Semantic Tool Search by Description
Describe your task and search all 136 tools by meaning — ranked matches with match reasons. Free, no sign-up, no external AI API.
Natural-language tool finder powered by local embeddings (no external AI API).
What is it? Semantic Tool Search lets you describe what you need in plain language and returns the best-matching Vecaria tools, ranked by embeddings served on Vecaria's own infrastructure. A relevance floor (currently 0.70, calibrated on real vs. garbage query score distributions) withholds weak matches — those get a keyword-search pointer instead — and every result card shows the tool's description plus why it matched. A transparency block on first paint states the catalog size, the embedding model and the ranking method. It searches Vecaria's own tool catalog — not your documents or code — so it finds tools, not information.
TL;DR: Search tools by describing your task in natural language — ranked matches with match reasons, free and no sign-up.
How to use
- Describe the outcome you want in one sentence — e.g. "turn scanned pages into text". Tool names are not required.
- Review the ranked tool matches — each card shows the tool's description and why it matched.
- Open the tool you need; if nothing clears the relevance floor, follow the keyword-search pointer.
FAQ
Does it use an external AI API?
No — matching is based on open embeddings (bge-small-en-v1.5, MIT) served by Vecaria's own VPS; your query text leaves your browser but is not stored.
Why do I sometimes get no results at all?
Semantic search only returns high-relevance results: candidates below the relevance floor (currently 0.70) are withheld rather than shown as noise. The page then points you to keyword search.
What does "semantically similar" mean?
The result is close to your description in embedding space but shares no literal wording with the catalog — open the tool page to confirm it does what you need.
What are the limits?
20 queries per minute per IP and up to 200 characters per query. There is no Pro cloud path and no cloud quota is consumed.
What if it is unavailable?
Use the standard keyword search; semantic search depends on the Vecaria search service being online.
How is this different from the site's keyword search?
Keyword search matches literal tool names and categories — fastest when you already know roughly what a tool is called. Semantic search reads your whole task description and ranks tools by meaning, so it works when you do not know the name; weak matches are withheld rather than shown as noise.
How is this different from the AI Tool Assistant?
Semantic search returns a ranked list of matching tools with match reasons, and needs no key — ideal for quick discovery. The AI Tool Assistant plans one step deeper: it returns a specific tool with parameters or an ordered step sequence, using your own LLM key. Start here, then use the assistant when you need parameters.
Do I need an account or an AI API key?
No — it is free with no sign-up and needs no LLM key. Ranking runs on open embeddings (bge-small-en-v1.5, MIT) served on Vecaria's own infrastructure, not on an external AI API, and queries are not stored.
Which tools does it search?
The full Vecaria catalog of browser-based file tools — CAD and 3D, GIS, PCB, CNC, PDF and document utilities. The transparency block on the page shows the current enabled-tool count, and each result links straight to that tool's page.
Examples
- Type "turn scanned documents into text" → get image OCR and similar tools
Limitations
- Your query text is sent to Vecaria's own VPS for embedding-based ranking — it leaves your browser; nothing is stored, and there is no Pro cloud path.
- Limits: 20 queries per minute per IP, up to 200 characters per query.
- Relevance floor is currently 0.70 (cosine similarity); results below it are not shown — meaningless queries return an empty result set with a keyword-search pointer instead of high-scoring noise.
- Purely semantic matches (no literal term overlap) are labeled "semantically similar" and are not guaranteed to match your intent — confirm on the tool page.
- Catalog size is finite (the transparency block shows the current enabled-tool count).
- Requires the Vecaria semantic search service. Results are tool suggestions, not generated answers.
Privacy
Your query text is sent to our own VPS for embedding-based ranking (no external AI API): nothing is stored, there is no Pro cloud path and no cloud-task allowance is consumed; real limits: 20 queries per minute per IP, up to 200 characters per query.