Www Project Agent Memory Guard

OWASP Agent Memory Guard is an open source runtime defense against memory poisoning in AI agents. Agents that persist memory across sessions treat anything written there as a privileged input on later turns, so an attacker who plants text in the wrong field can override instructions, exfiltrate data or hijack tool calls. The project notes that such an attack survives a context reset because the memory itself survives. The tool sits between an agent and its memory store and screens every operation through a pipeline of detectors and a declarative policy. This differs from defenses that inspect user input at the front of the loop, because it targets the memory layer instead. It is written in Python and released under the Apache License 2.0. It is aimed at developers and security teams building agents with persistent memory. The project is hosted by the OWASP Foundation as an Incubator project, classified as a tool, with a GitHub repository, two named project leaders and a contributor community. The listing page is a project page rather than a standalone product site, and the code and documentation live on GitHub. There is no charge. The software is free under its open source licence, and the OWASP site states that all of its resources are free and open to everyone. No paid plans are offered.