lf top

Real-time runtime telemetry for the LenserFight environment. Renders CPU, memory, GPU, service health, local battle load, and rolling graphs inside an alt-screen terminal UI that refreshes every second.

lf top                          # compact mode (default)
lf top --mode expanded          # all panels + per-core CPU + graphs
lf top monitor                  # subcommand alias for expanded mode
lf top battle                   # battle operations center
lf top graph                    # rolling 60-second sparkline graphs
lf top infra                    # infrastructure services + host info
lf top stream                   # pipe-friendly scrolling output (no alt screen)

---

Modes

Compact (default)

Single-screen summary: CPU bar, memory bar, GPU (when detected), service status dots, battle counts, and active alerts. Fits in ~76 columns.

Expanded (lf top --mode expanded / lf top monitor)

All compact panels plus:

• Per-core CPU breakdown (C00–Cnn in groups)
• Rolling 60-second CPU and memory sparkline graphs
• Full process info (PID, heap MB, uptime)

Adapts to up to 100 columns.

Battle (lf top battle)

Operational focus for active local battles: running/draft/executed counts, resource pressure summary (CPU/MEM/VRAM inline), and Ollama + Supabase service status. Useful during local battle run sessions.

Infrastructure (lf top infra)

Service connectivity panel — Ollama, Supabase, Cloud API, Docker — with probe detail (version strings, HTTP status). Shows host metadata: hostname, OS, architecture, Node version.

Graph (lf top graph)

Full-width Unicode sparkline graphs for CPU and memory drawn over a 60-second rolling window. The longer the session, the more context is visible.

Stream (lf top stream)

Emits one timestamped line per tick to stdout without clearing the screen:

2026-05-09T14:32:01.000Z  cpu=43%  mem=38%  gpu=61%  battles=2

No alt screen, no ANSI color by default in this mode. Designed to be piped:

lf top stream --interval 5000 | tee telemetry.log
lf top stream | grep "alerts="

---

Keyboard controls

Controls are active in all modes except stream.

Key	Action
e	Toggle expanded / compact
g	Toggle graph mode
b	Toggle battle mode
i	Toggle infra mode
r	Force immediate refresh
q / Ctrl-C	Quit and restore terminal

Switching modes is instantaneous — the active mode badge in the title bar confirms the current view.

---

Options

Flag	Default	Description
--mode	compact	Initial mode: compact, expanded, battle, infrastructure, graph, stream
--interval	1000	Refresh interval in milliseconds. Minimum 250.

---

Metrics reference

Metric	Source	Notes
CPU total	os.cpus() delta between ticks	Average across all cores. First tick shows 0% (no prior sample).
CPU per-core	Same delta	Visible in expanded and graph modes.
Memory	os.freemem() / os.totalmem()	Resident OS memory. Does not include kernel page cache.
Load average	os.loadavg()	1m / 5m / 15m. Shown as trailing three numbers next to the CPU bar.
GPU utilization	nvidia-smi (async)	Probed every 5 ticks. Silently absent when nvidia-smi is unavailable.
VRAM	nvidia-smi	Used / total in GB. Shown only when GPU is detected.
GPU temperature	nvidia-smi	Degrees Celsius, inline with VRAM row.
Ollama	GET /api/version (2s timeout)	Reports server version string when up.
Supabase	GET /auth/v1/health (2s timeout)	Reads supabaseUrl from project config.
Cloud API	GET /health (2s timeout)	Reads cloudApiUrl from project config.
Docker	docker version subprocess (2s timeout)	Reports client version string when up.
Battles	user runtime local-battles/*.json plus legacy project-root files	Counts by status: running (ready), draft, executed/voted.
Heap	process.memoryUsage().heapUsed	V8 heap of the lf process itself.
Uptime	process.uptime()	Time since the lf top session started.

Service probes run every 5 ticks and GPU detection runs every 5 ticks, both non-blocking. The displayed values between probes are the cached result from the last completed check.

---

Alerts

The dashboard generates inline alerts when thresholds are crossed. Critical alerts show in red; warnings in yellow. Each alert includes numbered recommended actions.

Condition	Level	Example recommendation
Memory ≥ 80%	Warning	Monitor Ollama VRAM; consider pausing idle agents
Memory ≥ 90%	Critical	Reduce active battles; enable low-memory mode
CPU ≥ 90%	Critical	Reduce concurrent executions; switch to a lighter model
GPU VRAM ≥ 90%	Critical	Reduce battles; switch inference provider; pause background agents
Any service down	Warning	Service-specific fix command (e.g. ollama serve)

Alerts are recalculated on every tick. They auto-clear as soon as conditions normalise.

---

Non-TTY behaviour

When stdout is not a TTY (e.g. inside a CI script or shell substitution), lf top emits a single compact JSON snapshot and exits:

lf top 2>/dev/null
# {"ts":"2026-05-09T14:32:01.000Z","cpu":43,"mem":38,"gpu":null,"alerts":0}

---

Related

• lf doctor — point-in-time environment health checks
• lf status — auth, config, and journey progress
• Operate LenserFight from the TUI dashboard — main lf command dashboard
• lf battle run — local battle execution (lf top battle shows load during runs)
• lf execution list — programmatic run state for scripting

lf top

Real-time runtime telemetry — CPU, memory, GPU, services, battles. Keys: [e]xpand [g]raph [b]attle [i]nfra [r]efresh [q]uit

lf top monitor

Expanded runtime telemetry — all panels, per-core CPUs, graphs.

Flag	Type	Required	Description
--interval	string	no	Refresh interval in ms

lf top battle

Battle operations center — battle load, agents, resource pressure.

Flag	Type	Required	Description
--interval	string	no	Refresh interval in ms

lf top graph

Rolling sparkline graphs for CPU and memory (60s window).

Flag	Type	Required	Description
--interval	string	no	Refresh interval in ms

lf top infra

Infrastructure view — service connectivity, host info.

Flag	Type	Required	Description
--interval	string	no	Refresh interval in ms

lf top stream

Scrolling telemetry stream — pipe-friendly, no alt screen.

Flag	Type	Required	Description
--interval	string	no	Refresh interval in ms