Documentation
¶
Overview ¶
Package model holds the core value types shared across the ingestion pipeline: the source descriptor, the parsed-document representation, and the persistence aggregates (Document, Chunk, Link). Keeping them in one upstream package lets parse/chunk/storage depend on a common vocabulary without importing one another sideways.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
func LastHeading ¶
LastHeading returns the final segment of an "A > B > C" heading path. An empty path yields "". It is the shared helper used to derive a display label (CLI citations, MCP result titles) from a chunk's heading path.
Types ¶
type Chunk ¶
type Chunk struct {
ID string
DocumentID string
Ordinal int
HeadingPath string
Content string
Tags string
DocType string
TokenCount int
StartLine int
EndLine int
MetadataJSON string
}
Chunk is a persisted chunks row. StartLine/EndLine are 1-based inclusive.
type Document ¶
type Document struct {
ID string
URI string
Collection string
ContentHash string
Title string
MimeType string
SizeBytes int64
ModifiedAt string
IndexedAt string
FrontmatterJSON string
}
Document is a persisted documents row.
type Link ¶
Link is a persisted links edge. DstDoc is a plain URI string; the target need not be ingested, so there is no foreign key on it.
type ParsedDoc ¶
type ParsedDoc struct {
Title string
Frontmatter map[string]any
// FrontmatterJSON is the raw frontmatter re-encoded as JSON for
// documents.frontmatter_json. Empty when there is no frontmatter.
FrontmatterJSON string
DocType string
Tags []string
// ModifiedAt overrides Source.ModTime when the frontmatter carries a
// timestamp. Empty means "use the file mtime".
ModifiedAt string
Sections []Section
Links []string
Kind Kind
// StructureOnly marks documents (e.g. OKF index.md) that should yield a
// documents row but no chunks, FTS, or links.
StructureOnly bool
// Metadata is stashed per-document context (e.g. the frontmatter
// `resource` key) propagated into chunk metadata_json.
Metadata map[string]any
}
ParsedDoc is the normalized output of a Parser: title, raw frontmatter, the ordered sections that chunkers consume, and the outbound link URIs. Kind lets chunkers dispatch without re-sniffing the extension.
type Result ¶
type Result struct {
ID string `json:"id"`
DocumentID string `json:"document_id"`
URI string `json:"uri"`
Collection string `json:"collection"`
HeadingPath string `json:"heading_path"`
StartLine int `json:"start_line"`
EndLine int `json:"end_line"`
Snippet string `json:"snippet"`
Score float64 `json:"score"`
}
Result is one ranked retrieval hit: the chunk that matched, its owning document's uri/collection, the heading path and 1-based line range for citation, a highlighted snippet, and the final relevance score (higher is better). It is the value a Retriever returns and the CLI/MCP layers render.
type Section ¶
Section is a contiguous region of a parsed document with a heading path and a 1-based inclusive line range. Markdown parsing emits one Section per heading scope; the text/code parsers emit a single whole-document Section.
type Source ¶
type Source struct {
AbsPath string
URI string
Collection string
Content []byte
ContentHash string
// ModTime is the file mtime in RFC3339, used as modified_at unless the
// frontmatter supplies a timestamp fallback.
ModTime string
}
Source describes a single file handed to a Parser. AbsPath is the absolute filesystem path used to read bytes; URI is the scan-root-relative path stored as documents.uri (the stable, portable identifier). Collection groups documents and participates in the deterministic document ID.
Source Files