A Canadian AI is getting better at finding information in your documents 🇨🇦

An interesting AI story slipped a little under my radar this week.

Cohere just launched Embed 5.

And what is especially interesting for us:

🇨🇦 Cohere is a Canadian artificial intelligence company.

But what does Embed 5 actually do?

🔎 Picture a huge library

Suppose a company has thousands of documents:

• PDFs
• internal policies
• manuals
• financial reports
• spreadsheets
• images
• technical documents

You then ask your AI assistant a question:

“What is our vacation policy?”

Before the AI can answer correctly, it first has to find the right information in that huge library.

That is exactly the kind of problem Embed 5 is meant to help with.

🤖 The AI has to find before it can answer

We hear a lot about models like ChatGPT, Gemini, or Claude that generate answers.

But in many AI systems used in business, there is a very important step just before that:

finding the right information.

Cohere describes Embed 5 as the foundation for improving that search for search systems, RAG systems, and AI agents.

In simple terms:

Your question → 🔎 find the right documents → 🤖 the AI prepares its answer

If the first step brings back the wrong documents, even an excellent AI will have a much harder time giving a good answer.

📄 And documents are not just text

This is probably one of the most interesting parts of Embed 5.

A PDF can contain:

• text
• tables
• charts
• images
• layout and structure

Simply turning the PDF into plain text can lose part of that information.

Embed 5 can work with text, images, and a combination of both. Cohere also points to support for more than 100 languages and documents that can hold a lot of context.

For a company with thousands of complex PDFs, that can become very interesting.

🇨🇦 And it comes from Canada

That is also what makes this announcement worth noting.

Cohere is an AI company founded in Canada and based in Toronto, focused in particular on the needs of businesses and organizations that want more control over their AI systems.

We talk a lot about the American AI giants.

• OpenAI
• Google
• Microsoft
• Anthropic
• Meta

But Canada also has an important player building its own AI technology.

And Embed 5 adds to that ecosystem.

⚡ Two versions

Cohere offers two variants.

Embed 5 Pro is mainly aimed at the best possible quality for analyzing and indexing large document collections.

Embed 5 Fast favors speed and cost when the AI has to run many searches.

The two can even work together: Cohere recommends indexing documents with Pro and using Fast for some real-time searches.

That is especially interesting with AI agents, which may run several searches to complete a single task.

☁️ And it is not limited to Cohere

Embed 5 is available directly from Cohere, but also notably through Amazon SageMaker and Microsoft Foundry. Cohere also offers private deployment options.

For companies already using those environments, integration is much easier.

🍳 The simple version

Picture our library again.

An excellent AI model can be extremely smart.

But if it looks in the wrong book, it may give you a wrong answer.

So Embed 5 is mainly about improving the librarian:

Find the right information, in the right place, before asking the AI to answer.

And that technology comes from a Canadian company. 🇨🇦

Worth keeping an eye on.

Go further

• Cohere’s original article: Embed 5
• What is RAG?
• What is an embedding?