llms.txt solved a real problem. It gave AI tools a way to understand a website. It has some traction. But there's nothing equivalent for people. I think this might be a thing. Let's see what HN thinks.

Every AI tool has some version of "custom instructions" but they're locked inside that tool. Switch from ChatGPT to Claude to Gemini and you're starting from scratch each time. Your voice, your preferences, your expertise, all gone.

identity.txt is a plain text markdown file (same philosophy as llms.txt, humans.txt, robots.txt) that captures who you are for AI tools. Drop it at yourdomain.com/identity.txt or paste it into any context window.

The spec is deliberately minimal:

  - H1 with your name
  - H2 sections for Voice, Expertise, Background, Preferences, Terms
  - A Terms section with machine-readable consent (open, attribution, prompt-only,
  restricted, none)

The Terms section is interesting, maybe. It's like robots.txt for personal identity. A social contract, not a legal one. We think consent and identity overlap in ways worth exploring. We def don't have the answers now. But open to the questions.

We're also experimenting with hosted identity files at identitytxt.org. You authenticate with Google, pick a handle, and get a permanent URL. But self-hosting is the primary use case. The hosted version is just convenience. Maybe. We don't know.

Look, this is early and experimental. We built it because we kept pasting the same context into every AI conversation and thought there should be a convention for it. The spec is CC-BY 4.0 and on GitHub: https://github.com/Fifty-Five-and-Five/identitytxt

We're genuinely interested in whether this resonates or if the problem gets solved differently. What are we missing?

https://identitytxt.org