Using Models Hosted on Together AI

You can use models, such as DeepSeek-R1, hosted on Together AI in your pipelines through DeepsetTogetherAIGenerator. First, connect deepset AI Platform to Together AI by passing the Together AI API key on the Connections page. Then, add the DeepsetTogetherAIGenerator to your pipeline and specify the model to use.

For details, see Use Together AI Models.

Tracing Pipelines in Weights & Biases' Weave

We've added an integration with Weights & Biases' Weave service so that you can easily trace your pipelines in W&B. All it takes is adding the WeaveConnector component to the pipeline you want to trace. The component doesn't require any configuration. You just pass the name of the pipeline you want to trace. This name is then used as the name of the Weave tracing project to help you easily identify which pipeline the traces belong to.

To learn more about how this integration works and how to use it, see Use Weights & Biases Services.

MongoDB Atlas Document Store

deepset AI Platform now supports MongoDB Atlas as a document store. This means you can store your data in your MongoDB Atlas database and retrieve it at search time using deepset pipelines. Just add the MongoDBAtlasEmbeddingRetriever to your query pipeline. For detailed instructions, see Connect to an External Database.

Deprecating V1 Pipelines

On January 15, we're deprecating v1 pipelines. This means the v1 pipelines stop running and are no longer supported. We encourage you to use our robust v2 pipelines which offer much greater flexibility, including loops and multiple branches. For more information, see Pipelines.

Haystack Experimental Features Available Through Custom Components

Haystack, deepset's open-source LLM framework which is the underlying technology for deepset AI Platform offers experimental features available in the [haystack-experimental](https://github.com/deepset-ai/haystack-experimental) package. While these features are not available in Pipeline Builder by default, you can add them through custom components. For details, see Custom Components and deepset AI Platform and Haystack. (to do: add links once these are published)

Environment Selection

When using API or the SDK, you can now choose the base URL (or the environment) for your deepset AI Platform deployment.

If you're making calls using our interactive API documentation, simply switch the Base URL option to the desired value:

You can choose between Europe and US deployments. For details, see API Structure and Organization.

When using the SDK, you can pass the environment in the api_url parameter or choose us, eu, or custom when setting up the SDK with the deepset-cloud login command. For details, see Set Up the SDK CLI.

Multiple Prompts on the Job Details Page

Based on your feedback, we've improved the Results table on the Job Details page. Now, when you click a job's name to view its details, you can see all the prompts used in the pipeline. Previously, you could only view a single prompt—now, you get a complete overview of the prompts used for each query, along with other key result details.