Olo Security
Olo Security provides a macOS application and browser extension that detects and redacts sensitive data on-device before it reaches AI tools. The tool reads text fields through the macOS Accessibility framework, classifies content locally using rule-based detectors and machine learning models, and offers to redact or generalize it, ensuring prompts and detected spans never leave the user's device. It is designed for individual professionals and teams needing centralized policy management and audit reporting.
- Cybersecurity
- Artificial Intelligence
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Sensitive data such as names, patient records, and API keys can be inadvertently exposed when users paste or type content into AI tools. Traditional data loss prevention solutions rely on cloud-based scanning that transmits content to external servers, creating additional privacy risks and compliance burdens.
Solution
Olo Security provides a macOS application and browser extension that reads text fields through the macOS Accessibility framework on-device, classifies potentially sensitive content locally, and offers to redact or generalize it before it reaches an AI tool. The application is architected as two separate processes: one that reads and classifies text with no network access, and a separate sync process that only transmits metadata such as timestamps, policy categories, and redaction counts. This design ensures that prompts, detected spans, and even derivatives like hashes or embeddings never leave the device. Team administrators can set detection and redaction policies and view aggregate metadata, but cannot see individual content.
Target Audience
Primary customers are individual professionals and organizations that use AI tools and need to protect sensitive data, including teams that require centralized policy management and audit reporting.
Features
- On-device classification using rule-based detectors and machine learning models that identify sensitive content without transmitting data
- Two-process architecture where the content-reading process has no network access, verifiable via network monitoring
- Team policy management with admin roles and aggregate redaction metadata reporting (e.g., counts by category, application, and team member)
- Redaction or generalization of detected spans including names, PHI, and credentials before content reaches AI tools
- macOS Accessibility framework integration to read active text fields in any application