Documentation search and snapshot identity

Documentation search and snapshot identity

Find a contract through documentation tools

EyeAuras Standard Chat searches a local Wiki snapshot. Publishing a page on the website and refreshing that snapshot are separate operations.

Find an operation and read its source text

  1. Enable the knowledge base in AI settings and start a Standard Chat session.
  2. Ask it to call doc_search with the exact API name and task, such as ScriptVariable Listen lifetime.
  3. Read the returned chunkId with doc_get_chunk_full. A heading and short preview help select a result; they do not replace the contract and code.
  4. Cite the result's Link. English pages use /en/... and Russian pages use /ru/...; the repository path ending in .md remains the internal document ID.

Use docName to restrict doc_search to one document. kind and tagsCsv filter metadata actually loaded into the index. Inspect doc_list_documents instead of inferring available tags from the page text.

A long chunk may require several calls. doc_get_chunk_full returns up to 20,000 characters per call and includes a continuation offset when text remains. Pass that offset with the same chunkId until no continuation remains; a partial response can end inside code and must not be treated as the complete example.

Which pages are indexed

Desktop and bot providers exclude pages explicitly marked published: false. An absent flag does not mean unpublished. Published scripting/api/AGENTS.md maps remain documentation; a filename alone does not make a page internal.

The loader reads flat description, kind, tags and published frontmatter fields. Tags accept comma-separated text or a bracket-wrapped list. Frontmatter is removed from searchable chunk text. Nested YAML objects are outside this metadata format.

Refresh the desktop snapshot and identify its version

For Standard Chat, click Save in AI settings, then start a new session or reset the current one. Saving resets the prepared knowledge base; the next initialization updates the Git directory and builds a new index. An existing session keeps its snapshot until it is recreated.

doc_get_index_status returns:

  • SourceRevision: the Git SHA loaded by the desktop provider;
  • IndexHash: the identity of indexed content and indexing settings;
  • DocumentCount, ChunkCount, IsIndexed and IsDirty: the state of this index.

A generic AiDocsKnowledgeBasePlugin created by another application may have no source revision: the source owner supplies it. doc_reindex rebuilds the index from documents already loaded; it does not fetch Git updates.

Rebuild under Guidance pack refreshes a separate set of files for external tools. That pack's SHA does not prove an existing Standard Chat session uses the same index. The website's 15-minute refresh from main also does not refresh an open desktop chat.

AiDocsKnowledgeBasePlugin, Bm25Document, Bm25SearchHit and Bm25IndexStatus are in EyeAuras.AI.BM25. A document's SourceUrl supplies its canonical link independently of its index name; generic loaders can supply their own URL. Publication policy belongs to the source adapter, not the ranking algorithm.