Settings

Use Settings to configure AI providers, email, API access and updates for this installation. Only the site owner can open this page.

Open Settings from the user menu (top-right) → Settings. Only the site owner sees this option.

Site identity

Your website's name and description, plus the browser favicon.

Favicon upload: Click the Upload button or drag-and-drop a .ico file onto the favicon area. A live preview shows immediately. To remove a custom favicon, click the trash icon next to it.

Click Save Identity to apply changes. Favicon replacement is an immediate live change, separate from website publication and snapshot rollback.

AI provider

Configure which provider powers built-in website generation. You can leave this unconfigured for manual editing or Studio MCP. External-client authentication is separate.

1. Select a provider from the list below
2. Enter your API key - encrypted at rest using AES-256-CBC
3. Test Connection - click to verify your key works before saving
4. Choose a Default model - VoxelSite loads available models from your provider automatically
5. Max Output Tokens - sets your output ceiling per generation request (default: 32,000). It is a maximum, not a target. VoxelSite never silently raises this setting; the model and action can lower the effective limit. Cost depends on actual usage.

You can switch providers and models at any time. Your previous API keys are preserved - switching back restores your configuration.

Output limits

Below Max Output Tokens, Settings shows your Configured ceiling, the effective limit for Full generation, and the limit for Section edits. It also shows the selected model's output limit, context window, and where those limits come from. Section edits, new sections, and focused AI Router edits use at most 16,000 output tokens, or your configured ceiling if it is lower.

VoxelSite uses documented limits for known model IDs and limits reported by your provider when available. Models without known limits use conservative defaults. The effective output limit also reserves space in the model's context window; it is not extra capacity on top of that window.

For OpenAI Compatible, site owners can use Advanced model limits to enter the Model output limit and Context window documented by your server. Enter both as positive whole numbers, with output no larger than context, or leave both blank to use detected limits. These overrides apply only to that endpoint and exact model ID. Changing endpoints does not transfer the old limits. Your Max Output Tokens ceiling still applies.

If a generation reaches its limit, review any partial result and ask the AI to finish the missing work. Increase your ceiling only if the selected model supports more output, or split the request into smaller steps. See Common Issues.

Expert review after generation

Below the token setting, an optional toggle enables a post-generation quality review. When turned on, VoxelSite makes an additional AI call after each generation to surface heuristic HTML, accessibility, and SEO suggestions - for example, heading hierarchy issues, missing alt text, or broken internal links.

Results appear in a collapsible "Expert Review" panel in the chat, below the list of modified files. Each suggestion includes the severity (error, warning, or info), the affected file and line, a description, and a suggested fix. Click Add to chat → on any suggestion to append a targeted follow-up prompt to the chat input.

This feature is off by default. It does not auto-fix issues - results are advisory, intended for users comfortable reviewing web-design advice. Enabling it adds another AI request and its token usage.

Design review after generation

The optional Design review toggle below Expert review requests feedback on the page design. When turned on, VoxelSite makes one more AI call after each generation that writes or restyles a page. The reviewer sees only the finished page, the design tokens, the navigation and footer, and the design brief the builder was given - never your prompt or the conversation.

It scores the page out of 10 against a studio bar (9 to 10 is work a senior designer would sign, 5 to 6 a competent template), names the three gaps that would move the score most with a one-sentence fix each, and lists any patterns that read as machine-made - gradients and glows, an icon on every card, identical card grids, brochure headlines.

Results appear in the same "Expert Review" panel in the chat. The score and the gaps show as warnings when the page scored under 8 and as suggestions otherwise. Click Add to chat → on a gap to turn it into your next prompt.

This feature is off by default and advisory only - nothing is changed automatically. It adds one extra AI call per generation. It does not run for visual-editor section edits, inline code edits, or SEO optimisation.

Click Save Settings to apply changes.

Supported providers

AI Router

AI Router sits below AI Provider. With it on, TypeSafe's Jev assesses requests so VoxelSite can choose between your Faster and Stronger models and select suitable prompt instructions. Both models belong to the provider you saved above.

Turn on Enable AI Router, enter your TypeSafe key, choose both models and click Save AI Router. Off uses your default model with no TypeSafe calls. If routing is unavailable or uncertain, VoxelSite reports the fallback and uses normal editing.

A stored key shows Remove key beside the field. Click it to stage removal and turn the router off; Undo cancels the choice. Save applies both changes. No removal control appears when no key exists.

The AI Router guide explains setup, model-family ordering, request types and what VoxelSite sends to TypeSafe.

Email & notifications

Configure how VoxelSite sends notification emails when forms are submitted. Four delivery methods are available:

• SMTP - presets for Gmail, Outlook, Yahoo, iCloud, Zoho, and custom servers
• PHP mail() - uses your server's mail function; delivery depends on the host's mail setup
• Mailpit - for local development and testing
• None - email disabled; submissions are still stored and visible in the Studio

See Email Configuration for detailed setup instructions, SMTP presets, security, and troubleshooting.

AI knowledge

After you've generated a site, the AI Knowledge card shows what VoxelSite has learned about your business and design preferences. Two files are displayed:

• Site Memory - facts the AI has picked up from your conversations: business name, phone number, opening hours, team members, key services. Shows the number of stored facts.
• Design Intelligence - the AI's design decisions: chosen fonts, color temperature, spacing rhythm, layout approach. Shows the number of tracked decisions.

Click either card to open a read-only viewer showing all stored values.

The Knowledge card is read-only. Update the underlying draft through Chat, the Code Editor or scoped external authoring. These private memory and design-context files are excluded from publication, public MCP reads and portable exports. Put customer-facing facts in the appropriate public site-data document.

