Frequently Asked Questions
Common questions about AI-Me, its scope, and how it works.
What is AI-Me?
AI-Me is an open specification that defines a machine-readable "vendor card" — a JSON document served at /.well-known/ai-me.json. It provides one-fetch, site-wide discovery for AI agents: who you are, what you offer, how to engage.
Who is this for?
Any website or service that wants to be discoverable by AI agents. SaaS companies, open-source projects, e-commerce stores, consulting firms, government portals — if you have a web presence, you can publish a vendor card.
Is this a replacement for schema.org?
No. Schema.org provides page-level vocabulary consumed by search engines. AI-Me provides site-level discovery for AI agents. They're complementary — you can reference your schema.org data from the vendor card's related field. See Why not schema.org? for more detail.
Does this compete with llms.txt?
No. llms.txt is human-readable guidance for LLMs. AI-Me is machine-readable structured data. Use both: include a pointer to your vendor card in your llms.txt file.
What about MCP and A2A?
MCP (Model Context Protocol) handles tool integration; A2A handles agent-to-agent communication. AI-Me handles discovery — it can point agents to your MCP server or A2A agent card via the related field.
Is this spec stable?
v0.1 is a draft. We're collecting community feedback before moving to v1.0. The core shape (required fields, canonical_pages, offerings) is unlikely to change significantly, but details may evolve. See the roadmap.
What fields are required?
Six fields: schema_version, name, website, summary, canonical_pages (with home and contact), and last_updated. Everything else is optional but recommended.
What if I have no commercial offerings?
The offerings field is optional. Open-source projects, government sites, and personal sites can publish a conformant vendor card without it.
How do I validate my vendor card?
Use the CLI: npx @ai-me/validator https://yourdomain.com, or paste your JSON into the online playground.
Can I add custom fields?
Yes. The spec uses an extension mechanism: add any fields you need, but namespace them with a prefix (e.g. x_mcp, x_mycompany) to avoid collisions. Clients MUST ignore unknown fields.
How do I contribute?
See the contributing guide. Open issues, propose changes via pull requests, and join the discussion on GitHub.