Documentation for AI assistants

Every documentation page has a Markdown copy at the same path with .html replaced by .md. For example, runtime/intro.html has a runtime/intro.md copy. The HTML page advertises this copy with an alternate link in its head, following the llms.txt proposal.

The Markdown contains the rendered documentation: code included from example files, every language tab, API signatures, tables, and notes. Links and image URLs are absolute, so content remains usable when read outside this website. Section and API headings link to their original HTML anchors for citation.

Choose a resource

These files are generated alongside each version of the documentation:

Resource	Use it to
llms.txt	Find the relevant guide or API page, with a description of each page
index.json	Search document titles, descriptions, and full text, then fetch a Markdown page
index.chunks.json	Retrieve individual sections or API symbols with their heading context and citation URLs
llms-full.txt	Download all Markdown pages, including API references, for local indexing
objects.inv	Look up documented symbols by name
sitemap.xml	Enumerate canonical HTML pages

Start with llms.txt and fetch the pages relevant to the task. The full corpus includes large API references; it may exceed an assistant’s context window. The exported corpus covers the Sphinx documentation. The API guides also link to the separate Rust and Doxygen references.

Match the documentation to the installation

Markdown metadata and JSON records carry documentationVersion, documentationChannel, canonical, and markdownURL. The latest channel tracks development; stable tracks the latest release. Match API usage to the installed LIEF version, platform, architecture, and enabled features.

Use the Installation and Integration guide to choose a build. Read What is LIEF Extended? for Extended features and Runtime for runtime availability. Basic runtime support is an independent build option. Memory-layout enumeration, assembly, and disassembly require Extended.

Retrieve and update chunks

Both JSON indexes declare a schemaVersion. Each chunk contains Markdown and plain text, its document title and description, a headingPath, and a url pointing to the canonical HTML section or API symbol. Chunk sizes follow the documentation structure; code blocks stay intact.

documentID identifies a canonical page, including its version channel. chunkID identifies a section within that page. These IDs remain stable across builds when the page URL and section anchor stay the same. position, documentChunkCount, previousChunkID, and nextChunkID allow retrieval of surrounding context. Positions start at zero; neighbors are empty strings at document boundaries.

Compare the index’s corpusHash to detect changes to content or metadata, then compare individual contentHash values to find changed Markdown bodies. Both use SHA-256. The chunk index describes the hash inputs and ordering. Build timestamps are excluded so unchanged content produces the same indexes.