Why a platform, not a tool
AI agents act, not just answer. They run with a developer's permissions: they read files, run commands, call APIs and push code, and any text they read can carry injected instructions. Guardrails that only filter prompts cannot see what the agent actually does, so FFaX rebuilds endpoint detection and response together with active deception for autonomous agents: observe every action, judge its intent, contain it, and turn a hijack attempt into intelligence.
Endpoint layer: cross-platform runtime daemon
The foundation. A daemon on macOS and Linux observes and controls every tool call, file access, process and connection an agent makes, and protects itself against tampering and injection.
Architecture
One daemon per machine, one view for the team
Agents connect through native hooks and an MCP proxy to the ffaxd daemon on each machine. The daemon enforces policy, keeps the action log and runs the decoy layer, while the gateway watches the network and the decision engine brings everything together in one view for the security team.
Network layer: gateway and deception proxy
Inspects agent traffic between the local network and the internet, blocks covert channels, high-entropy data smuggling and P2P probing, and routes suspicious sessions into decoys.
Intelligence layer: AI semantic review
AST analysis and a local security model judge whether AI-generated code hides malicious logic, obfuscation or jailbreak backdoors, and whether an agent's actions match its task.
Deception layer: active deception and honey-agents
When an agent tries to break out, FFaX can lead it into a realistic, read-only decoy environment and record its plan and attack path instead of only blocking it.
Decision engine: real-time analysis and response
Telemetry from endpoints, the gateway and agent conversations is correlated in real time. The engine recommends a response (block, isolate or deceive) and, where policy allows, executes it.
Enterprise audit and compliance
Every action, verdict and response is logged with full context, giving security and compliance teams an audit trail. Local-first by default, with self-hosted deployment for teams.
In one sentence
Runtime inspection and countermeasure software for AI agents: it analyzes an agent's full behavior chain in real time and responds precisely, so enterprises stay in control of autonomous AI.
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.