Context-Aware: Mesa is context-aware, meaning it can understand the context of the codebase and the code changes in a pull request.
Flexible: Mesa is flexible, meaning it can be configured to review specific in a particular way.
Secure: Mesa is secure, meaning it is secure and compliant with industry best practices.
Scalable: Mesa is scalable, meaning it can be used by any size team or codebase.
Self-Improving: Mesa is self-improving, meaning it can improve over time as it learns from the feedback of the code review process. Furthmore, it
is designed to benefit from the trend of improvements to underlying AI models and tools.
The workflow engine is responsible for generating the code review plan that is used to generate the comments that are displayed in the pull request. This is where
the Mesa assembles your expert agents, rules, and configuration into a coherent code review plan and processes the outputs of AI agents to generate the descriptions and comments that are displayed in the pull request.
The sandbox environment is a tenant-isolated, virtual developer environment in which our AI agents run.
This sandbox environment allows our AI agents to have full access to both your pull request diff and your entire codebase, including the ability to read, write, and execute files.
Unlike other LLM-based code review tools, our AI agents are not limited to the pull request diff, and can build their own understanding of the codebase and the changes in the pull request at review time
as they reason through changes step-by-step.
Once we have a set of instructions and a sandbox environment, the agentic loop is responsible enabling our AI agents to execute tools in the sandbox environment, track their progress, and manage their context window,
with context-window management, and communication between Mesa’s review coordinator agent and any custom expert agents that you have configured.
