Settings reference: options, defaults, and when they apply
Look up settings by section and understand saved values, installation defaults, scope, and experimental options.
Settings & helpHow to use this reference#
Open Settings from the sidebar and find the relevant section below. Tables explain each option and when it takes effect. For worked examples, see settings and data.
Your displayed value takes precedence. Upgrades generally preserve existing choices. The tables distinguish new-install defaults from fallback values used when a saved configuration is missing a field.
General: appearance, permissions, and daily behavior#
| Option | How to choose | Effect and scope |
|---|---|---|
| Theme | Light, dark, or system | Applied and saved when selected |
| Language | Choose the interface language | Not a universal response-language switch |
| Default Chat Process View | Full for debugging; hidden for reading results | Changes presentation, not tool execution |
| Always on Top | Keep results visible while using other apps | Applies to the window immediately |
| Message Navigator | Return to earlier messages in long chats | On by default; hover at the chat's edge |
| Default Permission Mode | Default, Auto, or Full Access; see permissions | New-session default; sessions can override |
| External editor entry | Show the toolbar action for opening the project | Hidden by default; verify in a project |
| Commit entry | Show the toolbar commit action | Hidden by default; Review can still commit |
| Default open tool | Choose a detected editor or remember the last tool | Controls the external-editor button's main action |
| Default Reasoning Effort | Choose the starting level for each model with multiple supported levels | New sessions; existing sessions keep their choice |
| In-app notifications | Show completion notices inside Onevium | Independent of desktop notifications |
| Desktop notifications | Receive alerts outside the window | Requires OS permission; Do Not Disturb can suppress them |
| Automatic title updates | Let conversation titles follow their content | Disable to keep titles fixed |
| Keep running in background | Prevent automatic sleep during active work | Uses more power; does not guarantee macOS lid-closed operation |
| Reduce visual effects | First adjustment for flicker or stutter | Reduces decorative animation immediately |
| Disable hardware acceleration | Try if reducing effects does not help | Requires restart; may reduce smoothness |
| Check for Updates | Inspect version and available updates | Download or view the release as offered; verify after restart |
Some defaults differ between fresh installations and older configurations:
| Option | Fresh-install baseline | Missing-value fallback in older configurations |
|---|---|---|
| Theme / language | Light / Chinese | Existing choice or environment initialization |
| Permissions | Full Access | Default, or compatible existing full-access configuration |
| In-app / desktop notifications | Off / on | On / off |
| Automatic titles | Off | On |
| Background activity | On | Off |
Choose the default model through provider and conversation model controls. Default Reasoning Effort is not another model selector. General also has no independent Enter-to-send preference.
Providers: model connections#
Add or adjust providers, models, and connection configuration here. Confirm the endpoint, credentials, and model, then send a short message. Saving does not prove a working model. See providers and custom providers for the fields.
Claude CLI: shared configuration and advanced options#
| Control | Purpose | Saving, scope, or limitation |
|---|---|---|
| Extended Thinking | Adjust extended reasoning on supported models | Saves immediately; model support varies |
| 1M Context Window | Enable a long-context option for supported Anthropic models | Experimental; account and model eligibility still apply |
| Agent Teams | Enable experimental CLI teams | Separate from Onevium's newer @Team workflow |
| Teammate Mode | auto, in-process, and platform-dependent tmux/iterm2 | Shown when teams are enabled; affects Claude in a terminal, not automatic in-app selection |
| Permission Rules | Edit CLI permission JSON | Valid JSON plus Save; a different layer from the session selector |
| Advanced → Environment Variables | Change variables passed to Claude | Edit JSON, save, and confirm; retain unrelated fields |
| Advanced → other fields | Edit existing advanced configuration | Keep unfamiliar fields unchanged; Reset discards unsaved edits |
| Edit Raw JSON | Edit the whole configuration | Format, review, and save; may affect CLI using the same settings |
An unchecked CLI boolean does not describe every model's underlying defaults. Verify changes in a new conversation. Configuration can contain credentials; remove them before sharing screenshots.
