Heads up This site is currently under heavy development.
← all tools
◆ AI/LLM Security

ADR

sensor-v1.0.0 open-source

ADR secures enterprise AI agents through observability, security benchmarking, and threat detection. Deployed at Uber.

Summary

ADR is a open-source cyber tool that monitors and evaluates the security of AI agents such as Cursor, Claude Code, and Codex, as well as customer-facing support agents. It runs as a sensor deployed alongside these agents, capturing intent, tool use, and execution traces, paired with a detector that combines high-recall triage with deeper reasoning on suspicious sessions; a companion benchmark, ADR-Bench, tests agent defenses against over 300 tasks and 133 MCP servers covering known attack techniques. It targets security teams responsible for AI agent deployments inside organizations, sitting in the emerging agentic AI security space alongside broader observability and threat-detection tooling. Discovery and prevention components remain closed-source. The project shipped its first tracked release 21 days ago.

ADR secures enterprise AI agents through observability, security benchmarking, and threat detection. Deployed at Uber.

What ADR answers

Which AI agents can this actually watch?

captures intent, tool use, and execution traces across 7+ coding tools on macOS, Linux, and Windows, plus internal automation and customer-facing support agents

Does it stop unsafe actions or just flag them?

the open-source release only observes and detects; the prevention component that would block actions is closed-source

Can it tell me what AI tools are running on employee machines before I even deploy it?

no, discovery of unsanctioned AI tools and endpoints is closed-source and not part of this release

What does it take to run the detector?

a sensor deployed alongside each agent feeds a two-tier detector, high-recall triage first, deeper reasoning only on sessions flagged as suspicious

How do I know its detection actually works against real attacks?

run it against ADR-Bench, which covers 300+ tasks and 133 MCP servers spanning all 17 known agent attack techniques

Is this proven outside a lab?

it runs in production at Uber and the underlying approach was accepted as a paper at MLSys 2026

Release history

  1. sensor-v1.0.0 Jul 31, 2026 · issue 005

    ADR Sensor v1.0.0: first public release of the Python telemetry package for Agentic Detection and Response

    └──▷ GET THIS VERSION
    $ git clone --branch sensor-v1.0.0 https://github.com/uber/ADR.git
    # already have the repo? check out this version:
    $ git checkout sensor-v1.0.0
    └──▷ TRY IT
    Install the ADR Sensor package to begin collecting normalized agent telemetry from AI coding tools on an endpoint.
    $ pip install adr-sensor
    • Introduces adr-sensor (installable via pip install adr-sensor), a Python package that collects and normalizes agent telemetry from Claude Code, Cursor, Codex, opencode, and Claude Desktop into a unified schema.
    • Supports agent observability across macOS, Linux, and Windows, capturing agent intent, tool use, and execution traces from 7+ AI coding tools.
my-toolchain — 0 tools
paste an install list to detect your tools

A brew list, a Brewfile, requirements.txt, a Dockerfile — or just the product names, free-form. Nothing leaves your browser.

    browse all tools →