RELEASE LOG
WilBot Changelog
Track shipped features, fixes, and reliability improvements. External-provider features depend on provider availability and configured quotas.
v4.2Latest release
14 October 2026 · since 3 October, 01:42 BSTA bigger home on the web, stronger tools everywhere
- A clearer WilBot homepage brings the two products together: start a private AI workspace or explore the Discord bot. Type on the homepage and your prompt is carried into WilBot Chat and sent there, so the conversation starts straight away.
- WilBot Chat is a full web workspace, not just a prompt box. Start without an account, keep guest conversations with your browser, then sign in to sync them. Search, rename, pin, archive, branch or delete conversations; responses stream as they arrive, with Markdown and code formatting, stop and retry controls, and a model selector that reports the provider actually used.
- Choose a focused workspace for everyday questions, coding, research, writing, study, support or shopping. Research can use live sources; image understanding and image creation are available through their dedicated tools when a compatible provider is configured. Web chats remain separate from Discord unless you choose to sync a conversation.
- Web Chat provider routing now prioritises configured OpenRouter, Gemini and Groq routes. Paid OpenAI use requires explicit opt-in, and available models and provider labels reflect the configured routes. Provider-side free quotas and availability can still change.
- Short follow-ups and corrections keep more relevant context, simple requests can use a lighter route, and current-information research is more likely to search, stay on the active subject and filter unrelated results. Ordinary replies use bounded provider attempts and move on from slow or unavailable routes.
- Image conversations better distinguish an edit request from a question about an image. Clear edits reach the image-edit route; questions remain on image understanding. Default canvas choices better fit gameplay, face-cam, portrait and story requests, and successful image cards identify the provider/model when available.
- Server teams can opt into AI assistance for eligible unclaimed support tickets, pause it per ticket and request private staff summaries. The dashboard assistant can prepare a bounded configuration proposal for an authorised administrator to review before applying it.
- Bedtime stories now have saved server preferences for length, mood, character theme and narration style. WilBot can queue and control its own generated music in voice, and supported live leaderboard questions can use current ranking data.
- Ticket and interview actions check staff access and current workflow state. Interview claims persist, duplicate message work is less likely across workers, quote delivery gives clearer errors and can retry in the source channel, and moderation and media paths received focused safety fixes.
- Ask what WilBot can do for an overview based on commands registered in the deployment. `/help search` and natural-language help find member-facing commands and guides, with relevant excerpts grounding follow-up answers; private operational notes stay excluded.
- The server dashboard now reads and saves the actual quiet-chat setting, reports failed saves clearly and works better on narrow phone screens. Conversation starters wait for channel activity to settle and never mention members, roles or everyone. Existing server settings remain in place.
- Eligible free Discord users have a small, bounded OpenAI text allowance before WilBot falls back to an eligible no-cost route. Request caps are not a provider spending limit. WilBot also improves provider cooldowns, status checks, reconnect reporting and command cleanup; it does not promise that external services or free quotas are always available.
v4.1.4Previous update
12 October 2026Activity-aware quiet-chat prompts
- Conversation starters now wait until the configured channel has been quiet for the chosen period, instead of posting blindly on a timer.
- WilBot checks timestamps only, skips if it cannot read channel history, and never mentions members, roles, @here, or @everyone.
- Server admins can set the quiet threshold with /settings engagement; the dashboard now reads and saves the real persisted settings and reports failures accurately.
v4.1.3Previous update
10 October 2026More natural image edits and clearer help
- Clear edit requests such as ‘make this guy eat tacos’ now route attached or referenced images into the AI image editor rather than image-understanding chat.
- Question-style prompts about an image remain on the understanding path; existing AI provider, safety, access, and Creator Unit rules are unchanged.
- Unspecified gameplay/face-cam images use a widescreen canvas; portrait/story requests use a vertical canvas, while ordinary image prompts remain square by default.
v4.1.2Previous update
10 October 2026Help that explains WilBot and finds the right feature
- Ask WilBot ‘what can you do?’ for an overview generated from the commands actually registered in the running deployment.
- Search commands and member-facing feature guides with `/help search` or natural-language questions, then ask follow-ups grounded in the matching documentation.
- AI-assisted help uses the live command catalogue and relevant guide excerpts; private operational and incident-response notes stay private.
v4.1.1Previous update
3–6 October 2026 · from 01:42 BSTClearer conversations, better support, and more dependable tools
- Image-model questions now read the live eligible image route directly, so a text-AI outage will not block the answer. Direct insults get a short, calm acknowledgment, and the fuse/Fuze Siege pun is resolved from the immediate conversation instead of guessed.
- Neutral youth image requests now use positive age-appropriate framing, avoiding safety-negation wording that can trigger upstream filters. On an upstream refusal, only an already-eligible no-cost fallback may be tried; there is no billable escalation, and harmful requests remain blocked. Provider policies may still reject benign youth depictions.
- Questions about image-generation models now use the current requester-eligible provider route, and generated-image cards identify the provider/model that actually returned the image.
- Questions about how a generated image was made now stay in image conversation instead of accidentally triggering a new variation; edits require a clear change instruction.
