Edictum

Edictum provides runtime governance for AI agents by allowing teams to set and enforce operational rules before agents execute actions. The tool focuses on enabling decision points such as allowing, blocking, requiring approval, or recording actions before any tool or agent operation is carried out. Its core purpose is to ensure that agent behaviors conform to defined policies and workflows, particularly in scenarios where agents interact with real systems and sensitive operations.

Features include runtime decisions on whether an agent action should proceed, rulesets that specify what agents are permitted to read, write, call, or return, and workflow gates that enforce ordered processes across a run. Edictum can require human approval for certain actions and maintains audit evidence for later review. The platform supports defining operational agent profiles, turning documented classifications into enforceable runtime rules, such as read-only, scoped write, or high-impact profiles that combine read/write access with approval requirements. It measures behavioral conformance to declared profiles, focusing on operational controls rather than output quality evaluation.

Edictum offers SDKs and adapters for Python, TypeScript, and Go, making it compatible with a range of agent stacks. Integrations are available for frameworks such as LangChain, OpenAI Agents, CrewAI, Claude Agent SDK, Google ADK, Agno, Semantic Kernel, and Nanobot, enabling enforcement with minimal code changes. The tool provides a CLI for validating rulesets in CI and managing command-line workflows, as well as a Gate CLI for controlling coding assistants and local tool runners. Enforcement can be configured through YAML-based rules, and the system supports live enforcement feeds, workflow gates, local decision logic, and policy failure blocking.

Edictum is distributed under the MIT license and can be installed via pip for Python, pnpm for TypeScript, and go get for Go. Its open-source approach and focus on runtime governance make it suitable for teams building or managing autonomous agent systems that require strict operational control and auditability.