wiki

type Doc ¶

type Doc struct {
	Path    string
	Content string
	Hash    string
}

func LoadAll(wikiDir string) ([]Doc, error)

type Index ¶

type Index struct {
	Docs    []Doc
	Vocab   map[string]int      // term -> index into each vector
	Vectors [][]float64         // one vector per Doc, same order as Docs
	Links   map[string][]string // extracted co-mention edges: path -> paths mentioned by that doc
}

Index is a TF-IDF vector search index over wiki documents. This is a deliberate v1 simplification of the spec's "embeddings index": the spec named semantic search as the goal but did not specify an embedding provider, and adding one would mean either an external API dependency (contradicting "Golem stores no credentials") or an arbitrary vendor choice. TF-IDF cosine similarity gets meaningfully better-than-grep retrieval with zero network calls and zero credentials — pure Go, no cgo.

func Build(docs []Doc) *Index

func EnsureIndex(wikiDir, indexPath string) (*Index, error)

func LoadFromFile(path string) (*Index, error)

func (idx *Index) Search(query string, topK int) []Match

func (idx *Index) SearchExpanded(query string, topK int) []Match

type Match ¶

type Match struct {
	Path  string
	Score float64
	Via   string // non-empty when this result was reached by link expansion, not direct TF-IDF
}