Two kinds of review
Structural analysis parses AI-generated code into syntax trees and checks it with Semgrep-style rules for dangerous calls, hidden control flow and obfuscation. A security-tuned model running inside your network then reads the code's meaning, and can tell when code that seems to add numbers also hides a time-triggered data exfiltration path.
Commit gate
Every AI-assisted change, and every commit before it reaches your Git server, passes semantic review. Only code that clears the review is accepted.
Intent alignment
What it was asked, versus what it is doing
FFaX compares what the agent was asked to do with what it is actually doing, and flags actions that drift from the task, a common sign of prompt injection.
Dynamic taint tracking
Secrets such as API keys in environment variables or sensitive files are marked as tainted at runtime. If tainted data flows toward an unauthorized port, process or address, the process is stopped before anything leaves.
Review pipeline
Rules, the local model and human approval form one pipeline, so high-risk findings reach a person with the evidence attached.
Private by default
The security model runs inside your own network, so code and prompts are reviewed without ever leaving the company.
Design partner program
Install the demo for free and shape the product
Engineering teams using AI coding agents can run the six-week demo at no cost: open-source client, an action log that never leaves your machine, three default policies, decoys and a one-page console.