Hey Product Hunt 👋 I'm the maker of agentrec. AI agents are getting shells, credentials, and the autonomy to run unsupervised — but everything guarding them watches the API (prompts, tool-call JSON). Nothing watches what the agent's process actually does on the machine, and the agent's own logs are self-reported — so you're trusting the thing you're auditing. agentrec is a flight recorder for exactly that. It's one eBPF agent — a Kubernetes DaemonSet or a single static binary — that records every file, command, and network connection an agent makes, attributed to the exact tool call that caused it. No SDK, no code changes; it reads the kernel, so it works with any agent framework. 20 detection rules built in (secret access, docker/containerd sockets, cloud-metadata SSRF, reverse shells, priv-esc) → findings on day one On a supported kernel it can block: deny an action in-kernel with a clean -EPERM before it happens ~6.5µs/event, memory-bounded node agent, tamper-evident hash-chained recordings It's live and self-serve — $0.10/node-hour, 100 node-hours free every month, no card to start. Would love your feedback, especially if you run agents in CI or production. Happy to go deep on the eBPF internals — ask me anything!

agentrec : Flight recorder for AI agents

See — and block — what your AI agents actually do

See — and block — what your AI agents actually do