At Cohere, we’re continuing to enable developers to accelerate generative AI application development. We’re expanding our open source Cohere Toolkit features, introducing HTML rendering, configurable authentication, and multi-step tool use.

The toolkit is an open-source repository of production-ready applications that you can deploy across cloud providers. It contains the components required to build an end-to-end AI-powered assistant including source code for interfaces, models, and retrieval. Users can leverage the toolkit to build applications that are conversational, grounded, and customizable for powerful business solutions.

Generate Interactive HTML Applications in the Chat UI

Users can now ask our Command R models to generate interactive HTML applications. Using the Command R model series, these applications can be rendered in the Chat UI for rapid prototyping or for visualizing data.

Watch Command R+ generate a functional to-do list in the Chat UI:

0:00

/0:39

Try this example yourself with the following prompt:

Write a functioning to-do app in HTML. The app should have an input textbox so I can write my tasks, a button to add the task to my list, and when the button is clicked, the task should be added to a running list with a checkbox. Use Arial font and make the background of the app #ebe9eb.

Developing a website or web application often involves creating a significant amount of HTML code. By having an AI model generate HTML, developers can save time and effort, allowing them to focus on more complex and creative aspects of the project. This interface can assist end users with:

Rapid prototyping: AI models can quickly generate HTML code for basic web components, such as forms, tables, and layouts
Code generation from natural language: Developers can provide natural language descriptions or instructions to the AI model, which can then generate the corresponding HTML code

Here is an example using Command R+ to prototype a simple game:

0:00

/0:43

Try this example yourself with the following prompt:

Write a guess the number game in HTML. The computer picks a random number, and the player has to guess it by entering a number. The game can provide feedback on whether the guess is too high, too low, or correct. This can be created using HTML forms for user input.

Configure Authentication Using Email and Password

Developers can now set up access permissions to a deployed toolkit for end users with simple email and password authentication. You can also configure authentication using Google OAuth, which enables you to use a Google account for SSO or OpenIDConnect for another identity provider.

Authentication enables organizations to deploy the toolkit with data sources that require individual user permissions to access. This is a necessity when creating an enterprise assistant for internal documents.

When configured, users will see a login page like the following:

0:00

/0:10

Build an Assistant That Can Use Multiple Tools to Answer User Queries

Users can now build an assistant that uses custom tools with multi-step tool use in the toolkit. This feature was previously only available on our platform via API, but now users can create RAG assistants that can do multi-step with their own custom tools using the Cohere UI using the toolkit.

Multi-step tool use is a Command R/R+ model capability to invoke external tools to generate an informed response. The model is also able to use multiple tools in sequence. When the model is given a list of tool definitions, it generates a plan of action and decides which tools to use, populates the required parameters, and defines the order of operations. Adding this functionality to the toolkit enables you to build a RAG application with custom tools and use it with your preferred privately hosted model.

Enterprise use cases of multi-step tool use include:

Data analysis
Market research
Customer support chatbots
CRM updates

Try the Toolkit Today

Get up to speed and get started with the Cohere Toolkit GitHub repo. Need help? Reach out to us on Discord.

Tomeu Cabot