Settings

Settings is where you choose Nova's models, tune memory and voice, connect providers and channels, and make backups. Open it from the sidebar (Settings, the gear) or from the gear on the right edge of the main screen. It has seven tabs:

How the window works

• Tabs are on the left. Each tab opens with its title and a one-line description.
• Search settings filters the open tab by section title, description, field label and value. Matching sections open on their own.
• Sections fold and unfold when you click their header, and each shows how many settings it holds. The window remembers the tab and sections you had open.
• Model fields are dropdowns filled from the provider's catalogue, with the context size shown, for example (128K). Refresh models reloads the list, and the pencil (Manual model ID) lets you type an ID yourself. When you change a provider, the list reloads, and the first model is picked if your current one is not offered.
• Save stores everything and closes the window. Cancel, the close button, Esc or a click outside the window discards your edits.

Saved changes apply to open chats straight away, with no restart.

Agent

The choices that shape everyday conversations. Profile & Behavior is open when you arrive.

• Profile & Behavior:

  • Default agent profile: the assistant new conversations start with (Nova by default).
  • User Name: the name shown in chat and in the greeting on the main screen.

• Chat Model:

  • The provider and model Nova uses for conversations, and its Thinking effort.
  • The advanced fields: base URL, rate limits, extra parameters, vision support and chat compaction thresholds.
  • See Models and providers.

• Memory: about thirty switches and numbers for how memory is recalled and kept. The ones people change most are:

  • Memory auto-recall enabled
  • Auto-memorize enabled
  • Journal consolidation enabled and Journal synth enabled, which keep the Journal and its notes up to date
  • Journal searches per turn
  • Semantic memory reranking
  • Temporal decay
  • Auto-memorize images

  Keep the defaults unless you have a reason to change them.

• Voice: how Nova listens, speaks and narrates.

  • Listening:
    • Microphone device is chosen in your browser, using Request permission to select device. It is kept in this browser rather than saved with the window.
    • Speech capture (Complete utterance or Streaming), language, the silence thresholds, and barge-in, which lets you interrupt Nova by speaking.
  • Speaking:
    • Voice Assistant Engine: Gemini Live (the default) or Classic STT + TTS.
    • TTS Provider: ElevenLabs or Groq. Each has its own voice choices, and ElevenLabs also has voice-tuning sliders.
  • Narration: Speech Runtime Mode (Voice Assistant or Narration Companion), the narration mode, and a style hint.

Voice language and personality for live voice are set in the composer's model picker, on the Voice tab.

System

Advanced operating controls for background work. The tab's own advice is the right one: keep the defaults unless you have a specific reason to change them.

• Support Model, Autonomy Model and Safety & Recovery: the provider and model for each background role, with the same advanced fields as the chat model. See the model roles.
• Browser Automation: how the built-in browser behaves.
  • Browser Profile Mode: Ephemeral (the default) starts clean every time. Persistent profile keeps sign-ins. Attach to existing Chrome uses a Chrome you already run.
  • Also here: whether all agents share one profile, how many tabs Nova may use per chat (1 to 5), the total tab cap (1 to 50), and whether the view should Follow Nova or Stay on my tab.
  • Some fields only appear after you save a mode and reopen Settings.
• Memory Search Model: the model that indexes memory. Changing it rebuilds the index.
• Runtime Limits: the folder knowledge is imported from, and token-budget controls for very long turns.

External

Connect messaging, APIs, provider accounts and outside services.

Social

Credentials for Nova's social research tools and the Social feed: Reddit, Instagram, X, GitHub, Telegram, Discord and TikTok. These are the same fields as Social → Settings → Platform Credentials. See Integrations for what each platform needs.

MCP and A2A

• External MCP Servers: Open shows the JSON editor for your servers. Apply now connects them straight away, and a live list shows each server's status and tools. Two timeouts set how long to wait for a server to start and for a tool call.
• SyntrofAI MCP Server: Enable SyntrofAI MCP Server lets MCP clients use Nova. The connection example shows the addresses.
• SyntrofAI A2A Server: Enable A2A server lets other agents call Nova.

The two server switches take effect when you press the main Save. The full walk-through is on Integrations.

Developer

Shell Interface chooses how Nova opens a terminal on your instance: Local Python TTY or SSH. Most people never need to change it.

Backup

Create a backup

The default selection covers your settings and keys, saved chats, scheduled tasks, uploads, and imported knowledge files. It is not a complete copy of memory. To keep memories as well, use Export Namespace in the Memory dashboard.

Restore a backup

Saved chats from the backup come back after the restore.

Related