Open-source Sponsio released for controlling LLM agents

The open-source project Sponsio has been released to solve the problem of enforcing tool-call boundaries in LLM agents.

Existing prompt engineering approaches showed a success rate of about 95%, but had the limitation that the remaining 5% of exceptional cases could lead to critical errors requiring an audit. In addition, post-hoc audit approaches have the drawback of discovering problems only after side effects have already occurred.

Key features of Sponsio:

• Deterministic contract layer: Before a tool call is executed, it is validated based on YAML rules to determine whether execution should proceed.
• Boundary control: Rules such as "X must be called before Y," "maximum N retries," and "approval required before destructive operations" can be enforced.
• Lightweight: Agent execution flow can be controlled without introducing a heavy workflow engine like Temporal.

The project is available as open source on GitHub.