CrowdStrike for AI agents — endpoint visibility and control over AI use
HQ: San Francisco, CA, United States | Founded: 2025 | Employees: 2 | Stage: Seed (Y Combinator S26) | Website: https://www.traceforce.ai
Traceforce builds security for how employees use AI inside companies, and positions itself as 'CrowdStrike for AI agents'. The problem is that tools such as ChatGPT, Claude, Cursor and Claude Code are now installed on employee laptops and wired into company data through connectors and skills nobody keeps track of, while the tools security teams already own — network gateways, proxies, enterprise controls — never see the risky activity, because it happens on the device before it reaches the network. Traceforce installs on the device itself, giving security teams a full inventory of every AI app, connector and skill across their machines within about 30 minutes, visibility into how each agent connects to data sources, and real-time blocking of unsafe actions such as uploading a client list to a chatbot. It also publishes mcp-xray, an open-source tool that detects vulnerable connectors. The origin story is a real incident: at their previous company, an AI coding agent on a developer's laptop scraped what appeared to be database credentials and pushed them to a public GitHub repository, which took weeks to investigate. The team validated the problem with more than 50 security executives before writing code. Reported results: more than 3,000 employee devices secured across six medium-sized enterprises, more than 40 active pilots including several Fortune 500 companies, and $3M+ in yearly subscription revenue represented by those pilots. Founded in 2025 in San Francisco with 2 employees. Chief executive Xia Hua was previously director of engineering at Clumio and holds a PhD in applied mathematics from MIT; chief technology officer Varun Wadhwa led database systems for AI retrieval at LinkedIn and distributed data infrastructure at Clumio.
Traceforce has the best traction-to-team-size ratio in this set: two people, more than 3,000 devices in use across six enterprise customers, and over 40 pilots representing $3M+ in yearly revenue. It also has the sharpest technical argument. Security tools that sit on the network cannot see what an AI coding agent does on a laptop before it touches the network, and cannot see how an agent is wired into company data through connectors and skills nobody tracks. Moving the software onto the device is the only way to find shadow AI and block an action as it happens. The origin story is lived rather than imagined: at their previous company, an AI coding agent scraped what looked like database credentials and pushed them to a public GitHub repository. Two risks stand out — endpoint security is the most crowded part of security, and two founders cannot cover enterprise deployment for long.
Browse: Home | Blog | About | All Startups A-Z | View full profile on StartupWiki