⚙️ Config v1.3.3

Restructured interface.prompts and interface.agents from boolean to object structure

• Added use field to control whether users can use prompts/agents (default: true)
• Added share field to control whether users can share prompts/agents (default: false)
• Added public field to enable public sharing of prompts/agents (default: false)
• Boolean values still supported for backward compatibility (treated as use setting)
• See Interface Configuration for details

Added public field to interface.mcpServers configuration

• Controls whether users can share MCP servers publicly (default: false)
• See Interface Configuration for details

Enhanced SSRF protection for actions.allowedDomains and mcpSettings.allowedDomains

• When not configured: SSRF targets are blocked by default, all other domains are allowed
• When configured: Only listed domains are allowed
• Blocked targets include: localhost, private IPs, link-local IPs, .internal/.local TLDs, and common service names
• Supports protocol and port restrictions (e.g., 'https://api.example.com:8443')
• See Actions Configuration and MCP Settings for details

Added guardrailConfig to endpoints.bedrock for AWS Bedrock Guardrails support

• Configure guardrailIdentifier with your guardrail ID
• Set guardrailVersion to specify the version
• Optional trace field for debugging ("enabled", "enabled_full", or "disabled")
• See AWS Bedrock Configuration for details

Added new fields to MCP server configurations

• Added title field for custom display name in the UI
• Added description field for server description
• Added apiKey authentication for SSE/HTTP transports with source, authorization_type, and custom_header options
• Added oauth.skip_code_challenge_check for providers like AWS Cognito that support S256 but don’t advertise it
• See MCP Servers Configuration for details

Added groupIcon property for model specs custom groups

• Specify built-in icon keys (e.g., "openAI", "groq") or custom URLs
• Only the first spec with a groupIcon in each group is used
• See Model Specs Configuration for details

Added xhigh (extra high) reasoning effort option for OpenAI models

• New option for reasoning_effort parameter: "xhigh"
• See Model Specs - Preset Fields for details

Bedrock promptCache parameter now defaults to true for Claude and Nova models

• Automatically enables prompt caching for supported Bedrock models
• Can be explicitly disabled by setting promptCache: false
• See Model Specs - promptCache for details