lms-mon

TUI monitor for LM Studio

Real-time visibility into loaded models, inference traces, GPU/CPU/RAM usage, and rolling sparkline graphs — all in a single terminal window.

---

Features

• 4-pane layout — models table, system meters, live log, sparkline graphs
• Tab / Shift-Tab to cycle panes; active pane gets an accent border
• ↑ / ↓ to select a model row → highlights that model’s traces in the log pane
• Live inference stats: tok/s (last / avg / peak), TTFT, request count
• GPU metrics via nvidia-smi (NVIDIA); gracefully skipped if absent
• lms log stream --json --stats parsed in real time — exact prompt/response traces
• Command palette (Ctrl+P) — Textual’s fuzzy-search launcher for every shortcut
• Shows helpful error if lms is not on PATH

---

Requirements

Dependency	Version	Purpose
Python	≥ 3.10	Runtime
textual	≥ 0.70	TUI framework
httpx	≥ 0.27	Async HTTP to LM Studio REST API
psutil	≥ 5.9	CPU / RAM metrics
LM Studio	≥ 0.3.26	lms CLI + REST API
nvidia-smi	optional	GPU metrics (NVIDIA only)

---

Install

Option A — uv (recommended)

uv provides the fastest isolated Python 3.12 environment.

# 1. install uv if not already present
curl -LsSf https://astral.sh/uv/install.sh | sh

# 2. clone
git clone https://github.com/dcolley/lms-mon.git
cd lms-mon

# 3. create venv with Python 3.12 and install deps
uv venv --python=3.12
source .venv/bin/activate          # Windows: .venv\Scripts\activate
uv pip install -r requirements.txt

# 4. run
python lms_mon.py

Or install as an editable CLI entry-point:

uv pip install -e .
lms-mon

Option B — pip + venv

git clone https://github.com/dcolley/lms-mon.git
cd lms-mon
python3.12 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python lms_mon.py

Option C — pipx (isolated global command)

pipx install git+https://github.com/dcolley/lms-mon.git
lms-mon

---

LM Studio setup

# Ensure lms CLI is on PATH (one-time bootstrap)
~/.lmstudio/bin/lms bootstrap

# Start the server if not running
lms server start

# Verify
lms ps
lms server status

lms-mon uses:

• GET http://localhost:1234/api/v0/models — models pane
• lms log stream --source model --filter input,output --json --stats — log pane

---

Usage

lms-mon                            # default localhost:1234
lms-mon --host 192.168.1.100       # remote LM Studio instance
lms-mon --port 8080                # custom port

Command palette (Ctrl+P)

Ctrl+P is a Textual feature, not something lms-mon implements itself. Textual enables it by default (ENABLE_COMMAND_PALETTE); lms-mon inherits it automatically.

Press Ctrl+P to open a fuzzy-search command palette. Type to filter, Enter to run a command, Esc to close. It lists every registered key binding, including shortcuts that are hidden from the footer — useful for discovering actions you might not remember:

Palette label	Key	Action
Select	Space	Toggle log filter on the model under cursor
Sort	s	Toggle alphabetical / loaded-first sort
Load	l	Open load dialog (lms load + extra flags)
Unload	u	Unload model under cursor
Log: model	1	Stream model prediction logs
Log: server	2	Stream LM Studio server logs
Log: runtime	3	Stream runtime logs
Log: input	i	Toggle input events (model source)
Log: output	o	Toggle output events (model source)
Log: stats	S	Toggle per-token stats in log stream
Theme	—	Change Textual colour theme
Quit	—	Exit the app

The palette also exposes Textual system commands (e.g. Theme, Quit). To disable it in a fork, set ENABLE_COMMAND_PALETTE = False on the App subclass.

Keyboard shortcuts

Key	Action
Ctrl+P	Open command palette (Textual; see above)
Tab	Focus next pane
Shift+Tab	Focus previous pane
↑ / ↓	Move model cursor (log pane highlights selected model)
Space	Toggle log filter on model under cursor
s	Toggle model list sort (alpha / loaded-first)
l	Load model (opens flag dialog)
u	Unload model under cursor
r	Force-refresh model list
c	Clear log pane
1 / 2 / 3	Switch log source (model / server / runtime)
i / o	Toggle log input / output filters (model source)
S	Toggle inference stats in log stream
q	Quit

---

GPU support

Backend	Status	Notes
NVIDIA	✅ automatic	Requires nvidia-smi on PATH
AMD ROCm	⚙ patch needed	Swap nvidia-smi call for rocm-smi --showuse --csv in _try_nvidia_smi()
Apple Silicon	⚙ patch needed	Use powermetrics (requires sudo)
None / CPU-only	✅ graceful	GPU row hidden; everything else works normally

---

Project structure

lms-mon/
├── lms_mon.py        ← single-file app (all TUI logic)
├── requirements.txt  ← pinned runtime deps for uv/pip
├── pyproject.toml    ← PEP 517 build metadata + entry-point
├── setup.cfg         ← legacy setuptools fallback
├── LICENSE           ← MIT
├── .gitignore
└── README.md

---

Contributing

PRs welcome. Obvious extensions:

• AMD ROCm / Apple Silicon GPU polling
• /api/v0/models/{id}/stats per-model VRAM breakdown
• Export inference traces to JSONL
• Prometheus metrics exporter mode (--prometheus-port)

---

License

MIT © 2026 dcolley