AWS has released Dogwood, a new open source policy language designed to manage sequences of AI agent tool calls. The announcement was made through the company’s open source channels, and the code is now available under the Apache 2.0 license. Dogwood extends the existing Cedar policy language with temporal conditions, allowing rules to consider an agent’s prior actions rather than evaluating each request in isolation.<\/p>\n
This development addresses a growing need in the field of AI agent governance. As agents are increasingly deployed to perform multi step tasks, such as booking travel or managing workflows, they make multiple tool calls in a session. Traditional policy languages, like Cedar, are designed to evaluate single requests. Dogwood aims to fill this gap by enabling policies that can reason about the entire sequence of calls.<\/p>\n
According to AWS, Dogwood’s capabilities include managing approvals, enforcing rate limits, and tracking running totals. For example, a policy could restrict an agent to a maximum of five API calls per minute or ensure that cumulative spending does not exceed a certain threshold. These functions are intended to give developers finer control over agent behavior and resource usage.<\/p>\n
Cedar is a policy language developed by AWS for access control, used in services like Amazon Verified Permissions. It is designed to be fast, safe, and easy to use. Dogwood builds on Cedar’s foundation by adding a temporal dimension. This allows policies to reference past events in a session, which is not possible with Cedar alone.<\/p>\n
Dogwood is currently supported in AgentCore Policy, an AWS service for managing agent permissions. However, AWS notes that the reference interpreter provided with Dogwood is not production ready. This indicates that while the language is fully specified and functional, further development is required before it can be used in critical production environments.<\/p>\n
The open source release includes the specification, a reference interpreter, and examples. Developers can use these tools to test and experiment with temporal policies. The project is hosted on GitHub, where users can contribute to its development and report issues.<\/p>\n
The release of Dogwood is part of a broader trend in the tech industry toward better governance of AI agents. As agents become more autonomous, there is an increased need for robust controls to prevent misuse and ensure compliance. By making Dogwood open source, AWS is inviting the community to adopt and improve the language, potentially setting a standard for future agent policy management.<\/p>\n
Industry observers note that the ability to reason about sequences of actions is a key requirement for many practical agent applications. For instance, in financial services, an agent might need to check account balances, initiate transfers, and confirm transactions in a single session. Policies that can account for the cumulative effect of these actions are essential for safety and regulatory compliance.<\/p>\n
However, some experts caution that temporal reasoning adds complexity to policy evaluation. Balancing expressiveness with performance will be a challenge for the community. AWS’s decision to not label the interpreter as production ready suggests that they are aware of these challenges and are looking for external input.<\/p>\n
Developers interested in using Dogwood will need to integrate it with their existing agent frameworks. The project provides documentation and examples to help with this process. Since it is open source, the community can also add features and improvements over time.<\/p>\n
The timing of this release is notable, as many companies are still figuring out how to manage AI agents at scale. Standards are only beginning to emerge, and early contributions like Dogwood could influence the direction of the industry. By publishing the language under Apache 2.0, AWS is allowing broad usage and modification, even in commercial products.<\/p>\n
Going forward, AWS plans to continue developing Dogwood based on community feedback. The company has not announced a specific roadmap or timeline for production readiness. Interested parties can track the project’s progress on GitHub and participate in discussions.<\/p>\n
In the short term, developers can start experimenting with Dogwood in their own projects to understand its capabilities and limitations. The examples included in the repository demonstrate common use cases, such as rate limiting and approval workflows. As the community grows, more sophisticated patterns may emerge.<\/p>\n
Ultimately, the success of Dogwood will depend on its adoption and the quality of its implementation. Whether it becomes a standard tool for agent governance will be determined by how well it meets the needs of developers and organizations. The open source model provides a pathway for continuous improvement, and AWS’s involvement adds credibility to the effort.<\/p>\n","protected":false},"excerpt":{"rendered":"
AWS has released Dogwood, a new open source policy language designed to manage sequences of AI agent tool calls. The announcement was made through the company’s open source channels, and the code is now available under the Apache 2.0 license. Dogwood extends the existing Cedar policy language with temporal conditions, allowing rules to consider an […]<\/p>\n","protected":false},"author":1,"featured_media":845,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[127],"tags":[1154,177,1153,1152,234],"class_list":["post-846","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-dev-news","tag-ai-agent-governance","tag-aws","tag-cedar","tag-dogwood","tag-open-source"],"_links":{"self":[{"href":"https:\/\/buildconsole.com\/blog\/wp-json\/wp\/v2\/posts\/846","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/buildconsole.com\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/buildconsole.com\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/buildconsole.com\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/buildconsole.com\/blog\/wp-json\/wp\/v2\/comments?post=846"}],"version-history":[{"count":0,"href":"https:\/\/buildconsole.com\/blog\/wp-json\/wp\/v2\/posts\/846\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/buildconsole.com\/blog\/wp-json\/wp\/v2\/media\/845"}],"wp:attachment":[{"href":"https:\/\/buildconsole.com\/blog\/wp-json\/wp\/v2\/media?parent=846"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/buildconsole.com\/blog\/wp-json\/wp\/v2\/categories?post=846"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/buildconsole.com\/blog\/wp-json\/wp\/v2\/tags?post=846"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}