2.16. Settings¶

The Settings page is your central hub for configuring all aspects of Backend.AI GO. Access it by clicking the Settings (gear) icon in the sidebar.

Overview¶

Settings are organized in a left sidebar with three groups:

App

Agent

System

Use the Search settings bar at the top to quickly find any setting by keyword.

Organization-managed settings¶

When Backend.AI GO is deployed in an enterprise environment, an administrator may lock specific settings through a machine or central policy. Locked settings show a "Managed by your organization" badge and cannot be changed by the user. The lock covers only the paths the policy specifies; all other settings behave normally. On a standard (unprovisioned) install no settings are locked and the badge never appears.

Appearance¶

Theme¶

Backend.AI GO offers multiple visual themes to suit your preference:

• Glass: A modern, translucent design with frosted glass effects.

• Stained: A vibrant, colorful theme with gradient accents.

Color Scheme¶

Choose between:

• Light: Bright background with dark text.
• Dark: Dark background with light text, easier on the eyes in low-light environments.
• System: Automatically follows your operating system's appearance setting.

Sidebar¶

• Collapse sidebar by default: When enabled, the sidebar starts collapsed to maximize content area. Hover or click the expand button to reveal it.

System Tray¶

• Show system tray icon: Display an icon in the system tray (menu bar on macOS) for quick access to Backend.AI GO without keeping the main window open.

Accessibility¶

• Reduce motion: Disables theme transitions and animations. This overrides the system preference and is useful for users who experience motion sensitivity.

UI Scale¶

Adjust the interface zoom level to make everything larger or smaller. The slider ranges allow fine-tuning for your display and preference. Click Reset to 100% to return to the default scale.

General¶

Language¶

Choose the interface language. Backend.AI GO supports:

• English
• Korean (한국어)

Timestamps¶

Configure how message timestamps are displayed throughout the application. See Chat Interface > Timestamp Display for details.

Models¶

For model-specific settings such as loading parameters and generation defaults, see Model Settings & Parameters.

Generation¶

The Generation section contains inference parameters and conversation behavior settings.

• Restore last chat session on entry: When enabled (the default), navigating to the chat screen without an active session automatically reopens your most recently updated conversation. Disable this if you prefer to always start at a blank chat.

See Conversation Management > Session Restore for more details.

Tools & Extensions¶

Configure tool permissions and MCP (Model Context Protocol) servers. See MCP Integration for details.

Memory¶

The Memory section configures the persistent memory system. For a complete guide to memory banks, the Memory Viewer, auto-extraction, and memory injection, see Memory.

Voice Input¶

The Voice Input section controls push-to-talk dictation in the chat composer.

• Enable voice input: Turns the microphone button in the chat composer on or off.

• Transcription language: Choose the spoken language, or leave it on Auto-detect to let whisper.cpp identify the language automatically.

• Transcription model: Path to a whisper.cpp model file (for example ggml-base.bin). Download a model from the whisper.cpp model repository and set the path here, or use the Engines page to install it through the engine registry.

• Insert mode: Insert places the transcribed text in the composer so you can review and edit before sending. Auto-send sends it immediately after transcription completes.

Advanced¶

The Advanced section contains network settings, logging, and developer options.

Port Ranges¶

Backend.AI GO binds several local TCP port pools. You normally never need to touch these, but if another application reserves the defaults — most commonly WSL2/Hyper-V on Windows — you can move the pools to a free range here.

• Inference Server Ports: The port pool for local inference servers (llama-server). Default 39000–39019.
• System Monitor Ports: The port pool for the system monitor service (all-smi). Default 39500–39509.

The API server / router port is configured separately under Settings → API → Port (default 39080).

If you hit a "No available ports" error — especially on Windows with WSL2 — see Troubleshooting → Windows/WSL2 port conflicts for the diagnostic command and the recommended workarounds.

Connectors¶

The Connectors section (tagged Experimental) configures external accounts that agents and automations can act on:

• Email connector (IMAP + SMTP): An email account with IMAP read access and optional SMTP sending (send, reply, save draft, delete). Passwords go into the operating system's secure keychain and are never written to disk in plaintext. A Test connection button verifies the account before saving.
• Google Calendar connector: An OAuth client for Google Calendar. Begin the consent flow, allow access in the browser, then paste the authorization code back into the form. Credentials are stored in the OS keychain.

Audit Log¶

Every operation a connector performs is recorded on the Audit Log page (sidebar > Audit Log). Each entry shows the connector, the operation, when it happened, its outcome (Succeeded, Failed, or Cancelled), and a rollback verdict: No rollback needed, Rollback available, Rollback (best effort), or Rollback unavailable.

Entries with an available rollback offer a matching undo action, such as undoing a send, deleting a created item, or restoring from a pre-call snapshot. A confirmation dialog spells out what the undo will do before it runs, and some undo windows (like message recall) close after a time limit. Sensitive fields are redacted in the log.

Maintenance¶

Updates¶

• Automatically check for updates: Check for new versions when the application starts. Disable this in offline environments.

• Update channel: Choose which release channel to receive updates from.

  • Stable: Production-ready releases (recommended).
  • Beta: Pre-release versions with new features being tested.
  • Canary: Bleeding-edge builds for early testing.

  Non-stable channels may contain bugs or incomplete features. On the Beta channel, if a stable release newer than the latest beta is published, the updater offers that stable release.

• Auto-update inference engines: Keep the bundled inference engines current automatically.

• Anonymous usage statistics: Share anonymous usage statistics to help improve the application.

Backup & Restore¶

Export your settings to a file, or import a previous backup. The import dialog lists the categories found in the file (General, Inference, Tools, Conversation, Advanced, and so on) and lets you choose which to apply; importing overwrites the current settings for the selected categories. Backups never include API keys or other sensitive data.

Reset¶

Reset to defaults restores every setting to its default value after a confirmation prompt.