Skip to content

index-ai / Compare

index-ai vs llms.txt

Different problems, different tools — and they coexist. Here is exactly what index-ai does that llms.txt does not, so you can decide with specifics rather than vibes.

The comparison ​

index-aillms.txt
What it isA structured, open specification proposalMachine-readable JSON Schema, published and tested with the specA human-readable text fileMarkdown text with links, suggested by a blog post
FormatValid JSON/.well-known/index-ai.json + /agent-index.json, both validated against the official JSON SchemasPlain text / Markdown/llms.txt, no schema, no validation, no structure contract
VerificationMeasured and checkedcontent_chars is an NFC code point count (exact or bounded max) verified against the served content; conformance is graded by a validatorDeclared, never verifiednothing in the file is checked against the actual site content
Budget signalKnown before fetchevery measured node declares its content size (exact or max bound); the agent budgets tokens before fetchingUnknown until fetchedan agent must fetch each linked page to learn its size
StructureNodes + optional graphtyped relationships (2b) let agents traverse without parsing HTMLFlat list of linkssections and links, no data model, no relationships
Query interfaceLevel 3 (MCP)an optional typed API over the Agent ViewNonea text file, no query interface defined
PolicyMachine-readablepolicy.usage_preferences declares search/summarization/citation preferencesAbsentno machine-readable usage or citation policy

The one-line difference

llms.txt tells an agent what links exist; index-ai tells an agent what content exists, how big it is, how it is structured, and how to query it — measured and verifiable.

Do they coexist? ​

Yes — index-ai is additive. Implementing index-ai MUST NOT require changes to your existing robots.txt, sitemap.xml, or llms.txt. You can run all of them at once; they solve different problems:

Same origin — four conventions, one job each.

Your siteone origin · one server
robots.txtsays what agents cannot access
sitemap.xmllists URLs
llms.txtgives human-readable context
index-aigives structured, measured, queryable content

All four sit on the same origin, side by side. index-ai is the only one that is measured and machine-queryable.

Not a replacement, not a fork

index-ai is a distinct, deliberate convention — not an llms.txt variant. It even offers an optional bridge: an llms.txt file may point to the index-ai manifest. The normative statement lives in §16 Compatibility of the specification.

Demonstration ​

The AI-readiness walkthrough — the Agent View in action.

Retry it yourself now