Mistral AI has announced significant advancements in language model customisation and the introduction of an early version of AI Agents.
Mistral AI's latest offering allows developers to customise any of their flagship and specialist models, including Mistral Large 2 and Codestral, through La Plateforme. This customisation can be achieved using base prompts, few-shot prompting, or fine-tuning, with the option for developers to use their own datasets.
This approach enables developers to integrate generative AI capabilities into their applications with specific domain knowledge, context, or tone.
In addition to model customisation, Mistral AI introduced an alpha release of Agents. This technology wraps models with additional context and instruction, facilitating custom behaviour and workflows through a simple set of instructions and examples. Leveraging the advanced reasoning capabilities of Mistral Large 2, developers can create increasingly complex workflows with multiple agents that are easily shareable within organisations.
The company also announced the stable release of their client SDK, mistralai 1.0, available for both Python and Typescript. This update aims to improve usability and consistency for developers working with Mistral AI's technologies.