Every prompt and tool call, checked inline.

Agent harnesses ship with no place to put a control. AI Traffic Control scans every prompt and tool call inline, and enforcement runs locally — PCI, PHI and PII never reach a model or a third party.

ai-tc inline
  • prompt pci/card-number redacted
  • prompt phi/patient-record redacted
  • tool fetch → vendor API blocked
  • tool read src/billing.ts allowed

The controls your team already runs cannot see inside an agent. Your egress proxy sees a request to a model provider and allows it. Your code review sees the diff, days later. In between, a prompt carrying a customer record has already left and a tool call has already run against production, at machine speed, on credentials your developers hold. Add a second harness and there are two of them. Nobody is watching either.

AI-TC puts a control in that path.

AI Traffic Control — Agent harness security

AKA's forward deployed security engineers decide the controls an environment needs, the Flightcrew agents write them, and AI-TC is where those controls meet live traffic. One engine covers every agent in the environment, and a control written once holds as that environment grows.

  1. 01

    Inline enforcement

    AI-TC sits in the path the agents already take.

    Every prompt and every tool call is scanned as it is made, and the decision lands before the model or the third party sees anything. A match warns, redacts, or blocks. There is no window between the call and the check, and no copy of the traffic sent anywhere to be scored.

  2. 02

    The data that cannot leave

    Scanning runs on local machines.

    Card numbers, patient records and personal identifiers are matched in the prompt and in the tool call, before either reaches the model, and the request continues redacted. Nothing is sent to AKA, and nothing is written outside the machine the scan runs on.

  3. 03

    Every agent, one engine

    A second harness does not need a second control.

    AI-TC runs locally, alongside the harnesses your engineers already use. Claude Code and Codex sit behind the same engine, and so does every additional agent running at the same time. One rule set covers all of them, and it lives in the repositories your team already reviews.

AI-TC is open source. Your engineers read every rule it evaluates and every action it takes, before AKA is in the room.

Read the source on GitHub