You can use generated results for commercial purposes. You own all rights to the content you create. The only restriction: make sure your prompt does not include copyrighted third-party material. You are responsible for respecting the rights to any input data.

It is an embedding model: it converts text into a numerical vector. These vectors are used to build semantic search, RAG pipelines, clustering, deduplication, and document classification. The context is 512 tokens, so split long materials into chunks beforehand and index them separately.

The output is not text or an image, but an embedding — an array of floating-point numbers with a dimension of 1024. This vector is placed directly into Qdrant, pgvector, or another vector storage, and you send requests via the unified OpenAI-compatible BotHub API.

There are no reasoning chains here: it is an encoder that encodes text into a vector in a single pass, rather than writing an answer. Reasoning mode is not indicated in our data. You build the ranking and re-ranking logic on your side.

This capability is not indicated in our data, and an embedder usually doesn't need it: the model does not call external tools. However, the API response comes structured — it is a JSON with an array of numbers, which is easy to parse and write directly to a vector database.

Acceptance of such files is not indicated in our data — provide text as input, up to 512 tokens per request. If a document is longer, cut it into overlapping chunks and index each fragment as a separate vector, otherwise, some meaning will be lost.

Russian is native to it, along with dozens of other languages — it is a multilingual model of the E5 family. A query will find a suitable fragment in a Russian-language database, and cross-lingual search also works: vectors of different languages fall into one common semantic space.