Documentation
¶
Overview ¶
Package index — embedding-backed semantic-search store for the SemanticSearch MCP tool (ADR-014 T6, design from the 2026-04-26 multi-CLI fan-out).
One in-memory chromem-go collection per repo, persisted to disk so `clawtool serve` boot can reload without re-embedding. The index builder walks the repo, chunks each file, embeds via the configured provider (OpenAI default, Ollama override), and adds each chunk to the collection.
Per ADR-007 we wrap [chromem-go](https://github.com/philippgille/chromem-go) (MIT, pure Go, no CGO) for the vector store and the embedding caller. We never reimplement HNSW / cosine / batching.
Index ¶
Constants ¶
This section is empty.
Variables ¶
This section is empty.
Functions ¶
This section is empty.
Types ¶
type Options ¶
type Options struct {
// Provider picks the embedding backend. "openai" uses
// text-embedding-3-small via the user's OPENAI_API_KEY; "ollama"
// uses a local Ollama daemon at OLLAMA_HOST (default
// http://localhost:11434) with the nomic-embed-text model.
Provider string
// Model overrides the per-provider default. Empty = pick from
// provider's stable default.
Model string
// PersistPath, when non-empty, persists the collection to disk so
// boot reloads skip re-embedding. Default
// ~/.cache/clawtool/index/<repo-hash>.gob.
PersistPath string
// MaxFileBytes caps the size of any one file the indexer reads.
// Files above the cap are skipped (binary blobs, generated
// assets). Default 200 KiB — enough for source files, tight for
// build artefacts.
MaxFileBytes int64
// Ignore globs (matched against the path relative to the repo
// root) skip files. Defaults filter common build / vendor /
// .git directories.
Ignore []string
}
Options drive the semantic search pipeline.
type Result ¶
type Result struct {
Path string `json:"path"`
LineStart int `json:"line_start"`
LineEnd int `json:"line_end"`
Snippet string `json:"snippet"`
Score float64 `json:"score"`
}
Result is one ranked hit returned by Search.
type Store ¶
type Store struct {
// contains filtered or unexported fields
}
Store is the single semantic-search index. Methods are safe to call from multiple goroutines after Build returns.
func New ¶
New creates an empty Store rooted at `repo` with the given options. Build populates it; Search queries it.
func (*Store) Build ¶
Build walks the repo and embeds every readable text file. Idempotent when a persisted collection at PersistPath already exists — that path is loaded and Build skips the walk entirely. Operators force a rebuild via `Rebuild`.
Source Files