AgentSpace

AgentSpace - AI-Powered Software Architecture Reviews As software systems grow, it becomes increasingly difficult for engineering teams to understand how their services depend on one another and what might break when something changes. Architecture diagrams often fall behind the actual system, while code-focused AI assistants don't always have the broader context needed to evaluate architectural decisions. We built AgentSpace to address that gap. AgentSpace is a collaborative workspace where engineers can visualize their software architecture, import Docker Compose configurations, explore dependencies, and review proposed changes together. Its architecture graph is versioned, so teams can track what changed and why. For the Meta Global AI Developer Hackathon, we integrated Meta Muse Spark 1.3 into our existing agent framework. Instead of asking a model to review isolated code snippets, we provide structured information about the system's components, connections, and configurations. The model can then identify potential security and reliability concerns and suggest changes grounded in that context. We added structured-output validation and checks on proposed graph modifications. AI-generated proposals require human approval before they can update the architecture specification, and accepted changes are recorded in the project's revision history. We tested this workflow with a sample e-commerce system, ShopSphere, where Muse Spark identified overly broad database access in the payment service and reliability concerns around Kafka event handling. Our aim is to make architecture reviews more accessible, collaborative, and actionable, particularly for teams that don't have a dedicated architecture or infrastructure specialist reviewing every change. Live application: https://agentspace-meta.vercel.app