Search is how agents find material in a library too big to read whole. Keyword search is on from the start; the other two are opt-in.
Keyword search
Nothing to set up. The Search popup offers By keyword and By meaning; keyword matching covers Markdown, plain text, HTML, and JSON files, plus the text StashBase prepares from PDFs, DOCX files, images, and recordings. Results point back to the visible source file.
Some files need preparation before they can appear. Search reports what is still being prepared, and failed preparation can be retried. Muted files stay out of Search and automatic Chat context.
Search by meaning
Search by meaning finds material that says what you mean in different words. StashBase builds a vector index beside the keyword index and combines both at query time.
Add your own OpenAI or OpenRouter key under Settings → Advanced → Search by Meaning. Your provider’s pricing and data terms apply. Default Agent credits do not cover search by meaning.
With setup complete, StashBase prepares eligible files across Library folders and keeps the index in sync as files change. This runs in the background and is independent of building Wiki Pages.
In Chat, the Search by meaning switch sits in the composer. Turning it off keeps keyword matching, including prepared document text, and does not pause indexing. When credits run out, keyword search and local workflows keep working.
Transcription
For audio or video, download a local speech model from Settings → Transcription. Small (465 MiB) is the default; Tiny (74 MiB) and Base (141 MiB) use less space and trade some accuracy for speed. Transcription runs on your machine with no API cost and produces timestamped text that both kinds of search cover.
Original media stay the visible files. They play directly when the format is supported; otherwise StashBase creates a compatible local audio preview.
What leaves your machine
Keyword search, OCR, transcription, and every file workflow run locally. Hosted search by meaning sends the relevant text to the selected embedding provider for indexing and each query for retrieval. Extracted text and indexes stay in app-managed storage, not in your folder, and removing a folder from the Library clears them without touching your files.
For how the two kinds of search divide the work, read Semantic search across your local files. To put the index to use, go back to building your first Wiki.
Frequently asked questions
Do I need an API key to search my files?
Not for keyword search, which works from the start with no account and no key. Search by meaning needs your own OpenAI or OpenRouter key, added under Settings → Advanced → Search by Meaning.
Can StashBase search scanned PDFs and images?
Yes. PDF text extraction and OCR run on your computer, and both keyword search and search by meaning cover the prepared text. A file that is still being prepared does not appear in results yet, and Search tells you what is pending.
How much does search by meaning cost?
You pay your embedding provider, not StashBase. OpenAI's small embedding model costs $0.02 per million tokens (as of August 2026), so indexing a large personal library costs cents. Default Agent credits do not cover it.
Are my recordings uploaded for transcription?
No. Transcription runs on your machine with a speech model you download once, and there is no per-minute fee. The original audio or video stays where it is.