agentcap
github.com/yeet-src/agentcap- Category
- Developer Tools
- Rank
- No. 3688Tools index
Previous survey · No. 3538 ·
- Type
- TOOL
- Use case
- Deployment & Operations
- GitHub
- 16 stars
- Date
About
agentcap records AI coding agent activity from the kernel using eBPF and exposes it as per-agent Prometheus metrics with a Grafana dashboard. It reports tools run, domains queried, destination ports, files read or written, CPU time and network bytes. It needs no SDK, works on agents already running, and attributes child processes to the agent that launched them.
At a glance
agentcap attributes commands, network connections and resource usage to coding agents and their child processes on Linux. It is worth evaluating if you already use Prometheus and want to see which agent spawned a tool or started talking to an unexpected destination.
Good fit
Linux operators investigating agent activity across several coding tools on a host they control.
Consider something else when
Choose another approach if you need to block commands, inspect encrypted traffic or monitor a non-Linux host.
Start here
On a dedicated test host, start with a loopback metrics endpoint. Run one harmless agent task that launches a known command and check whether its execution count appears under the expected agent label.
Cost and requirements
The documented setup requires a Linux kernel with the relevant eBPF hooks and the yeet daemon. Prometheus collects the metrics, and Grafana can display them.
Limits to know
Encrypted DNS can appear as a connection without a domain. The supplied service manifest binds to all interfaces, so restrict access before exposing metrics that reveal command or network activity.
Sources checked 2026-11-05. Source-based assessment.
Tags
Tech Stack
Media
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.