In plain words
Cohere is a company, based in Toronto, not a chatbot you download. It sells AI to businesses and governments, not to the general public. They use its models inside their own software and tools.
It makes three main kinds of model. Command models are large language models that write and answer. Embed models turn text into lists of numbers (embeddings) so a computer can find documents by meaning. Rerank models sort search results so the most useful ones come first.
Used together, they let a company build an assistant that answers from its own files, with references back to the source, a pattern called RAG. Cohere's selling point is privacy: its models can run inside a customer's own data centre or private cloud, so sensitive data stays in-house.
Everyday example
A law firm's best junior
A senior lawyer asks: "What did we agree with this client on late fees?" A good junior pulls every relevant file, puts the three best on top, and writes a short answer with page numbers. Cohere's models split that job: Embed finds, Rerank sorts, Command writes the answer with citations.
Where the example breaks: A junior lawyer understands the law and the client. The models only match patterns in text, so an answer can cite the wrong page or miss a file, and someone still has to check.
How it actually works
A typical Cohere set-up is a search pipeline with a writer at the end.
Index the documents
Each document is split into chunks and passed through an Embed model, which turns each chunk into a vector. These vectors are stored in a vector database. Similar meanings end up close together.
Find candidates
A question is embedded the same way, and the database returns the chunks whose vectors sit nearest to it. This is semantic search: it can find "late payment charges" even when the question says "late fees".
Rerank
A Rerank model reads the question and each candidate together and scores how well they really match, then reorders them. It is slower per document than embedding, so it runs only on the shortlist.
- Founded
- 2019 in Toronto by Aidan Gomez, Nick Frosst and Ivan Zhang
- Link to the transformer
- Aidan Gomez was one of the eight authors of the 2017 paper that introduced the transformer
- Model families
- Command (writing and agents), Embed (search), Rerank (sorting), Aya (many languages)
- Customers
- Businesses and governments, rather than individual consumers
Cohere's research arm releases Aya, multilingual models covering many languages, some with open weights for research.
Questions people ask
Does Cohere have a chatbot like ChatGPT?
Not for the public. Its workplace product, North, gives employees a private assistant and agents that work with company data. Cohere sells to organisations, unlike ChatGPT, which anyone can sign up for.
What is Cohere Command?
Command is Cohere's family of large language models for writing, answering, summarising and calling tools. Earlier versions were called Command R; newer ones carry the name Command A.
What is Cohere Rerank used for?
It improves search. After a fast search finds possible matches, Rerank reorders them by real relevance to the question. Many companies add it to existing search systems without changing anything else.
Where is Cohere based?
Cohere was founded in Toronto, Canada, and has offices in North America, Europe and Asia. A planned combination with Germany's Aleph Alpha would add Berlin as a second headquarters; as of November 2026 the deal is waiting for regulatory approval.
Was this clear?
1 is confusing, 5 is crystal clear. One tap, no account needed.
Spotted something wrong, or know a better example? Suggest a change