FOR DEVELOPERS & LOCAL AI OPERATORS

Your models.
Your machine.
Your measure.

Manage GGUF models and run local inference from one terminal. Find a model, start llama-server, and keep an eye on the process.

Setup guide Watch a real session

Pre-release · Private source · Python 3.11+

FIG. 01 / YOUR WORKFLOW AT A GLANCE

yardstyx / field guideILLUSTRATION

DISCOVER / LOCAL WORKFLOW

Find your starting point.

Search Hugging Face from the same terminal you use to run your models.

yardstyx search qwen 7b instruct

Browse first. Choose a model that fits your machine.

Explore model management

RUN / LOCAL WORKFLOW

A preset. A local model. Go.

Launch your most recent local model with GPU layers disabled.

yardstyx run fast --ngl 0 --ctx 256

Requires a downloaded model and a separate llama-server installation.

Set up your first session

OBSERVE / LOCAL WORKFLOW

Keep the process in view.

Check server health, then open the live local resource monitor.

yardstyx health --port 18081
yardstyx monitor 18081

Run these commands separately. Leave the monitor with Ctrl+C.

See serving and monitoring

Explore actual commands. This guide does not execute them or show live telemetry.

GGUF MODELSLLAMA-SERVERCLI + TERMINAL UIGATEWAY IN BETA

01 / THE EVERYDAY WORK

From a file on disk
to a server you can inspect.

Start with the local model workflow. Explore beta routing and activity tools when you need to bring more endpoints into view.

Find and inspect.

Search Hugging Face, resume a download, and inspect the metadata of the GGUF files you already have.

Model management

Run and check.

Choose a preset, start llama-server, and check health and local resources from the terminal.

Serving & monitoring

Inspect activity.

Explore beta gateway routing, request logs, and usage history. These operations are under active hardening.

Beta gateway operations

02 / THE ACTUAL TERMINAL

One small model.
One complete local session.

Download, inspect, launch, health-check, and stop. This recording uses a tiny CPU test model so the process is easy to follow.

RECORDED LOCAL SESSIONLinux · CPU · b11904

$ yardstyx --version
yardstyx 0.1.0
$ yardstyx download ggml-org/models --file tinyllamas/stories15M-q4_0.gguf --dir ./models

  Downloading: tinyllamas/stories15M-q4_0.gguf
  → models/stories15M-q4_0.gguf

  stories15M-q4_0.gguf ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 19.1/19.1 MB 12.9 MB/s 0:00:00


  ✓ Downloaded: stories15M-q4_0.gguf  (18.2 MB)

  ✓ SHA256 verified
$ yardstyx info ./models/stories15M-q4_0.gguf
╭────────────────────────────── stories15M-q4_0.gguf ──────────────────────────────╮
│                                                                                  │
│   Architecture             llama                                                 │
│   Name                     llama                                                 │
│   Context length           128                                                   │
│   Embedding length         288                                                   │
│   Layer count              6                                                     │
│   Attention heads          6                                                     │
│   Tokenizer model          llama                                                 │
│   File size                18.2 MB                                               │
│   GGUF version             3                                                     │
│                                                                                  │
╰──────────────────────────────────────────────────────────────────────────────────╯
$ yardstyx run fast ./models/stories15M-q4_0.gguf --ngl 0 --ctx 256 --port 18081

  Preset: fast  ctx=256  ngl=0  port=18081
  Model:  stories15M-q4_0.gguf

  ✓ Server up on :18081  alias: stories15M-q4_0
$ yardstyx health --port 18081
  ● Server up  :18081

  llamacpp:prompt_tokens_total                        0
  llamacpp:prompt_tokens_cached_total                 0
  llamacpp:prompt_seconds_total                       0
  llamacpp:tokens_predicted_total                     0
  llamacpp:tokens_predicted_seconds_total             0
  llamacpp:n_decode_total                             0
  llamacpp:n_tokens_max                               0
  llamacpp:spec_decode_num_draft_tokens_total         0
  llamacpp:spec_decode_num_accepted_tokens_total      0
  llamacpp:spec_decode_num_drafts_total               0
  llamacpp:prompt_tokens_seconds                      0
  llamacpp:predicted_tokens_seconds                   0
$ yardstyx stop --port 18081
  Stopped PID 689651
  Stopped 1 process(es).

Actual output captured on October 29, 2026. Small smoke-test model, fast preset with GPU layers disabled, 256-token context. Idle gaps shortened and temporary paths normalized; output is not a benchmark. Download the recording · Read the transcript

Test conditions and verification

Yardstyx 0.1.0, llama.cpp b11904, Linux x86_64. The model is TinyStories 15M Q4_0, for checking the serving path rather than useful assistant answers. Every recorded command exited successfully. Its downloaded file’s SHA256 was independently checked against upstream metadata.

Test model · Backend release · Verification details

03 / BEFORE YOU START

Know what
you’re bringing home.

Explore capabilities and limits

Who is Yardstyx for?

Developers and local AI operators who want to manage GGUF models and llama-server from one terminal. The product is pre-release; begin with a local test workload.

What do I need?

Python 3.11+, the Yardstyx source, and a separate llama-server installation. Model size and context determine memory needs. The setup guide includes a small CPU smoke test.

What stays local?

Local model files and local serving stay on your machine. Searching or downloading models, connecting remote endpoints, and using external integrations introduce network traffic.

Is there a browser UI?

Yardstyx has terminal interfaces. A native browser console is not available today; a separately installed Open WebUI can connect through a generated profile.

Your first local session

A model. A server. A clear next step.

Have an authorized source checkout? Follow the setup guide and check your first local endpoint.

Setup guide

Your models.Your machine.Your measure.