AI Support Assistant

• Fixed the missing Create Article button for authorized administrators.
• Added optional BookStack knowledge base search in selected books, with source links and safe failure handling.
• Added optional Autopilot replies to guest-created tickets. Disabled by default; staff-created guest tickets remain excluded.
• Added purchased product add-ons to client context for replies, Custom Instructions and Autopilot.
• Added Arabic and right-to-left display in drafts, previews, saved replies and emails; technical text stays left-to-right.
• Fixed incomplete service and domain lists, with exact domain matching and completeness warnings in Sources.
• Added customer images, PDFs and text attachments as context, with per-file status and provider compatibility checks.
• Autopilot skips merged or stale requests and refuses incomplete attachment, named-domain or BookStack context.
• Hardened Autopilot delivery with locked final checks and durable delivery records to prevent unsafe retries.
• Cached verified licenses and preserved previously verified access during temporary licensing outages; confirmed rejections remain enforced.
• Added Anthropic prompt caching. Savings depend on prompt length and repeat requests.
• Clarified licensing, provider and cut-off response errors; shortened connection and language checks while retaining reply-generation timeouts.

• Added GPT-6.1 Sol with mandatory reasoning, saved None mapped to Low, and at least 2048 tokens for Low-effort utilities
• Added Claude Sonnet 5.5 with Low-effort, text-only utilities and Medium-effort adaptive main requests, keeping existing token limits
• Applied both new model contracts to the effective per-request model without changing defaults or older selections
• Added Claude Opus 5.5 with low-effort, 2048-token utility requests and medium-effort main requests, without remapping Opus 5
• Added GPT-6 Sol and GPT-6 Luna to the model selector without changing the default or older selections
• Enabled workload-aware reasoning for both models, including saved None, on the direct OpenAI Chat Completions API
• Kept Astra's forced-Low utilities and minimum token budget; Sol and Luna retain the smaller utility limits
• Grouped OpenAI and Claude models by generation with shorter labels, keeping existing options and saved selections
• Added a description below each model selector that updates when the selection changes, including Fable's 30-day retention requirement
• Removed the Recommended models banner; custom endpoint guidance remains beside the endpoint field
• Preserved exact automatic reply text and sent Markdown through WHMCS AddTicketReply
• Prevented stale AI results from replacing staff drafts; changed drafts require confirmation
• Used submitted connection-test endpoint and model while keeping saved keys masked
• Encrypted provider keys with WHMCS native encryption and made settings/customization writes atomic
• Added a per-customer-request Autopilot attempt ledger, writer advisory lock, pre-send recheck, completion checks and an Open staff-follow-up status option
• Aligned automatic account context with its enabled setting; private notes and other customers' tickets remain excluded

• Fixed KB phrase indexing and empty search terms that could return unrelated articles
• Aligned retrieval and reply-language selection on the latest client message, with stable ordering for replies sharing a timestamp
• Kept enabled support context and bounded conversation history in Custom Instructions replies
• Added shared guidance to use applicable KB evidence, avoid unsupported claims, and keep internal notes private
• Removed canned acknowledgement replies that mistook short problem reports for thanks
• Prioritized recent conversation context while retaining the latest client message and a recent staff reply within existing history limits
• Preserved KB line, list, and table boundaries, technical indentation, and literal configuration values in provider request context
• Added Norwegian (no), Bokmål (nb), and Nynorsk (nn) detection, reply greetings and signoffs, and KB generation language support for both providers
• Mapped all three Norwegian codes to WHMCS norwegian KB metadata

• Added GPT-6 Astra and Claude Fable 5.1 to the model selectors without changing defaults or remapping existing selections
• Extended workload-aware reasoning controls to Astra, including safe mapping of saved None to Low
• Updated Astra request parameters and model-specific timeouts, and increased Astra and Fable 5.1 language-detection and connection-test budgets for mandatory reasoning
• Documented direct API credentials, separate OpenAI API billing, and Fable 5.1 retention requirements

• Added an Improve Writing reply enhancement that improves clarity and flow while preserving the original meaning and tone
• Simplified the purple enhancement menu by grouping Professional, Friendly, Empathetic, and Technical under Change Tone and renaming Custom Prompt to Custom Instructions
• Added a generated reply source summary for matched KB articles, products, and past-ticket context
• Added centralized administrator and CSRF validation for module POST actions
• Removed browser retrieval and display of stored provider API keys
• Restricted custom API endpoints to HTTP or HTTPS URLs without embedded credentials, fragments, or redirects
• Excluded private admin notes and other customers' past-ticket excerpts from Autopilot context

