Execution security kernel for AI agents.
Website · Documentation · Contact
Models request actions. LatchGate decides which ones execute — authenticated, policy-checked, budgeted, sandboxed, and signed. Credentials never enter the agent process. Every decision produces cryptographic evidence.
Assume the model is compromised. Design the system so that it doesn't matter.
| Repository | Description |
|---|---|
| latchgate | Core enforcement kernel — Rust, Apache-2.0 |
| latchgate-integrations | Framework adapters: LangChain, CrewAI, Vercel AI SDK, OpenAI Agents, Pydantic AI |
| latchgate-action | GitHub Action for agent CI pipelines |
| homebrew-tap | brew install latchgate-ai/tap/latchgate |
| latchgate.ai | Website source |
