Custos is an open-source safety and control layer designed specifically for AI agents, acting as a reliable 'bouncer' for your applications. As AI agents increasingly interact with critical infrastructure, Custos sits directly between the agent and its tools—such as APIs, files, databases, shell commands, and MCP servers—to intercept and verify every single action before it is executed.

Built for teams deploying AI agents in real-world environments, Custos allows you to explicitly define rule sets determining what an agent is allowed to do, automatically blocking anything outside of those bounds. Furthermore, every action attempted or performed is recorded in a secure, tamper-evident audit trail. With native support for both Python and TypeScript, Custos provides developers with the robust security infrastructure needed to prevent AI agents from performing unintended or harmful actions.

Pros of Custos — The bouncer for your AI agents

Acts as an interception layer between AI agents and critical tools like APIs, files, databases, shell commands, and MCP servers.
Provides a tamper-evident audit trail recording every agent action.
Open-source software built for teams running real-world AI agent environments.
Supports both Python and TypeScript development stacks.

Cons of Custos — The bouncer for your AI agents

Relies entirely on developer-defined rules to catch unexpected agent behaviors.
Limited to Python and TypeScript language support based on current specifications.
Requires integration work to sit between existing agents and their target tools.

Frequently Asked Questions

What is Custos?

Custos is an open-source safety and control layer for AI agents that sits between the agent and its tools to check every action before it runs, blocking anything outside of defined rules.