Network proxy: routing and validation#
| Control | What to enter or choose | Effect and check |
|---|---|---|
| Mode | Off, App Only, Global; normally Off without configuration | App Only is for Onevium; Global changes system proxy settings |
| Proxy Type | HTTP, HTTPS, or SOCKS5 as supplied | The proxy protocol, not the destination site's protocol |
| Server Address / Port | Separate host and port fields | Enter a hostname and port number; leave fields to save |
| Username / Password | Only when the proxy requires authentication | Password is saved separately and can be cleared |
| Extra Direct Addresses | Additional domains or hosts that should bypass the proxy | Local addresses already bypass it |
| System Proxy Status | Applied or Not Applied in Global mode | Check this, not just the selected mode |
| Test Proxy Server | Save parameters and select Test | Scope explained below |
| Check Browser Exit IP | Open the check page in Global mode | Verify the browser's exit separately |
Test scope: the Test action creates a proxy tunnel to ifconfig.me:443, then attempts /ip. An established tunnel can return success even when IP retrieval fails. If an IP is shown, it describes that test request's exit.
This does not test Anthropic or a custom model endpoint, validate a model account, or prove a separate browser uses the same route. Verify model access with a real conversation separately.
Switching away from Global attempts to restore the previous system proxy. Windows and Linux SOCKS have system-authentication limitations; other apps may request credentials separately or omit them.
Original proxy settings with isolated example values. No system proxy change or test was performed. This version tests an IP echo service, not the model account.
Activity: local usage and cost observations#
| Control | What it shows | Useful approach |
|---|---|---|
| Year and heatmap dimension | Messages, sessions, tokens; defaults to current year and messages | Compare the same year and dimension |
| Time range | Today, yesterday, 7/30/90 days; defaults to 7 days | Narrow the range to investigate a spike |
| Provider filter | All or one provider | Identify which connection contributes usage |
| Include cache | Whether chart data includes cached tokens | Off for that chart layer by default; compare like with like |
| Cost, tokens, cache hit rate | Local records and price estimates | Notice unknown-price exclusions and pricing dates |
Estimated cost, cache savings, and an official invoice are different. This section also differs from subscription limits below.
Tools: built-in capability groups#
| Group | Purpose | Default or behavior |
|---|---|---|
| Schedules | Manage automations | On by default |
| Widget | Create visual artifacts | On by default |
| Recap | Generate work recaps | On by default |
| Services | Manage services | On by default |
| Channels | Let the assistant manage channel configuration | Off by default; not the switch for connected bots |
| Browser / Sessions | Separate browser and conversation groups | Enable the group needed for your task |
| Continuity / channel runtime | Automatically supplied runtime tools | Informational rows, not switches, in the current interface |
Disabling a group prevents further calls; newly enabled tools are available with the next message. Switches do not undo completed actions or stop an active task. Tool counts and token estimates need not match fixed numbers in a tutorial.
Personalization and memory#
| Control | Purpose and when to change it | Default, scope, and effect |
|---|---|---|
| AI Autonomy master switch | Controls related autonomous processing | On by default; user rules also depend on it |
| User Rules | Stable preferences and habits | Off by default; up to 4,000 characters; autosaved; user scope |
| Memory extraction | Extract durable information from conversations | On by default; not a switch for saving all chat history |
| Extraction frequency | High: each turn; balanced: every 3; economy: every 5; or off | Balanced by default; off stops new extraction, not recall |
| Deep extraction | Include reasoning, alternatives, and tradeoffs | Off by default; adds extraction usage |
| Extraction language | Language of new memories | Follows interface by default; does not translate existing items |
| Recall threshold | Strict, balanced, relaxed | Strict for fresh installs; balanced fallback for older configurations |
| Cross-project search | Retrieve relevant memories from other projects | Off by default; not blanket cross-project injection |
| Memory context length | Compact, balanced, detailed | Balanced by default; memory volume, not the model window |
| Synthesis frequency | Daily, weekly, off | Daily by default; off does not stop old-memory decay |
| Included extraction projects | Choose projects that create new memories | All included by default; exclusion does not delete existing items |
| Extraction model | Model and provider for background extraction | Defaults to a low-cost role; must be callable |
| Daily budget | Limit automatic extraction spending | Fresh baseline $10; older fallback $0.10; not total chat spending |
| Memories list | Search, filter, add, edit, pin, disable, restore, delete | At the bottom of Personalization; project options come from existing memories; adding without a project can be global |
Tool Assessment still has a settings row, but its related public tools are disabled; do not build a new workflow around it. Buddy is hidden. The separate “inject pinned memories” switch is also hidden; this differs from pinning an item in the list.
@Memory initiates work on Claude Code rule or memory files; it does not mean saving a database memory to this list. See memory and recaps. Configure Daily Briefing through its own page's gear.
