How do CoveAI features work, and what underlying technologies do you use?
CoveAI features are built using a combination of modern machine learning technologies, including large language models (LLMs), document processing tools, and structured data extraction services. Cove leverages both secure third-party APIs and our own cloud infrastructure to deliver this functionality. These technologies enable functionality such as content generation, summarization, document analysis, and structured data extraction.
This is not the consumer ChatGPT product; we are leveraging API-based enterprise integrations designed for software applications. Cove controls how the models are prompted, what data is shared for each request, and how results are structured and presented in the product. Our architecture is designed to be flexible, allowing us to expand AI capabilities over time and, if appropriate, change or add model providers without impacting client data ownership or security controls.
How does Cove ensure data security and privacy when using AI, and is our data used to train models?
CoveAI features are designed with data minimization and privacy in mind to protect client data. These features only process the information intentionally provided for a specific workflow (e.g., a user’s prompt for generating an Announcement). AI subprocessors do not have broad or autonomous access to tenant or portfolio data and they do not use Cove’s client data to train their models. Client data remains stored within Cove, and when external providers are used, only the minimum necessary data is transmitted to fulfill the request. These providers are also tracked as subprocessors within our SOC compliance program.
How does Cove ensure AI-generated content is professional and appropriate?
Cove embeds structured guidance and guardrails into its AI workflows to promote clear, professional, and audience-appropriate outputs. Default instructions guide tone (e.g., tenant-facing and professional), discourage speculation, and prevent the model from inventing facts or policies. Users can further adjust tone within each feature, and we are expanding controls to support property- and portfolio-level preferences.