Skip to content

Settings

Settings live behind the account avatar → Settings at the far right of the top bar.

The settings page has two dimensions: scope decides who a setting affects, section decides what you are configuring.

Four Scopes

ScopeStored inAffectsRestart needed
Boot Configconfig.yaml in the State RootStartup parameters such as port and authYes
Global Config<State Root>/workspace/.nbook/config.jsonAll projectsNo
Project Config<project>/.nbook/config.jsonOnly the current project; overrides globalNo
Browser StateLocal to the browserOnly this browser on this machineNo

The priority order is project overrides global. Preferences like font size usually belong in browser scope, models and API keys in global scope, and anything specific to one book in project scope.

Boot Config is read-only here

Settings only shows the boot state of the current process; it does not hot-write it. To change the port or auth, edit config.yaml and restart the service.

Sections at a Glance

SectionWhat it configures
Password ProtectionLogin and auth status
FrontendTheme, interface language (Chinese / English), default editor view mode
EditorProse font (including Chinese serif / sans / kai faces), size, line height, prose width, first-line indent, Monaco editor preferences
ModelsProvider, base URL, API key, default model, model library
EmbeddingEndpoint of the vector retrieval service
Cost DisplayToken pricing and currency conversion
Web ToolsConfiguration for web_search / web_fetch
Agent Profile ModelsAssign a separate model per Agent
Novel DataEndpoint of the rankings research service
ObservabilityRequest trace recording switch and how many records to keep
NeuroBook AccountAccount linking and cloud backup, see Account and Cloud Backup

Models and Cost

The Models section is mandatory — without it the Agent cannot work. A provider is whoever serves the model API; you get an API key from them and paste it in here. Entries that share an endpoint can reuse a key you already saved.

Agent Profile Models lets you spend compute by role: writer, which writes the prose, gets your strongest model, while summarizer, which grinds out titles and summaries in the background, gets a cheap one. Over months of use this is the most practical way to cut cost.

The Cost Display section handles pricing. Tokens are metered separately as input / output / cache write / cache read — the unit prices for those four usually differ a lot, and lumping them together distorts the total badly. Once prices are set, the interface converts usage straight into USD and CNY.

Observability (Request Traces)

Turn it on and every model call is recorded end to end: the normalized context, the raw request body, HTTP status, token usage, time to first token and total duration. The Trace button in the top bar opens the viewer, where you can filter by session, group by invocation, read the details across six tabs, and jump between a trace and its session in either direction.

When you are tuning a prompt or working out "why did the AI answer like that", this is the only place that shows you the truth.

Traces contain your full prose

A trace holds the complete prompt and the novel text inside it. The default is the most recent 100 per session, stored under <State Root>/workspace/.nbook/agent/traces/. Clear them before you share the machine or the directory. They never enter the shareable log bundle, but they do stay on disk. See Privacy Boundaries.

Interface Language

The Frontend section switches between 简体中文 and English. This only changes the interface text; it does not change which language the Agent speaks to you in — that comes from the model and the prompt.

Keep Reading