Semantic search
Search & navigation ›Semantic search
April 7, 2026Updated: April 7, 20265 min read1 views
Find documents by meaning not just exact words
Semantic search
Semantic search, AI-based, understands the meaning of your question — not just the exact words.
When to use it
Semantic search is ideal when:
- You don't remember the exact words in the document
- You're searching for a concept, not a specific term
- You want to find documents similar to something you have in mind
Practical examples
| You type | What it finds |
|---|---|
| "documents about delay penalties" | Contracts with penalty clauses, even if they don't contain the word "penalties" |
| "agreements expiring soon" | Contracts and certifications expiring in the coming months |
| "quality issues with suppliers" | Reports, complaints and communications about supply defects |
| "how much do we pay for maintenance" | Invoices, contracts and quotes for maintenance services |
Full-text vs Semantic
| Full-text | Semantic | |
|---|---|---|
| Searches for | Exact words (with stemming) | Meaning and concepts |
| Speed | Milliseconds | Slightly slower |
| Prerequisites | None (always active) | AI connector with embedding |
| Best for | Precise terms, codes, names | Concepts, questions, exploratory searches |
How it works
- Your query is transformed into a numerical vector (embedding)
- The vector is compared with those of all indexed documents
- Documents with the most similar meaning appear first
Hybrid search
Theka automatically combines full-text + semantic in every search. The final ranking considers:
- Text relevance — word matching
- Semantic similarity — meaning proximity
- Boost — current version, final/signed status, active project, recent document
You don't have to choose which to use: type your search and Theka uses both.
💡 Tip: If semantic search is not active, Theka uses only full-text (no errors, still valid results). To activate semantic search, the administrator must configure an AI connector with embedding capability.