尚未收集仓库信息。
尚未检测到更新。GitHub 发布版本后会显示在这里。
Open-source macOS tool that independently records what AI agents run: processes, file changes and network activity. Also a macOS EDR.
翻译准备中
正在加载完整 README。
查看完整 README ↗RawTrail is an open-source endpoint detection and response (EDR) system for macOS fleets. It gives security teams real-time visibility into process, network, and DNS activity on Apple Silicon Macs, runs behavioral detection rules against a materialized process graph, and can block execution or kill a process on the endpoint. It is fully self-hosted: your own server, your own data, no SaaS dependency.
Deploy it and enroll real Macs, watch your own AI agent on your laptop, or try it first with the Mac-free demo.
Two steps. Stand up a server, then push the agent to your Macs.
1. Stand up the server.
Single VM with your own domain (recommended). One Linux VM, one script: Caddy gets a Let's Encrypt certificate automatically and reverse-proxies to the server, so you manage no certificates and no content-inspecting edge WAF flags your agents' telemetry as attacks. Full walkthrough: docs/quickstart-vm.md.
EDR_DOMAIN=edr.example.com EDR_VERSION=latest ./bootstrap.sh
For production, pin a specific signed release instead of latest: see docs/quickstart-vm.md.
Or self-host the container on any container host (Docker, Kubernetes, AWS ECS/EKS, GCP, Azure, or your own VM). The server is a standard multi-arch Linux image. Setup, secrets, and TLS: docs/install-server.md.
2. Deploy the agent to your Macs (Apple Silicon, macOS 26+). The agent ships as a Developer ID-signed, notarized .pkg plus three .mobileconfig profiles, delivered by your MDM.
Once agents enroll, open the server's /ui/ to watch hosts, process trees, and alerts in real time.
Run the server and the agent on your own Mac to see what your coding agent actually ran: every process, file change and connection behind each prompt, and any credential file it opened. Nothing leaves the Mac. One command with Docker running:
scripts/laptop/rawtrail-laptop install
What you see, what it costs and how to remove it: docs/watch-your-agent.md.
Want to look before deploying? Evaluate the full server, UI, and detection pipeline in about five minutes. No Apple Silicon Mac, MDM, or Apple entitlement needed.
docker compose -f docker-compose.demo.yml up
Open https://localhost:8088/ui/, accept the self-signed certificate warning, and sign in through the bundled IdP with the demo account [email protected] / demo. It signs in as an admin: user invitations, policy authoring, and every host and alert action, so you can drive the full operator surface.
You'll see two real macOS hosts (an engineer laptop and a CI build server), each with a deep process graph and correlated network and DNS activity drawn from genuine scrubbed captures. Woven into that ambient activity are five fired ATT&CK detections: a credential keychain dump and a suspicious process phoning home (exec, DNS, and outbound connection correlated across all three streams), plus sudoers tampering, launchd persistence, and an application-control block. Every alert comes from the real ingestion and detection pipeline, not hand-inserted rows, and the benign activity raises no false alarms.
Notes:
以上为 README 的开头部分。其余内容请阅读原文。 查看完整 README ↗