eBPF-powered telemetry with near-zero runtime overhead. Use fine-grained CPU, GPU, scheduler, SSD, network, and syscall metrics to surface real issues.
Everything about the application has been optimized to a T. But you still need to be sure that nothing in the environment could ruin the day. Rezolus uses the world's fastest histogram to catch the smallest delay.
Perception and planning loops stick to a strict deadline. Rezolus runs well on embedded systems and monitors the possible sources of contention: scheduler, GPU, block I/O, locks, with recall in hindsight.
On a real-time kernel the chase is often down to the last cycle. Run-queue delay, involuntary context switches, CPU migrations, and TLB shootdowns are all included out of the box.
Complex workloads stress systems in unpredictable ways. Capture both workload characteristics and device runtime behavior in their full dynamic range with histograms, plus drive health and temperature info for debugging.
Local-first design fit inside one static binary. No time-series database, nothing phoning home, no need for third-party dashboards. Recordings land on local disk, and the viewer opens them readily in your browser.
Light enough to provide fine-grain visibility into a box that runs close to line rate. Interface and ethtool counters sit alongside per-CPU scheduler and PMU behavior, so the box can speak to you when something is not right.
Collects system metrics via eBPF with near-zero overhead. Exposes data over HTTP for downstream consumers.
Serves Prometheus-compatible metrics with configurable histogram percentiles. Scrape-ready.
Captures high-fidelity metric snapshots to Parquet files for benchmarking and performance analysis.
Maintains a rolling ring buffer on disk. Trigger snapshots via signal or HTTP for post-incident analysis.
Interactive web dashboard with PromQL support. Explore recordings or stream live from a running agent.
AI-guided analysis tools. Query recordings with PromQL, detect anomalies, and correlate metrics.
curl -fsSL https://install.rezolus.com | sudo bash
This adds the package repo, installs Rezolus, and starts the agent as a systemd service. Supports Debian, Ubuntu, Rocky Linux, and Amazon Linux.
rezolus view http://localhost:4241Streams live metrics from the agent
into
an interactive web dashboard. The agent listens on 0.0.0.0:4241, so
you can also point the viewer at a remote host.
Requires Linux kernel 5.8+ and root. See full installation docs for building from source or managing services individually.
Covering the full Linux performance stack. Plus bring your own metrics.
Usage, frequency, perf counters, cache, TLB
NVIDIA and Apple Silicon metrics
Usage, vmstat, NUMA allocations
Traffic, errors, ethtool counters
Connect latency, jitter, retransmits
Latency distributions, request counts
Runqueue latency, context switches
Counts and latency by type
Self-monitoring: CPU, memory, I/O
CPU, Scheduler, and Syscall metrics are also reported per cgroup (container).
Bring your own metrics by ingesting from any OpenTelemetry endpoint.
Everything you need to deploy and operate Rezolus.
Packages, source builds, platform support
Long-running, one-off, and offline analysis
Modes, samplers, eBPF integration
Samplers and metric types, one by one
Commands, flags, TOML config reference
HTTP endpoints for every mode