What PoisonZero does
PoisonZero is a small program that runs on the same machine as your AI coding agent and protects the files the agent trusts as its own memory. This page is the short orientation: what it is, what it protects, and who it is for. How it decides is a separate page.
What it is
PoisonZero is a local guard against memory poisoning, the quiet planting of instructions an agent later carries out as if they were its own:
- Local: it runs on the same machine as your agent, so no file has to leave the device to be protected.
- Focused: it guards one thing, the files an agent reads as instructions, and does not touch the rest of the machine.
- Automatic: a poisoned change is caught as it lands, before the agent acts on it.
What it protects
The scope is an allowlist of the files an AI agent treats as its own memory and rules, for example:
- Memory files: the notes and long-term memory an agent keeps between sessions.
- Instruction files: the standing instructions an agent reads on every run, such as an agent's rules or guidance file.
- Configuration files: the settings and tool definitions that steer what an agent may do.
Who it is for
Anyone who runs an AI coding agent that keeps memory or reads instruction files on the machine:
- Individuals running an agent locally on Linux, macOS or Windows.
- Teams and fleets that need the same protection on every device, managed centrally or fully offline.
Read next
How PoisonZero decides on each change: how PoisonZero protects. Cloud or fully private: choosing a mode. Put it on a device: installing PoisonZero.
A local guard for the files your agent trusts.
Watched on the machine, with poisoned changes stopped before your agent acts. Free for Linux, macOS and Windows.
Sign me up