AgentGate is an open-source control layer for autonomous AI agents and automations. It acts as an intermediary between agents and the external world — including filesystems, processes, APIs, databases, GUIs, and more — enabling:
- Fine-grained permissions
- Action logging and audit trails
- Policy enforcement
- Alerting and human-in-the-loop controls
- Transparent verification of actions
AgentGate is designed to improve the trust, safety, and observability of agentic systems. It provides a modular TypeScript SDK and a schema for defining and enforcing what agents are allowed to do, and verifying that they actually did it.
As agents are given more power, it's critical to control how they use it. Without oversight, they can access sensitive data, trigger unintended consequences, or fail silently. AgentGate provides a practical framework for safe deployment in real-world workflows.
- A schema format for defining agent permissions, scopes, logging, and escalation.
- A TypeScript SDK for enforcing policies and logging actions.
- Example use cases and integrations (e.g., API access, file writes, external tools).
- Documentation and guides to get started.
AgentGate is under active development. The initial public version is expected in 2025.
MIT