AWS offers local, open source leash for agent harnesses

AWS has introduced an open-source software library, Dogwood Local Engine (DLE), designed to add a layer of policy control to AI agents. This library can be embedded into an agent's harness or gateway, and it issues allow/deny verdicts for each tool call made by the agent. DLE doesn't enforce policies itself but checks them against policies defined using Dogwood, an open-source governance language AWS published in August.

A key feature of both DLE and Dogwood is their consideration of temporal conditions. DLE tracks agent tool call events over time, logging them to disk so it can retain its state even if the system crashes or restarts. AWS provides an example of using DLE to control coding agent Git pushes, defining a policy that allows a push only when the most recent test run has passed within the past 15 minutes.

The library is designed to minimize delay in agentic workflows, with evaluation time around 20 microseconds for a 15-minute window and about six milliseconds for a 24-hour window. DLE also prevents concurrent submission collisions by using a lock that allows only one event submission at a time. AWS published the library to address concerns about AI agents potentially going off safety rails, especially as they scale to settings where they work autonomously for longer intervals with more tools.