███╗ ██╗██████╗ ███████╗ ████╗ ██║██╔══██╗╚════██║ ██╔██╗ ██║██║ ██║ ██╔╝ ██║╚██╗██║██║ ██║ ██╔╝ ██║ ╚████║██████╔╝ ██║ ╚═╝ ╚═══╝╚═════╝ ╚═╝
A flight recorder for AI coding agents.
Coding agents such as Claude Code and Codex now run shell commands, edit files and reach into servers on their own. When something goes wrong, there is no reliable record of what they actually did. nd7 is that record.
It captures what an agent intended to do, from its own tool calls, and what actually happened on the machine, from the kernel, and writes both into one small, tamper-evident, append-only log that a person can read afterwards.
What we're building, in order
- An audit log that is fast, tiny, and impossible to edit unnoticed. Intent first, kernel-observed effects next: every process, file and network operation, attributed to the agent that caused it.
- Continuity across machines. When the agent SSHes into another host, the session follows it and the log stays one timeline.
- Enforcement. Kernel-level policy, credential proxying and network filtering, generated from what real sessions actually needed rather than written by hand.
- Undo. Revert what an agent did: file contents first, then service and system state, with an honest account of what cannot be reversed.
The engine is open source. Nothing is released yet; we are at step one.