatible subset on Webfold's own engine.
Intercepting proxy →
Block, pause and log traffic with a forward proxy.
Embedded view engine →
`webfold view` acquires, converts and queries a document's structure with no browser.
Documents and OCR →
Word, Excel, PowerPoint, ebooks and PDFs to Markdown, with OCR for scans.
PII and secrets →
Scan and redact personal data and credentials, as a CLI or an HTTP service.
Corpusforge →
A governed pipeline from raw sources to a deduplicated, redacted corpus.
Glossa →
Localisation extraction, translation and verification in your repository.
Computer use →
Sandboxed desktops an agent drives and a person can take over, over MCP, a JSON-line adapter or a local socket.
Search and research →
Merge web and academic engines, and read social and open-source intelligence sources under a collection policy.
Hosted API and SDKs →
Web, research, bookmark, document, OCR, PII, corpus, localisation and computer endpoints over HTTP, with TypeScript, Python and Rust clients.
API playground →
Run endpoints and see results rendered, or build the commands for the local tools.
Where each one runs today.
Local and hosted
- The command line, library, MCP server, CDP endpoint, proxy, documents and OCR, PII service, Corpusforge, Glossa and the computer-use gateway all run on your own machine or servers.
- The hosted Webfold API, in early access, covers web, research and bookmarks, and also documents, OCR, PII, corpus building, translation checks and sandboxed computers. Document, PII and computer calls send your data to the API; run the tool locally to keep it on your machine.
- The API playground runs the hosted endpoints. For the tools that need your files or folders, such as scanning a directory or starting a gateway, it builds the exact local command.
FAQ
Frequently asked questions
Does Webfold need a browser?
No. Plain pages convert with no browser at all. For pages that build themselves with scripts, Webfold has its own engine that runs them and speaks Chrome's protocol, so Playwright, Puppeteer and Selenium drive it unchanged; it passes all 1,928 tasks of the independent Lexbench benchmark. For the few pages that need canvas, WebGL, web workers or WebAssembly it can connect to a Chromium you run.
What makes it trustworthy?
Four things you can check: the arithmetic at its core is machine-proved from the Rust that runs; one process handles your data with no third-party model in the path; every fetch can be archived byte for byte and every claim carries its source; and every number on this site links to a benchmark with its protocol and limits.
Can I run it myself?
Not yet from a public download. It is one program under the Apache 2.0 licence. The source and first binaries open with the public release. A hosted API is in early access.
How is it different from Firecrawl or Exa?
Firecrawl is a hosted web-data API with a managed browser and curated indexes. Exa is a search API with its own index. Webfold reads, crawls and archives pages with no browser farm and merges the engines you choose. See the comparison pages for the full picture, including where they are ahead.
What do I get back?
Markdown with links resolved, JSON, or chunks sized for retrieval. Tables stay as one block. PDFs, Word and Excel files convert too.
Is the output safe to give straight to an LLM?
Treat it as untrusted. It is text from someone else's page, not sanitised HTML, and not a defence against prompt injection.
How should I read the benchmark numbers?
Each multiple is two measured values divided, on the workload written beside it, with its protocol and limits one click away. The HTML figures are conversion timing on 54 public fixtures. They do not cover networking, browser rendering or whole-service speed, and the PII and Glossa figures were run by those projects, not independently.
What does it cost?
The engine is free to run. Pricing for the hosted API is coming soon; join the waitlist.
How do I get hosted API access?
Join the waitlist or email us. The console is in early access, with subscriptions billed through Stripe.
Pick a face and start.
Run it yourself today, and join the waitlist for the hosted platform.