LLM-Ready File Specification (LRFS)
Version 1.0 · Released · Published 2026-04-22
LRFS defines how a file embeds a signed, tamper-evident semantic layer in its XMP metadata, so any AI tool can read the file's meaning without re-parsing or re-OCR.
Status
This is v1.0 released. The LRFS namespace URI
https://llmind.org/ns/1.0/ is stable and will remain unchanged throughout
the v1.x family. Breaking changes will use /ns/2.0/ and
leave 1.0 untouched forever.
Scope
The LRFS specifies:
- The XMP namespace URI, fields, and value types
- The per-layer HMAC-SHA256 signing scheme and file-checksum binding
- The version-chain format (immutable layer history)
- How readers detect, validate, and extract layers
Consolidated view
The full specification on one page is the citation-ready URL covering all three chapters in a single document.
Chapters
LRFS v1.0 is organized into three normative specification chapters:
- Payload format — XMP packet structure, RDF layer schema, canonicalization rules, and backward-compatibility for in-file semantic metadata.
- Signing scheme — HMAC-SHA256 and ed25519 per-layer signatures, SHA-256 file checksum, and verification algorithms.
- Conformance — L1/L2/L3 conformance levels for reader/writer implementations, test vectors, and self-reported compliance.
Reference implementation
The Python CLI at github.com/dmitryrollins/LLMind is the reference implementation. Third-party implementations (e.g., in Rust, Go, or JavaScript) are welcomed and encouraged.
Implementations
See the LRFS implementations directory
for a list of known LRFS-conformant libraries. Third-party implementations
are welcome — submit a PR adding a markdown entry to
src/content/implementations/.
Namespace
The stable XMP namespace for LRFS v1.x is https://llmind.org/ns/1.0/.
Roadmap
- Sprint 1: v0.1 stub — scope, namespace reservation
- Sprint 2 (this page): v1.0 release — full specification, three chapters
- Sprint 3: Third-party implementation guidance, test vector publication
- Sprint 4+: arXiv submission, awesome-list entries, independent implementations
Contact
Feedback: spec@llmind.org
Related
- File enrichment glossary — definitions of terms used throughout the specification