Guardrails for every agent.

AI Traffic Control (AI-TC) scans every prompt and tool call before it runs. Detection and enforcement run locally, so PCI, PHI, and PII never reach a model or a third party.

View on GitHub See editions
ai-tc data: protected
  • readsrc/api/webhooks.tsallow
  • detectpci/card-numberprotected
  • detectphi/patient-recordprotected
  • egressapi.stripe.comallow
  • egress169.254.169.254block

AI-TC enforces at every layer of the agent harness, from orchestration to data access.

01

Four layers. One attack surface.

Agent Harness Security covers every layer that runs an agent, from the orchestrator down to the data it touches.

Orchestration

The harness that runs your agent.

  • Claude Code
  • Codex
  • Cursor
  • In-house runners

Tool calls

Every action the agent takes.

  • shell
  • read / write
  • network
  • MCP tools

Agent identity

Who the agent is and what it may do.

  • service accounts
  • scoped tokens
  • least privilege
  • rotation

Data access

Every file and API the agent touches.

  • source files
  • .env / secrets
  • external APIs
  • production data

A guardrail changes the picture.

Unmonitored agent: claude-code
  • read src/auth/session.ts executed
  • shell rm -rf /var/log executed
  • network api.stripe.com executed
  • write /etc/hosts executed
No guardrail. 4 executed.
AKA
ai-tc agent: claude-code
  • read src/auth/session.ts allowed
  • shell rm -rf /var/log blocked
  • network api.stripe.com blocked
  • write /etc/hosts blocked
AI-TC on. 3 blocked.

Enforcement runs inline, every time.

Every prompt and tool call clears the rule set before it completes. The rules pull from the same detection library that ships in the open, updated without a redeploy.

Inline, before every call Rules update without a redeploy
02

Capabilities.

  1. 01

    Sensitive-data detection

    PCI, PHI, PII, and secrets are caught before they reach a model or leave the machine.

  2. 02

    Local enforcement

    AI-TC runs on the machine where the agent runs. Regulated data never phones home, and the audit log stays with you.

  3. 03

    Warn, redact, or block

    Every prompt and tool call gets an inline decision. Redaction rewrites the sensitive value; block stops the call.

  4. 04

    Multi-agent

    One engine in front of every agent you run.

03

Community and Enterprise.

AI-TC runs the same on every machine. The community edition is free and installed locally; the forward deployed enterprise edition adds org-wide control, a curated detection feed, and compliance reporting.

AI-TC ships alongside a growing family of open-source security tools.

See the family

Build Your Own Security

Try AI-TC
or