Articles ยท Gemini AI Tools
Score two texts with embeddings, not with a meaning guarantee
A similarity meter is easy to treat as a plagiarism verdict or a legal sameness test. It is not. Semantic Text Similarity sends Text A and Text B to Google Gemini using the configured embedding model (gemini-embedding-2). DevNestro then computes cosine similarity from the two vectors. The number is a distance in embedding space. The page states that this reflects embedding similarity and is not proof that two texts have the same meaning. Vectors are not written to disk. There is no archive and no analytics of the private passages.
Why this is worth doing in the browser
People search for semantic similarity because they want to know whether two paragraphs “match.” A 0.9 score can still pair a joke with a policy, or two translations that disagree on a key noun. Mixing that score with a local-privacy story is how a visitor pastes customer emails. This route says the opposite: cloud embeddings, a text rate limit, and an instruction not to treat the score as official. Related tools include semantic search for a list of entries and multimodal similarity when you have a document plus a query.
How to use the tool
Paste two texts. Confirm both may be sent to Google Gemini for embeddings. Compare. Read the cosine score and the meaning disclaimer. Clear / Start Over drops the session text.
- Do not paste confidential passages unless you accept Google processing.
- Do not treat a high score as proof of paraphrase, authorship, or legal equivalence.
- A refused request is often a missing key or an empty box.
- Use semantic search when you have many entries and one query.
Privacy
Both texts are sent to Google Gemini for embeddings. DevNestro computes the score in memory and does not persist the vectors. This is not processed only in your browser. Logs may include feature, model, duration, and token counts—not the passages. Free-tier training language follows the configured tier.
Have two texts you are willing to send? Open Semantic Text Similarity and read the score as a hint, not a verdict.