Cohere Introduces Open-Source Toolkit to Empower AI Developers 

April 25, 2024 -- Cohere recently announced the availability of Cohere Toolkit for developers. In this recent blog post, Cohere's Elaine Gao and Beatrix De Wilde describe the toolkit's features and potential impact on AI application development.

Today, we’re making the Cohere Toolkit available for developers to build AI applications faster. This toolkit is an open-source repository of production-ready applications that you can deploy across cloud providers. These applications can access Cohere’s Command, Embed, and Rerank models across AWS, Azure, and the Cohere platform, with more platforms coming soon.

Applications can be deployed in your environment, allowing you to meet your organization’s security standards, and you can connect them to your own data sources. Additionally, the Cohere Toolkit contains individual components which are the building blocks that can be used to build unique applications and deploy at scale.

The initial application included in the toolkit is a knowledge assistant, just like the demo you see on the Cohere platform. A knowledge assistant connected to enterprise data and tailored to specific teams can significantly improve productivity by providing quick access to information, automating tasks, and enabling seamless collaboration within teams. A knowledge assistant built with the Cohere Toolkit is:

• Conversational: Cohere’s models power this application by default and are trained to understand the intent behind conversations, remember conversation history, and complete enterprise use cases using RAG.
• Grounded: Out of the box, the assistant adds fine-grained, relevant citations to responses from your custom data sources.
• Customizable: Developers can use Cohere’s 100+ pre-built connectors to add custom data sources that augment the assistant’s responses, or custom tools so the knowledge assistant can take action.

A core challenge with building AI applications is that developers need to stitch together AI models, prompt templates, retrieval pipelines, and user interfaces with relevant backend code, all within a secure environment. This development process takes months because developers must experiment with how each of these components fit together.

With the Cohere Toolkit, developers now have the source code to accelerate their development and quickly set up an application within just a few days.

Core Components

The toolkit consists of plug-and-play components, and source code for:

• Interfaces: These are UI components with backend integration code. We are starting with open-sourcing our knowledge assistant-based interface, seen on Cohere’s demo, with chat UI supporting multi-turn conversations, fine-grained citations, document upload, and conversation history.
• Models: This module enables developers to interact with Cohere’s proprietary Command R and R+ models hosted on any platform where AI models are available to power your application.
• Retrieval: This set of components can be used for building state-of-the-art retrieval systems that form the backbone of effective RAG pipelines, all within the confines and security of your own environment. They include:
  • A set of 100 free-to-use connectors with OAuth authentication to major enterprise data sources
  • The ability to integrate tools from popular libraries such as LangChain and LlamaIndex
  • The ability to use Cohere’s Embed model hosted on cloud AI services along with vector databases, such as OpenSearch, Pinecone, Weaviate, etc
  • The ability to use Cohere’s Rerank model to improve retrieval of existing search systems using a single line of code

The Cohere Toolkit is ready to deploy with instructions for specific platforms such as AWS, Microsoft Azure, GCP, and more. Developers only have to point the code to where models live to achieve a scalable application ready to deploy in their environment. You can also customize the application, build and contribute new ones to the repository, or just reskin them to reflect your unique branding or voice.

Get Started

To get started with the Cohere Toolkit, follow the setup instructions in the repository.

We’re excited to invite users to contribute new integrations and features by submitting pull requests to GitHub. Your contributions, no matter how big or small, will help shape the future of this project and the wider open-source community.

We’re also putting the toolkit into action at our upcoming Cohere Build Days for developers. Join us for a workshop and build a powerful RAG application at one of our offices, registration here.

Source: Elaine Gao and Beatrix De Wilde, Cohere