Add Cohere Provider support

[infinityrobot]

What this does

This PR adds a new Provider for Cohere and adds 21 Cohere models across multiple model families that support chat, embeddings and vision, including:

• Command models – including Command A (256K context) and Command R (R, R7B and R Plus)
• Aya Expanse (multilingual)
• Aya Vision (multimodal)
• Embed 4 (text embeddings)
• Fine-tuned models created and managed on Cohere's platform

Implementation overview

Added a new RubyLLM::Providers::Cohere module that follows existing Provider patterns with API key configuration and Capabilities, Chat, Embeddings, Media, Models, Streaming and Tools modules.

All features have full test coverage with VCRs. Relevant documentation has been updated.

Key highlights enabled by adding Cohere as a Provider:

• Chat completion with streaming, tool and tool support with Command and Aya models
• Text embeddings with Embed models, including multilingual and image support (to be included in future contribution)
• Vision capabilities with Aya Vision models
• Ability to use fine-tuned models using Cohere's platform (leveraging assume_model_exists: true)

Note

I have opted to use Cohere's native API endpoints rather than their OpenAI-compatible interface so we have a more future-proof implementation and the optionality to support Cohere specific features across chat (documents, citations, safety mode, etc.), embeddings (input types, image embeddings, etc.) and other features like reranking and classification.

Usage example

Usage follows the usual RubyLLM::Providers patterns.

# Configure Cohere API key
RubyLLM.configure do |config|
  config.cohere_api_key = ENV['COHERE_API_KEY']
end

# Chat with Command models
chat = RubyLLM.chat(model: "command-a-03-2025")
chat.ask("Why is Ruby an incredible programming language?")

# Generate embeddings
RubyLLM.embed("Ruby", model: "embed-v4.0")

# Vision capabilities
chat = RubyLLM.chat(model: "c4ai-aya-vision-32b")
chat.ask("What is in this image?", with: { image: "spec/fixtures/ruby.png"  })

Other changes / bug fixes

• Fixed a bug in generating models.json due to Parsera returning models with a nil id.
  • Also raised in Reject models from Parsera that does not have :provider or :id #271
• Refactored spec/ruby_llm/embeddings_spec.rb to use an EMBEDDINGS_MODELS constant defined in spec/spec_helper.rb to align specs with chat_spec.rb and handle per-model dimensions. The 768 previously hardcoded is invalid for Cohere models

Future plans

Note – this PR is the first in a series I intend to contribute. In the future I would love to also add support for the below features that I think would be useful to other RubyLLM users if you are open to it @crmne!

• Reranking – via Cohere, TogetherAI and other potential providers
  • Proposed Cohere Rerank implementation ready in Add global Rerank interface with Cohere Rerank model support #276
• Classification – via Cohere
• Image / multimodal embeddings – via Cohere, Ollama models and other potential providers

Type of change

• Bug fix
• New feature
• Breaking change
• Documentation
• Performance improvement

Scope check

• I read the Contributing Guide
• This aligns with RubyLLM's focus on LLM communication
• This isn't application-specific logic that belongs in user code
• This benefits most users, not just my specific use case

Quality check

• I ran overcommit --install and all hooks pass
• I tested my changes thoroughly
• I updated documentation if needed
• I didn't modify auto-generated files manually (models.json, aliases.json)

API changes

• Breaking change
• New public methods/classes
• Changed method signatures
• No API changes

Related issues

No related issues.

[infinityrobot]

@crmne, I'd love your thoughts on this when you get a moment. I also have working Provider implementations for Google Vertex and xAI so would love to add those once I know I've got the patterns right.

[Eric-Guo]

I think as robot, it's much more polite to declare first, you are not human.

[infinityrobot]

Hi @Eric-Guo – I've updated the PR to be in line with main and updated a few spots where logic had changed due to #267 and #271 being merged.

Apologies, I'm not sure what you meant by the robot comment. I am a developer based in Sydney and given it was my first contribution to the repo and relatively new to OS contributions I had made a lot of effort to follow the contribution guide and write up what I had done in a lot of detail.

[Eric-Guo]

Sorry, my mistake, a random change in previous comment is very like claude 4 behavies....😂

[infinityrobot]

Haha, no worries, Claude definitely does that! I had just added that small fix as I was having the same problem raised in #270 and couldn't run the model, alias or docs rake tasks. Looks like the fix is in main now so have updated to use that.