Case Study: Fireworks AI accelerates agentic runtime development with Mastra

Why Fireworks uses Mastra in their agentic runtime

Fireworks AI, an inference platform processing over 140 billion tokens daily, faced a challenge in building a hosted agentic system. Their staff needed a TypeScript-based state machine foundation that offered control and programmability, but found existing solutions like LangChain were either too verbose or did not fit their ecosystem. This led them to begin building a custom solution from scratch.

By adopting the Mastra framework as their foundation, the Fireworks team was able to quickly build and launch AIML, their open-source, prompt-based agent framework. Using Mastra's workflows to execute logic from XML-tagged prompts allowed them to create complex agentic systems without writing code. This solution saved significant development time and is now used by some of Fireworks' largest customers, enabling the team to shift its focus from internal infrastructure to improving the user experience.

Fireworks AI

Matt Apperson

Staff Software Engineer