yardstyx.
Menu
THE TOOLS FOR TODAY’S WORK

Find it. Run it.
Keep it in view.

Local model management and serving come first. Gateway operations and the newer terminal interface are beta; their limits are marked where they matter.

01
Core workflow

Know what’s on disk.

Find a GGUF model and keep your local collection understandable.

See the setup
  • Hugging Face search and resumable downloads
  • Local model inventory and GGUF metadata inspection
  • SHA256 verification when an upstream hash is available

In practiceSearch and downloads use the network. Check the upstream license, file size, and hardware requirements before choosing a model.

02
Core workflow

Bring up a local endpoint.

Launch llama-server with a task preset, then check its health and local resources.

See the setup
  • Task presets, including cpu, coder, and chat
  • Explicit model paths and start, stop, and health commands
  • CPU/RAM monitoring and GPU telemetry where supported

In practiceInstall llama-server separately. Hardware fit and performance depend on your model and machine; the recorded walkthrough uses a small CPU smoke-test model.

03
Core workflow

Use the controls that fit.

Stay interactive in the terminal or use commands in your own scripts.

See the setup
  • Scriptable yardstyx CLI
  • Default interactive Rich terminal interface
  • A separate, newer Textual terminal interface in beta

In practiceRich and Textual have different coverage. There is no native browser console today.

04
Beta

Inspect model traffic.

Explore routing, activity, and usage across model endpoints.

See the setup
  • Gateway routing and endpoint operations
  • Fleet and tenant inventory
  • Request logs, metering, and work history

In practiceThese paths are beta. Gateway conformance and evidence completeness are still being hardened; tenant inventory does not establish production multi-user isolation.

05
Beta

Keep your existing client.

Generate a connection profile for a separately installed Open WebUI.

See the setup
  • Dry-run inspection before writing a profile
  • A sourceable environment file for client bootstrap
  • Local OpenAI-compatible endpoint connection settings

In practiceYardstyx does not install or start Open WebUI. An existing client may retain settings in its own database.

FIT & BOUNDARIES

A few things to know.

What does local-first mean?

Your local files and local serving use your machine. Model searches/downloads, remote endpoints, and integrations can send traffic elsewhere. Review endpoint settings before handling sensitive data.

What hardware is supported?

The code includes CPU operation and NVIDIA, AMD, and Apple telemetry backends. Fit, tuning, and throughput depend on the backend, model, context size, and hardware. This site’s recorded session validates one small Linux CPU setup; it does not establish a hardware performance matrix.

Can I use a browser dashboard?

A native Yardstyx browser console is not available. Use the terminal interfaces, or connect a separately installed Open WebUI using the profile helper. Open WebUI has its own installation and settings.

What is still experimental?

Planner/agent workflows, evaluation-driven routing, replay, attestation, billing, and adaptation are experimental. Some planner paths simulate actions. Complete evidence exports and production agent controls remain planned.

Is this ready for shared production workloads?

Yardstyx is pre-release. Begin with a local test workload. Gateway hardening, installation/upgrade validation, and complete evidence linkage remain active work; tenant inventory is not a guarantee of team isolation.

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