Vibe
Guard
The free, open-source firewall between your AI agents and everything they touch. Scan LLM traffic, block prompt injection, and guard your MCP toolchain — 764 detection rules, zero paid APIs. One command, the whole wall.
Runs across the full AI stack — Node runtime, MCP over JSON‑RPC stdio, evals, benchmarks — and speaks to every agent client: Claude Code, Codex, Hermes, OpenCode, and any MCP consumer.
Full-Stack Shield
Prompt-Injection Defense
764 hand-tuned rules catch jailbreaks, indirect injection, and tool-abuse attempts before they reach your model or agents.
Agent Firewall
A transport-level wall that watches every tool call your agents make — allow, deny, or sanitize per rule, per tool.
MCP Gateway
82 MCP tools over JSON‑RPC stdio — sits between any client and any server. Wire it once, every skill and agent is protected.
Integrity Hashing
Hashed runtime state detects tampering — including other agents dirtying your files mid-run. If it changed, you know.
Evals & Benchmarks
One command re-runs the attack corpus (corpus truth in manifest.json) and regenerates the README benchmark block.
Node Runtime
Zero-dependency Node core, free forever. No SaaS, no telemetry, no keys — the wall runs on your machine, not theirs.
One Command
Run the Wall
Every agent you spawn. Every tool it reaches. Every prompt that tries to break out. Guarded.