For more on how the AI uses this knowledge, see The AI Engine → Site Memory.

AI usage

Track your AI token consumption across all models you've used. The usage card shows:

• Total requests - how many AI generations you've triggered
• Input tokens - total tokens sent to the AI (your prompts plus site context)
• Output tokens - total tokens generated by the AI (the code and content it writes)

If you've used multiple models, a per-model breakdown appears below the totals.

Usage data persists across design switches and site resets - starting a new design or resetting the website does not clear your usage history. The reviewed Reset Installation flow can remove installation data; inspect its exact scope and final boundary before approval.

Token counts help you estimate costs. Check your AI provider's pricing page for per-token rates.

System status

A snapshot of your runtime environment:

Use this information when reporting issues or checking whether your server meets the requirements.

Connections

Open Settings → Connections to activate the private workspace and enable Studio MCP. Choose the account and access needed by each connection. Save a newly revealed credential once in the client’s secret storage; do not put it in website source or shared configuration.

Connection status distinguishes authentication from successful tool use and permission failures. Copying setup text does not prove a connection works. Follow the guide for Codex, Claude Code or another MCP client. These setup instructions are separate from version-specific client qualification.

Review pending administrative requests here. An approval names the connection, operation and exact effects and expires after a short period. A general connection grant does not authorize every owner action.

Update

Stage the upgrade ZIP in Settings, or choose Prepare Git update on a supported host. Staging does not activate code. Complete validation, review the affected files and preservation list, then approve the exact update as owner.

Use Continue update or Refresh status if needed. Reload Studio after success. Existing installations need the upgrade archive; the fresh-install ZIP includes default homepage and hosting files. See Updating.

Interrupted changes

Open the recovery controls when an operation is interrupted or its original actor no longer has access. Inspect the retained operation, affected paths and available recovery action. Owner recovery can restore retained originals; it does not grant permission to continue another actor’s edit.

Keep recovery archives until the operation is resolved. See Administration and Troubleshooting.

Server logs

Download or delete server log files for debugging. Each log entry shows the filename, line count, file size, and last modified date.

• Download - click the download icon to save a log file to your computer
• Delete - click the × icon on individual logs, or Delete all to clear everything

Log files are stored in _studio/logs/ and rotate daily. Share only relevant redacted excerpts; remove secrets and personal data.

API access

Enable and configure the Agent API - authenticated REST endpoints for external AI agents and automation tools.

Enable

Toggle Enable Agent API to allow external tools to connect. The API is disabled by default and blocked in Demo Mode.

API keys

Click Generate Key to create a new API key. Each key has:

• Label - a name to identify the key (e.g. "Zapier", "My Agent")
• Role - sets the permission ceiling: Agent, Editor, or Viewer
• Scopes - granular permissions within the role (e.g. pages:read, publish:trigger)

The key is shown once after generation. Copy it immediately - it cannot be retrieved later.

Optional capabilities

Below the role dropdown, an Optional capabilities section lets you grant additional scopes that are not part of any role's defaults:

• AI Prompt Execution - allows the key to run AI prompts that can create pages, edit content, and modify your site. This is an opt-in capability because the prompt engine has full workspace write access.

Viewer keys cannot receive AI Prompt Execution. Because the viewer role is read-only, granting prompt execution to it would violate that contract. The checkbox is disabled when Viewer is selected.

This scope requires exec() and a working PHP command-line executable on the server. See Common Issues → Prompt execution if you encounter a 503 error.

Keys retain their audience and scopes. Existing REST keys also need an explicit current-account binding. Use Settings → Connections → Connect existing REST keys to an account when prompted. Revocation or a change to the bound account’s role affects future calls and retained retries. See Permissions.

Allowed origins (CORS)

The Allowed Origins textarea controls which domains can make cross-origin requests to the Agent API. Enter one origin per line:

https://yourdomain.com
https://app.zapier.com

Leave as * (default) to allow all origins.

Nginx users: The Agent API requires Authorization header forwarding in your Nginx config. See Nginx Configuration for details. Routing works automatically - no Nginx rewrite rule is required.

A View API schema link in the card header opens the public OpenAPI 3.0 specification (/_studio/api/agent/v1/schema) - no authentication required. Import it into agent frameworks, Postman, or code generators.

For the full endpoint reference, authentication details, and code examples, see Agent API.

Danger zone

Both reset flows require an exact review and owner approval. A confirmation applies to the reviewed state; a changed site requires a new review.

Reset website

Review the website, history and business data selected for removal. The reset can clear managed draft and published website content, conversation and revision history, saved designs and snapshots, form definitions, Action definitions and business records.

The review lists preserved accounts, credentials, settings, provider usage, personal notes, shared Board records, uploaded images/files/fonts, unrelated hosting files and retained recovery records. It also preserves retry and delivery records so an old request cannot become a new send. Check the actual list for this installation before proceeding.

Reset installation

Factory reset prepares an exact inventory and a recovery window before its final irreversible step. Inspect what will be removed, what remains and the final boundary. Keep a full hosting backup; a website snapshot does not back up the database or application.

If interrupted, use the retained status and owner recovery controls. Do not start another reset because a response was lost. See Administration.

Related

• Agent API - REST endpoints for external AI agents and automation tools
• Email Configuration - detailed SMTP setup, presets, and testing
• Updating VoxelSite - full update guide with FTP fallback and troubleshooting
• The AI Engine - how Site Memory and Design Intelligence work
• Requirements - PHP version, extensions, and server setup
• Customizing the Studio - modifying the Studio source code