FAQ¶
Is MFS a filesystem?¶
No. MFS does not mount, move, sync, or store your files. It builds a search index over files that already exist on disk.
Does MFS modify my project?¶
No. Runtime state lives under ~/.mfs/ by default. Your project directory is
not given generated metadata files.
Why does mfs search require a path or --all?¶
Explicit scope prevents surprising searches against the wrong corpus. Use:
What happens if background indexing is interrupted?¶
The queue is lightweight local state. Re-run mfs add . to continue, or use
mfs add . --force when you want a stronger rebuild pass.
Are PDF and DOCX files supported?¶
Yes. PDF uses pymupdf4llm; DOCX uses python-docx. Both are converted to
Markdown before chunking and cached in ~/.mfs/converted/.
Are JSON, JSONL, and CSV indexed semantically?¶
Not by default. They are readable with mfs cat and searchable with mfs grep.
This avoids embedding noisy structured data unless a future workflow chooses to
index it explicitly.
Does MFS call an LLM while indexing?¶
Not by default. Embedding is required; LLM summaries and VLM descriptions are opt-in.
Can I use Zilliz Cloud?¶
Yes. Set milvus.uri to your Zilliz Cloud endpoint and milvus.token to the
token.