Settings

Open Settings — Defaults to change the defaults used by new conversations.
An Unsaved Changes badge appears after an edit.

The normal tabs are General, Models and Providers, Generation,
Files and RAG, Appearance, and About.

General

Default Agent Type chooses the runtime for new requests:

• Local Model uses a model installed on this computer.
• Online Model uses the selected online provider.

Changing this default does not install a local artifact or add online
credentials. Check Models before
starting a conversation.

Models and Providers

When the agent type is local, Local Model selects the model ID. This list
comes from model catalog data and can include a model that is not installed.
Use the Models page's Use action to select a concrete installed artifact.

When the agent type is online, the tab shows Online Provider and Online
Model. Changing either also changes the default agent type to online. The
available provider list can include OpenAI, Anthropic, xAI, Groq, Hugging Face,
Moonshot AI, Z.AI, DeepSeek, OpenRouter, a self-hosted server, or a custom
provider, depending on live catalog data.

There is no API-key, endpoint, or sign-in field here. Selecting a provider does
not configure its credentials.

Generation

Generation Parameters controls sampling defaults:

• Temperature, 0 to 2, controls randomness. Lower is more focused and
  higher is more creative.
• Max Tokens, 1 to 4,096, limits response length.
• Top P, 0 to 1, controls nucleus sampling.
• Frequency Penalty, 0 to 2, discourages tokens in proportion to how often
  they have already appeared.
• Presence Penalty, 0 to 2, discourages tokens that have appeared at all.

These values are sent with normal Chat requests. A provider or model can still
apply a smaller supported limit.

Files and RAG

This tab displays Enable RAG by Default and Default File Archives. The
archive list shows uploaded files and has Select All and Clear All.

Known v1.0 limitation: the app saves these values, but Chat does not currently
read them. They do not automatically add file context to a new conversation.
Use an explicit @ file reference as described in
Files and RAG.

Appearance

UI Preferences exposes:

• Font Size — Small, Medium, or Large
• Show File Previews
• Enable Notifications
• Compact Mode
• Show Workflow Thumbnails

Known v1.0 limitation: these preferences are saved but are not applied by the
current workspace pages. Changing them does not yet resize text, hide the
explicit Files preview, enable desktop notifications, compact the layout, or
add workflow thumbnails. Treat them as stored preferences for future UI
wiring.

About

About shows the Pitchblend and Geist versions plus information reported by
the active local service. It has no editable settings and therefore no action
footer. See
About and System Information.

Save, cancel, or reset

On every normal tab except About:

• Save Changes writes all edited settings. It changes to Saving... and
  then shows Settings saved successfully.
• Cancel restores the last values read from the service.
• Reset to Defaults asks Are you sure you want to reset all settings to
  their default values? This cannot be undone. A successful reset shows
  Settings reset to defaults successfully.

Known v1.0 quirk: Reset to Defaults sets Max Tokens to 16, even though a
new settings record normally defaults to 4,096. Open Generation after a
reset and choose the response limit you actually want.

Loading and error states appear as Loading settings... or Error loading
settings: ... with Retry. Save and reset errors appear in the status
notice and can be dismissed with X.

The app also stores backup-provider data internally, but there is no normal UI
to edit it. A conditional Developer tab can appear in development builds;
it is not a packaged customer setting. See
Developer and Attach Modes.