Skip to content

Backends, apps, monitoring, and settings

App overview

Manage inference backends

Backends groups runtime variants by capability. The rail filters All Backends, Installed, Available, Updates, and Experimental. It can also show technical details, unsupported backends, and backend logos.

Use Install, Update, Update all, or Uninstall as available. A backend that needs manual setup provides Setup guide instead. Installation progress and failures appear on the backend card and in Downloads.

Some installed variants provide a terminal button for persistent, backend-wide CLI arguments. Those defaults are different from a model's Configuration and from the session-only Effective settings override.

Browse applications

Apps loads Lemonade's curated marketplace and groups entries by category. Featured apps appear first. Use Visit to open an app or its site, Guide for setup instructions, and Video when one is provided. These actions open external links; The app does not install third-party applications.

The title-bar search searches app names, descriptions, and categories and prioritizes app results while Apps is open.

Monitor the server

Monitor data describes the connected server, not only this GUI window.

  • Performance shows server health and uptime, prompt and generation throughput, active inference slots, system CPU/RAM/GPU/VRAM, cache use, last inference details, loaded models, and configured model limits. Pause freezes dashboard updates; Resume restarts them.
  • Telemetry captures request traces. Search or filter by request kind or errors, inspect overview and messages, replay a request with changed parameters, compare results, or use Improve with an eligible model. Requests can also be created manually, deleted, cleared, or exported to the clipboard as JSON.
  • Logs streams server output. Filter by text, minimum level, capture level, or source; clear the displayed buffer or reconnect when the stream drops.

Some cards remain empty until a compatible model is loaded or a request has run. Telemetry capture and server logging can include prompt content; review your privacy requirements before enabling or exporting them.

Configure Settings

Settings is divided by purpose:

  • Server: endpoint, optional API key, connection status, and local-data cleanup.
  • Chat: optional browser history, collapsed-reasoning default, default TTS model, and automatic speech playback mode.
  • Memory: server-owned automatic eviction, eviction threshold, and maximum loaded models per type. Pinned models are protected from eviction.
  • Model storage: the server's model-cache and external custom-model directories, plus the server-startup model-update check.
  • Cloud providers: server-owned OpenAI-compatible provider definitions. Provider keys come from LEMONADE_<PROVIDER>_API_KEY environment variables or ephemeral server memory and are not written to config.json.
  • MCP gateway: the built-in Lemon-Tools MCP endpoint and discovered tools. The current GUI reports external MCP administration as unavailable while its local client host is being introduced.
  • Help & support: documentation, release notes, GitHub, and Discord.

Save settings applies changes in Chat, Memory, and the model-update controls under Model storage. Server connection and directory changes instead use Connect and Save directories. In Chat, Reset defaults resets the local draft; save it to persist the reset. In Memory and model-update settings, Discard changes restores the server's current values.

Theme is a title-bar control and is local to the GUI client. The app does not currently expose a zoom setting in Settings.

Downloads, progress, and errors

Open the title-bar download icon to see active, paused, completed, cancelled, and failed model or backend downloads. Rows show file progress, bytes, speed, and estimated time when the server supplies them. Expand a row for file details.

Depending on state, a row can be paused, resumed, cancelled, retried, removed from the list, or have its partial files deleted. Clear completed removes terminal history rows. Finalization can continue briefly after the byte count reaches 100%.

Model preparation also appears inline in Chat, and model, backend, connection, MCP, and download failures appear near the control that initiated them. Lemonade App does not currently provide a general-purpose Jobs workspace; use the Job Engine API for server job sequences.

Remote servers and multiple clients

A single lemond can serve several desktop or browser clients. Server-owned state is shared:

  • downloaded and loaded models, pin state, and backend installations;
  • persistent Models > Configuration values;
  • memory, storage, model-update, and cloud-provider configuration; and
  • Monitor data and server logs.

Client-owned state is independent:

  • endpoint, session API key, layout, theme, and navigation state;
  • Favorites, chat history, and Chat display or speech preferences;
  • request-time Chat sampling and tool selection; and
  • session-only Effective settings backend overrides.

One client's load, unload, delete, configuration, or server Settings action can therefore affect other clients. Its theme, history, Favorites, and chat sampling do not.