JohnXu22786/semantic-search
Local semantic code search for DSH (and any Node script): sema builds a fragment-level index of a workspace with symbol-aware chunking and embedding, then serves hybrid vector+BM25 retrieval through the sema_search, sema_reindex and sema_stats tools.
dsh-semantic-search is local semantic code search for DeepSeek Harness (and any Node script). sema builds a fragment-level index of a workspace: source is tokenized by a language-aware tokenizer (camel/snake/kebab splitting and CJK n-grams), chunked with symbol-aware boundaries so functions and classes stay intact, and embedded into fixed-dimension vectors. Queries run hybrid retrieval (vector cosine + BM25, fused with reciprocal-rank fusion) so meaning-based search works even when exact terms don't match. It registers sema_search, sema_reindex and sema_stats on the harness tool registry, ships a standalone CLI, and supports 16 languages plus common data/markup formats. Incremental refresh diffs by size and mtime, an optional watcher keeps the index live, and the same workspace and options always produce the same index and ranked answers.
Install
dsh plugin --profile demo add github:JohnXu22786/semantic-searchCopied verbatim from the README's 'As a dsh bundle' block (it also shows npm install -g dsh-semantic-search and a local-checkout add /path/to/semantic-search). The README notes the npm name is reserved but publishing is pending access setup, so the github: spec is the working route.
Compatibility
Offline by default — the built-in lexical provider needs no network, no model download and no API key; an OpenAI-compatible embedding endpoint can be configured instead, with fallback to the local provider when it is unreachable (allowFallback). The index is stored under <root>/.sema atomically and is deterministic.
Details
- Repo: JohnXu22786/semantic-search
- Category: Tools & Capabilities
- Stars: 2
- Version: GitHub repo JohnXu22786/semantic-search (install is a github: spec to the repo itself; the npm package name dsh-semantic-search is reserved with publish pending access setup). Also ships a standalone sema CLI.
Alternatives
clouwer/dsh-memsearch · tsonglew/dsh-workspace-search · a1073097082/dsh-model-search