• Added OpenAI GPT-5.6 Sol, GPT-5.6 Terra, and GPT-5.6 Luna support
• Added configurable GPT-5.6 reasoning effort controls with workload-aware defaults
• Added Anthropic Claude Fable 5, Claude Opus 5, and Claude Sonnet 5 support
• Preserved per-request OpenAI model, token-limit, and temperature settings
• Sanitized generated KB and triage HTML before previewing or storing it
• Backfilled all newer settings columns when upgrading an older installation
• Raised the minimum supported PHP version from 7.4 to 8.1
• Removed a no-op global Exception import that emitted a PHP warning

• Added custom OpenAI-compatible API support: connect any provider with a base URL and model name (Z.AI, DeepSeek, OpenRouter, Groq, Mistral, vLLM, and more)
• Added common endpoint reference table in settings for easy configuration
• GPT model dropdown auto-hides when a custom endpoint is configured
• Fixed connection test failing on providers that require at least one user message (e.g. Z.AI/GLM)

• Fixed AI responses using Markdown emphasis that failed to render in WHMCS ticket replies and email notifications; Markdown is now converted to HTML for both OpenAI and Anthropic providers
• Fixed "Generate AI Response" replying with mismatched context from unrelated tickets, and cases where the current ticket's message was omitted from the prompt entirely (causing the AI to ask the customer to resend their request)

• Refactored both AI provider classes (OpenAI.php, Anthropic.php): extracted 11 shared methods into a ProviderUtilities trait, eliminating ~1000 lines of duplication
• Fixed temperature logic in OpenAI provider that was always setting temperature for GPT-4o models
• Fixed OpenAI searchPreviousTickets missing the include_tickets settings guard
• Fixed OpenAI constructor defaulting to the deprecated gpt-3.5-turbo model
• Fixed PHP 8 "Constant already defined" notices from define() calls inside sanitizeUtf8
• Removed dead gpt-3.5-turbo and gpt-4-turbo model normalization branches
• Cached license validation per-request (previously hit the licensing server 2-3 times per AJAX call)
• Fixed KB article creation using string "on" instead of integer 1 for the private column
• Removed dead TicketOpen and TicketUserReply hooks that only logged without taking action
• Replaced mt_rand with random_bytes in the license check token
• Removed unused variables in Anthropic searchProducts
• Improved KB article retrieval: removed hasTechnicalIndicators bias (+10 per phrase for any technical-looking content), added distinct term coverage scoring that rewards articles matching multiple query words

• Added Claude Opus 4.8 support (Anthropic's newest and most intelligent model)
• Added GPT-5.5 support (OpenAI's newest and most advanced model)
• Removed superseded models from the selector: GPT-4o, GPT-4o Mini, GPT-5.1, and GPT-5.2 (OpenAI) and Claude Opus 4.6 (Anthropic); previously saved selections are automatically remapped to the closest current model

• Fixed intermittent cross-ticket context bleed where "Generate AI Response" could reply using a previously closed ticket
• Similar-ticket retrieval now requires 2+ keyword overlaps, excludes the current ticket, and restricts to the same department
• Similar-ticket block in the AI prompt is now explicitly fenced as reference-only with hard "do not respond to these" instructions to both Anthropic and OpenAI

• Added Claude Opus 4.7 support (Anthropic's newest and most intelligent model)
• Enhancement Tool now respects the admin-configured Custom Greeting and Custom Signature; tone rewrites no longer substitute "Dear..." / "Best regards..." over the configured verbiage
• Reworked the enhancement dropdown: added Shorten, Expand, Empathetic, and Translate (prompts for target language); renamed Quick Fix to Fix Grammar; removed Direct, Polite, and Simple

• Added GPT-5.4 support (OpenAI's newest and most advanced model)
• Replaced Claude Sonnet 4.5 with Claude Sonnet 4.6 (Anthropic's latest balanced model)
• Fixed Anthropic language detection bug (undefined property reference)

• Added Claude Opus 4.6 support (Anthropic's newest and most intelligent model)
• Added ticket custom fields support: AI now sees custom field values submitted with tickets

