Last Week at Maxim (Weekly Updates)

Last Week at Maxim (Weekly Updates)

This week at Maxim, we’ve rolled out some powerful updates focused on dynamic scripting, enhanced LLM flexibility, and deeper error visibility across our SDKs and libraries.

Script-Driven Variables in Evaluations

Evaluations just got a whole lot smarter! All variables defined via custom scripts are now surfaced directly in workflow evaluations. This unlocks fully dynamic, script-based variable injection, giving teams greater flexibility and control when designing custom test flows and scenarios.

Mistral LLM Integration is Live

We’ve officially added support for Mistral as an LLM provider! This includes:

  • Backend validation of Mistral API keys with proper error handling
  • A frontend UI to add/remove Mistral keys and choose between available Mistral models
  • Model seeding: Supported Mistral models are now available in the LLM-models sheet
  • Schema and environment-level updates for seamless integration

This brings even more choice to teams who want to evaluate or deploy lightweight, high-performance models in their workflows.

Standardized Error Component (JS/TS & Go SDKs)

We’ve added a unified error component to both our JavaScript/TypeScript SDK and Go library. This ensures consistent error formatting and debugging capabilities across different client implementations, making it easier to handle edge cases and improve observability.

Test-Run Export Summary

Exporting evaluation results now includes a high-level summary of each test run in the spreadsheet output. This makes it easier to track performance across scenarios and communicate insights with your broader team.

That’s it for this week! As always, we’re committed to building the most reliable and flexible evaluation platform for agent workflows. Stay tuned for more updates next week.