Extended MistralAI Conversation - A New Integration for Home Assistant

What is Extended MistralAI Conversation (EMC)?

Extended MistralAI Conversation (EMC) is a new integration forHome Assistantdeveloped to harness the power of Mistral AI's large language models (LLMs). Unlike traditional intent-based integrations, EMC allows the LLM to directly manage user requests, offering a more flexible and powerful approach, similar to what we have already seen with ExtendedopenaiConversation (EOC). The goal is to provide a complete package, also integrating Speech-to-Text (STT) and Text-to-Speech (TTS) features derived from other existing integrations, such as Mistral AI Conversation.

EMC's approach deviates from the default intent list, a method that some users, especially former Rhasspy users, have found cumbersome. The integration is based on the idea that the LLM can interpret and act on requests more directly, improving interaction with the smart home. The author of the integration chose to move to a more "local" AI provider such as Mistral AI, also following impressions on the maintenance of EOC.

Key Differences from Extended OpenAI Conversation

For those familiar with Extended OpenAI Conversation (EOC), EMC has some key differences that improve its flexibility and integration with the Home Assistant ecosystem.

Prompts and Custom Functions

One of the main innovations of EMC is the management of the prompt, which is divided into two parts: a static prompt and a dynamic prompt, both configurable via YAML files. Custom functions and scripts are also defined in a separate YAML file. This allows for greater customization and more granular control over how the LLM interprets requests and interacts with the system.

Integration with Home Assistant Scripts

EMC introduces an advanced mechanism to help the LLM select the appropriate scripts present in the filescripts.yamlof Home Assistant. Instead of duplicating script logic within EMC, the integration sends the script description to the LLM via dynamic prompting. This means that if the scripts are well documented with a clear "description" field, the LLM can use them directly, especially for simple scripts without parameters. Additionally, custom "script" functions are executed directly by Home Assistant's script engine, allowing you to use any available action or instruction.

Requirements and Configuration

To use Extended MistralAI Conversation, you must have a Mistral AI account. Initially, a free account may be sufficient, using templates likemistral-small-latest. However, if there is a high load on Mistral AI's infrastructure, free accounts may not guarantee a timely response. To overcome this problem, it is advisable to switch to a "Pay-as-you-go" account, which only charges for intensive use, allowing you to overcome any resource-related blocks.

The EMC configuration is saved in a JSON file. The integration is distributed with example files directly taken from the author's configuration, making installation and customization easier. The user will need to replace the demo values ​​with their own specific credentials and configurations.

Mistral AI Account Considerations

As mentioned, the choice between a free and paid account for Mistral AI depends on the intensity of use. Although the modelmistral-small-latestis generally able to handle requests, traffic spikes may limit access for free users. Adopting a "Pay-as-you-go" account with a small initial top-up (for example, a few euros) guarantees greater reliability and prevents blocks related to resource availability, ensuring a smoother experience with Extended MistralAI Conversation.