I help people make their websites and APIs work with AI agents. I run turva.dev, where I test what an agent can access and work out what needs changing when it gets stuck.
I also build the tools below. The code is open source so you can try it on your own site or see how mine works.
Shopify agent storefront check, audit, advisory, implementation, agent operations and MCP server design.
| Project | What it does | Try it |
|---|---|---|
| markdown-parity-check | Compares the main content of HTML and Markdown pages, with source locations for differences and JSON output for CI. | npm package · Hosted check, turva.dev pages only |
| llms-txt-validator | Checks llms.txt structure from the command line or Node, with JSON output for CI. | npm package · Hosted validator |
| turva-worker | Runs my website, serving HTML and Markdown from shared content sources. It also publishes the site's discovery metadata. | Live site |
| turva-mcp | Gives MCP clients read-only access to turva.dev's published information and evidence. | Connect your client |
The two command-line tools are available on npm. With Node.js 22 or newer installed, replace the example addresses with your own domain and page:
npx --yes turva-llms-txt-validator example.com
npx --yes markdown-parity-check --url https://example.com/pageThe first command checks llms.txt structure. The second compares a page's HTML and Markdown responses. For a separate Markdown address, add --markdown-url https://example.com/page.md to the second command. Use the command for your own site, since the hosted comparison only accepts turva.dev pages.
I use turva.dev as a reference implementation. Its source code and verification instructions are public, along with dated measurements and their limits. You can inspect the implementation and repeat the checks.
In the scan dated 2026-09-14, turva.dev scored 100/100 and reached Level 5 on isitagentready.com. The security scans from the same day passed all 24 categories on Hardenize and gave 98/100 on the Internet.nl website test.
I write about what I build and what happens when agents try to use websites.
- llms.txt explained.
- The /.well-known directory for agent discovery.
- What a website and API agent-readiness audit covers.
- HTML and Markdown can disagree.
- The twin is the page.
- Thirty-day follow-up: 201 comparable readings from 210 sites.
- What four AI assistants call an agent readiness audit.
- Website agent readiness, measured on 567 company sites.
I'm based in Tampere, Finland. I work in writing and explain the findings so the people maintaining a site can follow the reasoning and test the changes themselves.
Send me your website and tell me what you're trying to do. That's enough to start: info@turva.dev.