Crate studio_worker 

Library surface for the studio-worker binary.

Exposes the worker’s modules so integration tests (and downstream tooling) can drive the contract without going through the CLI.

Modules

admission

Admission: does a load or transient job fit in device memory now? (see docs/runtime/model-lifecycle.md).

auto_register

Auto-register state machine — the only registration path.

autostart

Autostart-on-login toggle. Writes per-OS artefacts so the desktop UI (tray) comes back after a reboot:

catalog

Local model catalog — the offline equivalent of the studio’s studioModels registry.

cli

Clap CLI definitions, kept out of main.rs so they’re testable.

config

Persistent config in ~/.config/minis-studio-worker/config.toml (Linux/macOS) or %APPDATA%\minis-studio-worker\config.toml (Windows).

control

What the tray UI can see and do in the daemon, independent of HTTP.

daemon_api

Wire types of the daemon-control routes on the local API, shared by the daemon (which answers them) and the tray UI (which reads them).

daemon_client

Blocking client for the daemon’s local API, used by the tray UI.

daemon_link

The tray UI’s link to its daemon.

daemon_lock

One daemon per config directory.

engine

Pluggable inference engines, generalised to all task kinds.

host

The model host: owns loaded models, drives each model’s lifecycle, persists residency and enforces admission and exclusive groups (see docs/runtime/model-lifecycle.md).

http

Thin reqwest wrapper around the studio API.

job_gate

The worker’s single-job reservation gate.

job_log

Per-job log capture.

job_run

One job’s bookkeeping, shared by every job path (studio offers, local transient jobs, lane requests, streaming sessions).

lifecycle

Per-model lifecycle state machine (see docs/runtime/model-lifecycle.md).

loaders

The worker’s in-process model loaders, one per engine, behind the model host’s ModelRuntime (see docs/runtime/model-lifecycle.md).

local

Local image generation — submit a prompt straight to the engine, no studio.

local_api

Always-on local HTTP API for image generation (127.0.0.1 only).

net

Shared transport-level guards for everything the worker downloads.

residency

Persisted residency: which catalogue models the operator wants loaded (see docs/runtime/model-lifecycle.md).

runtime

Long-running auto-update task + one-shot CLI helpers.

secrets

Shared entropy primitives for every locally-minted credential.

service

OS service install/uninstall. Linux: systemd –user. macOS: launchd plist template (written but not loaded — operator runs launchctl load). Windows: schtasks template (written but not registered — operator runs the printed command).

stt_stream

Streaming speech-to-text: the /transcribe session protocol, voice activity detection and the LAN listener that serves loaded models.

sys

Host-system probes: hostname, OS user, VRAM.

telemetry

Sentry telemetry — opt-in error/panic reporting.

thumbnail

Thumbnails of image jobs, so the tray UI can show what a job made.

types

Shared wire-format mirrors of the studio API.

ui

The tray UI: an egui window + system tray that is a client of the daemon (studio-worker run). See docs/runtime/daemon-and-tray.md.

update

Auto-update: poll a GitHub Releases feed, download cargo-dist’s platform installer when a newer semver is available, and re-exec ourselves so the new binary takes over.

ws

WebSocket worker-channel module.

Constants

AGENT_VERSION

RELEASE_NAME

Sentry release identifier in the <pkg>@<version> form expected by Sentry’s organisation-wide Releases feature. Bare version strings collide across projects in the same org, so we namespace with the crate name. Matches what sentry::release_name!() would expand to.

Functions

run_cli

Dispatch table for the CLI subcommands. Lives in the library so we can drive it from tests without invoking the binary.