Preloop

Preloop is a control plane for AI agents that provides governance, observability, and security for agentic workflows. It solves the problem of managing AI agents' access to codebases, infrastructure, and business systems by offering policy enforcement, approval workflows, and audit trails. The product targets engineering, platform, security, and operations teams in companies adopting AI agents in production. It is positioned as an open-source, self-hostable alternative to AWS Bedrock, with a focus on keeping humans in control of consequential actions.

Key features

• Onboard existing agents with preloop agents discover
• Govern tool access with MCP Firewall
• Route model traffic with budgets and attribution
• Approve sensitive actions from mobile or Slack
• Keep searchable runtime evidence
• Define allow, deny, and approval policies
• Require human sign-off for sensitive operations
• Route approvals to the right person or team
• Inspect runtime sessions, usage, and spend