SponsioLabs/Sponsio

Sponsio is an open-source runtime safety tool for AI agents. It enforces deterministic rules, called agent contracts, on what an agent may do. AI agents built on large language models (LLMs) behave probabilistically. They can take harmful actions, such as deleting data, running unsafe commands or breaking stated constraints. Sponsio addresses this risk by checking every agent action against defined rules before it is executed. It is developed by Sponsio Labs and is currently in an early (alpha) stage of release.

Each agent contract is a rule that applies over the course of an agent's actions, not only to a single step. Contracts are based on formal methods: a monitor built on fuzzy linear temporal logic checks each tool call. The monitor then allows the action, blocks it, issues a warning, escalates it for review or redirects it to a safe alternative. Every decision is logged in an audit trail. Because checks are rule-based rather than performed by another LLM, they add minimal delay and no model cost at runtime.

Users can start from 16 ready-made contract bundles, which cover common risks such as shell commands and file system access, or combine 46 underlying patterns. Rules can also be drafted in plain English and converted into contracts, which users review before enforcing them. Sponsio can run in observe mode before switching to enforcement. It works with frameworks including LangChain, OpenAI Agents, CrewAI and the Model Context Protocol (MCP), in Python or TypeScript. The code is released under the Apache 2.0 licence.

Auto-discovered on 2026-07-01 by OECD Catalogue Automation