• Added GPT-5.2 support (OpenAI's newest and most advanced model)
• Fixed bug where similar past tickets were not included in AI context

• Added client context feature: AI knows client's services, domains, and invoices for personalized responses
• Replaced GPT-5 with GPT-5.1 (latest OpenAI model)
• Replaced Claude Opus 4.1 with Claude Opus 4.5
• Fixed timeout issues with GPT-5 models (increased to 180s)

• Include private admin notes (per ticket) as additional context (configurable)
• Added Claude Sonnet 4.5 and Haiku 4.5 support
• AI enhancement tools available in new ticket editor
• Removed older Anthropic models (Sonnet 4, 3.7 Sonnet, 3.5 Sonnet, 3.5 Haiku)

• Added GPT-5 and GPT-5 Mini support with proper API compatibility
• Added Claude Opus 4.1 support
• Admin signatures now auto-apply when custom signature is disabled
• Updated model selection with GPT-4.1 Mini and GPT-4.1 Nano
• Streamlined available models list
• Improved error handling

• Ticket triage analysis with troubleshooting suggestions
• KB article generation from ticket conversations
• Multi-style reply enhancement (Professional, Friendly, Direct, Technical, Quick Fix)
• Fixed KB search for non-Latin languages (Cyrillic, Japanese, etc.)
• Language-specific stop word filtering
• Autopilot ticket status control (keep Open for manual review)
• Consolidated toolbar buttons with dropdown menus
• Improved OpenAI pseudo-HTML to real HTML conversion

• Implemented customizable greeting system
• Added enhance-only mode for refining existing replies
• Added o3 models support (OpenAI's most powerful reasoning models)
• Improved hooks.php implementation with better button injection and diagnostic logging
• Enhanced jQuery compatibility and timing issue resolution

• Added Japanese language support
• Improved language detection

• Added Claude Opus 4 and Claude Sonnet 4 models + reasoning variants
• Removed deprecated Claude 3 models
• Enhanced custom prompt feature: now detects and properly handles existing text in editor
  • Shows preview of existing text when using custom prompts
  • Automatically cleans greetings/signatures to prevent duplication
  • Allows seamless modification of AI-generated responses
• Improved AI suggestion with better context handling for longer conversations

• Improved KB search algorithm
• Added technical query detection
• Added custom prompt support
• Removed o1 models
• Security improvements

• Autopilot Enhancements:

  • Department targeting: Autopilot can now be restricted to specific departments via the settings UI (no database changes required)
  • Selective automation: Autopilot logic (scheduled and real-time) only replies to tickets in selected departments
  • Staff ticket safety: Autopilot will not auto-reply to Admin/Staff-initiated tickets

• OpenAI Model Updates:

  • New model additions: GPT-4.1 (most advanced) and o4-mini (ultra-fast, cost-effective) now selectable

• Autopilot Mode:

  • Full automation with customizable settings: Configurable delay for responses, multiple scope options (initial only, limited exchanges, unlimited)
  • Intelligent conversation handling: Complete context awareness, professional formatting with proper spacing and URL handling

• OpenAI Model Updates:

  • New model addition: GPT-4o Mini for fast, economical responses
  • Improved organization: Models categorized into Recommended, Reasoning, and Legacy groups
  • Enhanced usability: Clear model descriptions and simplified naming conventions

• Response Formatting Improvements:

  • Paragraph-based processing: Robust solution ensuring consistent response formatting
  • Universal language support: Language-independent greeting and signature detection

• Language Processing:

• Added support for 11 new languages: Finnish, Maltese, Latvian, Lithuanian, Estonian Slovenian, Irish, Greek, Croatian, Slovak, Bulgarian

• New Features: AI Assistant name and custom signature.

• AI Model Support:
  • **Added Sonnet 3.7 & Sonnet 3.7 with Reasoning models
  • **Improved system prompt for the Sonnet 3.7 reasoning model
  • **Minor bug fixes

• AI Model Support:

  • OpenAI Integration: Added support for new OpenAI models including o1 series
  • Model Consistency: Improved response consistency across different AI models

• Language Processing:

  • Detection System: Enhanced core language detection capabilities
  • Pattern Recognition: Added robust pattern-based language detection
  • Multilingual Support: Updated system prompts for improved multilingual responses

2022/12/01: Initial release.