Fulcrum

You'd reach for this when you need to put guardrails around AI systems in production. It handles policy enforcement to prevent unwanted model behavior, tracks and caps spending across API calls, and gives you visibility into what your AI is actually doing. The specifics of what APIs it exposes aren't detailed in the source, but the governance angle suggests rule definition, budget thresholds, and logging hooks. If you're running Claude or other LLMs at scale and need to control costs or ensure compliance, this sits between your application and the model to enforce those constraints.