Browser: backend, viewport, and profiles#
| Control | Purpose | Operation and limitation |
|---|---|---|
| Browser backend master switch | Enable AI browser-control entry points | On by default; off does not clear site data |
| Native / External / Auto | Native by default; Auto prefers native, then external | Next browser call; external needs a working Chrome connection |
| Default viewport | Preset dimensions for new tabs | Defaults to panel size; existing tabs can override |
| Clear browsing data | Clear cookies, cache, and site data in the associated shared partition | Affects shared login state; does not guarantee clearing every profile |
| Profiles → New | Create separate cookie storage | Does not automatically switch the current or next conversation |
| Profiles → Import cookies | Import from detected browsers/profiles | Depends on platform and actual results; verify login afterward |
| Profiles → Delete | Delete a non-default profile and its login data | No delete action for the default profile |
| Control Chrome | Configure external Chrome control separately | Install the offered extension, enable, wait for readiness, and check tools in a new conversation |
Regular conversations currently have no visible Profile selector; default browser calls still use the default profile. Creating a profile does not establish account isolation between tasks. See the browser guide.
Voice: recognition, polish, and history#
| Control | How to configure it | How to verify |
|---|---|---|
| Recognition models | Download a suitable language model, then select Use | Verify with a short recording |
| AI transcript polish | Enable for homophone and punctuation cleanup | On by default; sends text to the model provider |
| Polish model | Choose a working model | Defaults to a low-cost role |
| History and recordings folder | Inspect records and local audio | Check the latest record and open its audio file |
| Enable voice input | Off hides the microphone and disables its shortcut | Check the composer; on by default |
| Microphone selection | Authorize access, then choose system default or a detected device | Verify the selected device with a recording |
| Copy transcript | Copy a record's text | Paste into an empty draft to check |
| More actions → Download recording | Export the selected audio | Open the exported file |
| More actions → Delete recording | Delete the selected recording record | Confirm the target first, then inspect the list |
| History pagination | Browse more records | Check dates and content across pages |
Local recognition and AI polish are separate stages. There is no visible recording-retention-days setting. See voice dictation.
Privacy: diagnostics and cleanup#
Diagnostic event logging is off by default. Enable it for a specific investigation, reproduce the issue, then disable it. Review session metadata before sharing logs.
Clear Local Data currently deletes the database and companion files, media, uploads, and images. It is not a way to clear CLI configuration, project files, all browser profiles, or voice recordings. See settings and data for locations and cleanup distinctions.
Local storage does not mean offline processing; data destinations still depend on the services you use.
License: account, license, and devices#
Inspect email, license status/type, start and expiry dates, and remaining days. Replacement is controlled by account eligibility. When offered, enter the new activation code, submit, then verify the updated status and dates.
The device list shows name, platform, version, and recent activity. It has no delete-device action. Signing out does not clear local conversations.
Archived sessions: restore or permanently delete#
Use search, project, and date filters to find a conversation. Preview it before restoring or deleting. After restoration, check the content in its project. Bulk actions are supported; Delete all applies to the filtered results.
Permanent deletion removes the selected conversations and messages and cannot be undone. Practice with a demo conversation rather than real history.
Keyboard shortcuts: edit, resolve conflicts, and reset#
- Search for a command, select edit, and press a combination.
- A valid, conflict-free combination saves directly; Escape cancels recording.
- Transfer resolves a conflict by removing that binding from the previous command.
- Clear a binding, restore one default, or confirm a reset of all bindings.
Commands have global, chat, file-panel, browser, and other scopes. Some entries, including double-Escape stop and Shift+Tab Plan switching, are fixed references. macOS screenshots and the double-Command gesture also require their respective system permissions.
Usage and billing: subscription limits#
This section shows the official Claude subscription account, limit windows, and behavior analysis, with refresh or retry. Relevant limits require a Pro/Max subscription login. API keys and third-party providers may not supply this data.
Temporarily unavailable does not mean zero quota. Read local token/cost statistics in Activity separately from these subscription limits.
If an option is missing#
- Check the applicable version at the top of this page and whether the relevant feature is enabled.
- CLI Agent Teams and the 1M context option remain experimental and depend on model, account, or platform support even when a control is visible.
- Platform-specific options appear only where supported; inspect the settings on the device where you will run the task.
- After changing a setting, verify it with the relevant message, notification, browser action, or recording.
Next steps#
Return to the settings map for common adjustments, use permissions to choose execution controls, or follow workflow tips.