budget-enforcement

该论文形式化了分布式多智能体委派中的容错预算守恒机制，提出基于量化资源向量、排他托管信用与委托DAG的防超额花费方案，并通过TLA+模型检查与崩溃注入实验证明了账本与效果守恒等安全属性。

The article describes how a company tackled the problem of tracking LLM costs across teams by first implementing a proxy for visibility and then using Orqai for real-time budget enforcement.

A practical guide to implementing pre-call budget enforcement for LLM API calls, covering estimation, reconciliation, fail-open decisions, scoped budgets, and concurrency handling to prevent runaway costs.