Search by meaning,not by file name.
DesignVault's AI semantic search reads a plain-language description — “dark hero we made for the fintech launch” — and returns the assets that match the idea. No tag required. No naming convention to remember.
3 matches · 0.3s · by meaning
- 94%
Hero — Nova launch
figma · website
- 91%
final_v3_FINAL
figma · website
- 87%
Untitled frame 47
figma · website
Your best work is filed under final_v3_FINAL.
Nobody names files for the person searching six months later. Keyword search can only find the words someone remembered to type. Semantic search finds what the asset is.
0 results
The frame exists. It's just named final_v3_FINAL and nobody tagged it.
- final_v3_FINAL94%
- Untitled frame 4789%
- hero-export-285%
same query · same library · 0.3s
Embeddings do the remembering for you.
No setup, no training run, no prompt engineering. Import an asset and it's searchable by meaning within seconds.
- 01
Every asset becomes a vector
On import, DesignVault reads the thumbnail, title, tags, and description, and computes an embedding — a numeric fingerprint of what the asset means, not just what it says.
- 02
Your query becomes one too
Type what you remember in plain English or French. The query is embedded the same way, in real time, before it ever touches the index.
- 03
Closest meaning wins
A cosine-similarity scan ranks every asset in your org by semantic distance, fused with classic full-text rank. Best matches first, scored as a percentage.
Meaning and keywords, fused in one query.
Semantic search doesn't replace exact matching — it covers its blind spots. When you search, DesignVault runs both engines and merges the rankings: precise hits for precise queries, conceptual hits for vague ones.
Exact when you are. Search “Q4-banner-728x90” and full-text rank puts the literal match first.
Fuzzy when you're not. Search “that warm illustration style from last summer” and vectors take over.
Bilingual either way. Query in French, find assets described in English. Embeddings don't care which language the tag was written in.
one query → two engines → one ranking
- titlets_rank
- tagsts_rank
- clientts_rank
- descriptionts_rank
- visual contentcosine
- subjectcosine
- stylecosine
- intentcosine
One result list, ordered by combined relevance — never two tabs to check.
Social — spring campaign
instagram · 1080×1350
Viewing one asset? The sidebar already found its visual siblings.
- 93%
- 88%
- 84%
One good asset leads to the rest.
The same embeddings power a “similar assets” panel on every asset page. Found one banner from the campaign? Its siblings — same palette, same style, same subject — are one click away, each with a similarity score.
It also works in reverse: before importing a near-duplicate, DesignVault flags assets that already cover the same ground. Your library stays lean without anyone policing it.
Read the asset docsThe questions designers actually ask.
Semantic search ships with the Business plan — compare all plans.
Something missing? Write to hello@designvault.net — a human answers.
- How is semantic search different from keyword search?
- Keyword search matches the exact words stored in a title, tag, or description. Semantic search converts your query and every asset into vectors and compares them by meaning — so "dark hero for the fintech launch" finds the right frame even if it was saved as final_v3.fig with no tags at all.
- What data is sent to the AI model?
- Only the asset’s title, tags, description, and thumbnail are used to compute its embedding. Embeddings are stored in your organization’s row-scoped tables, isolated per tenant. Your assets are never used to train any model.
- Which plans include AI search?
- Semantic search ships with the Business plan ($199/month, unlimited assets). Full-text search across titles, tags, clients, and descriptions is included on every plan, Free included.
- Does it work in French and English?
- Yes. Embeddings are language-agnostic, so a French query finds assets described in English and vice versa. Keyword search uses a bilingual dictionary too — DesignVault is built by a bilingual team, for bilingual teams.
- Do I still need tags?
- Yes — tags power filters, smart collections, and team conventions. Semantic search complements them: it catches everything tags miss, and AI auto-tagging suggests tags on import so the library stays structured without manual effort.
- How fast is it?
- One embedding lookup against an indexed vector column. Results return in well under a second, fused with full-text rank in a single query.
Stop guessing file names.Start describing what you remember.
14-day trial · no credit card · 100 assets free forever