AWS Adds Local Instance of Dogwood Tool for Governing AI Agents as Code
Techstrong.ai, Monday, October 5th, 2026
AWS's Dogwood Local Engine embeds policy-as-code in local agent harnesses to allow or deny tool calls based on past actions.
AWS has released Dogwood Local Engine, an extension of its open source Dogwood policy-as-code framework that can be embedded as a library in an agent harness on a local machine.
It enforces policies with temporal conditions based on a history of timestamped tool-call events, such as allowing a coding agent to git push only if tests passed within the last few minutes, and the harness runs a tool only when the engine allows it.
AWS VP Marc Brooker says the goal is easier policy enforcement with lower latency.