Haldir

Haldir is a governance, security, and control layer for AI agents that interact with external tools and resources. It enforces scoped permissions, spend caps, access to secrets via an encrypted vault, and maintains a tamper-evident audit log for all agent actions via Merkle tree verification. Designed for developers, enterprises, and teams building with AI agents, Haldir integrates with tools and frameworks like Claude Code, LangChain, CrewAI, and the Vercel AI SDK, and can be self-hosted or accessed in the cloud.