Six months ago I was building a Python agent with database access. One bad instruction later, live credentials were gone. No warning. No checkpoint. The agent just decided, then acted.
I looked for something that sat between the agent and its tools. It didn't exist. So I built it.
Grimdall is a runtime execution proxy — every tool call your agent makes gets checked against policy before it executes. Shell commands, GitHub actions, database queries, deploys. Blocked or allowed. Logged either way.
Open source. One command. No signup to try it. Would love brutal feedback from anyone shipping agents into production.
Top comments (0)