- Quote cards now respect thread-specific send permissions, retry in the source channel when Discord explicitly denies the configured archive, and show clearer upload errors instead of incorrectly blaming permissions for every failure.
- Fast/simple chat routing now learns from recent observed provider latency instead of relying only on static speed estimates. Existing free/paid eligibility and circuit breakers still apply; provider speed varies, so sub-second replies are not guaranteed.
- AI routes that return HTTP 402 now enter the existing quota cooldown instead of being retried on every chat. No provider credentials or billing settings are changed.
- Harmless questions about furry or femboy identity now get a respectful answer that avoids guessing from appearance, rather than an unrelated crisis warning. Assistant answer controls acknowledge clicks and provide a fallback notice if Discord cannot refresh them.
- Ordinary AI replies now use bounded provider attempts and a total route budget, then fail over to another eligible provider rather than retrying the same slow route. Deep reasoning and image/specialist tasks keep larger or existing timeouts; this reduces avoidable waits but is not an instant-response guarantee.
- AI ticket help is now available as an explicit server opt-in with per-ticket pause/resume and private staff summaries. It assists only with eligible unclaimed support tickets; staff claims, sensitive topics, and requests for a person take precedence.
- Interview claims are persistent and protected against competing claims. Waiting reminders are routed to staff operations instead of repeatedly filling the applicant-facing interview conversation.
- Discord message handling now uses persistent deduplication across workers, reducing duplicate replies and recovering abandoned work after a worker crash.
- Generated-music tracks can be queued and controlled in voice with queue, skip, pause, resume, stop, and leave actions. WilBot plays its generated tracks; it does not scrape streaming services.
- Bedtime stories now have saved server choices for length, mood, character theme, and narration style. Optional local narration is documented for deployments that install the required model and runtime.
- Short follow-ups and corrections retain more of the active conversation context, and straightforward requests can take a lighter response path without unnecessary research.
- Current-information research keeps follow-ups tied to the active subject and filters unrelated material more carefully before answering. Supported live leaderboard questions can use current ranking data.
- Free AI requests use no-cost/local routes when the small OpenAI allowance is exhausted or disabled. No other paid providers are silently opened for free users.
- The dashboard assistant can prepare bounded configuration proposals for authorized staff to review before applying.
- Ticket and interview controls check staff authorization and current workflow state instead of trusting channel visibility or an old button.
- When a Streamer role is missing, WilBot now gives practical account-link and role-hierarchy troubleshooting rather than an unrelated generic answer.
- Supervisor health reporting refreshes during healthy operation and reconnect recovery; startup command cleanup removes redundant legacy aliases while keeping active commands.
- Status API checks share one timeout across HEAD and fallback GET requests, preventing slow providers from doubling dashboard refresh latency.
- Private web chat now routes through configured OpenRouter, Gemini, or Groq providers first; paid OpenAI is opt-in, and the UI reports the provider actually used.
- Free Discord users now receive a small OpenAI text allowance (two attempts per user per UTC day by default), protected by shared deployment caps of 20/day and 300/month. It is a single OpenAI attempt, capped at 650 output tokens and 12 seconds; an OpenAI failure falls back to the no-cost route, and failed paid attempts count against the limit. These request caps do not replace an OpenAI project dollar budget.
- Thermal-throttling replies no longer suggest WilBot can read a user's live GPU sensors. It labels general temperatures as examples and asks for the user's actual model and sensor reading; this correction is immediate and provider-independent.
- Moderation and media workflows received focused false-positive, safety, and provider-routing regression fixes.
v4.1.0Previous release
September 2026More personal. More capable. More connected.
- WilBot now works properly in Discord DMs with personal context, memory controls and no canned server greetings.
- Multi-image conversations, safer image search, AI editing, improved video/music routing and better TikTok/media handling.
- Stronger moderation review, nickname audit history, welcome fallbacks and privacy-safe beta diagnostics.
- Server Apps and WilScript now support draft, validate, simulate, install confirmation, sandboxed run and admin rollback.
- New activity, message, channel, member and growth charts, plus expanded Rainbow Six Siege guidance.
- A site-wide redesign, dark mode, private web chat, server-side AI routing and optional Discord continuity.
- Tier-aware quotas, clearer upgrade messages, rotating presence and improved larger-server reliability.
v4.0.0Major Release
August 25, 2026WilBot v4: The Complete Architectural Rebuild
- Rebuilt Public Web Platform & Unified Dashboard on https://wilbot.link
- Deterministic Warning Escalation Engine with configurable points and time windows
- Contextual AI Moderation with four configurable enforcement modes
- Natural Language Automation Engine (WHEN / IF / THEN block compiler)
- Permission-Aware Message Intelligence with channel exclusion and retention controls
- Config Doctor Diagnostic Suite checking role hierarchies and deleted channel links
- Staff Interview Queue with quiet hours and customizable snooze to eliminate spam
- Level Role Discord Synchronizer with dry-run member qualification previews
- LTX Generative Video Studio supporting T2V and I2V modes with daily free quotas
v2.0.6Patch Release
August 15, 2026Voice AI & Anti-Spam Enhancements
- Added RMS Amplitude Voice Activity Detection (VAD) with energy thresholding
- Fixed global STT rate-limiting with exponential backoff on 429 errors
- Optimized supervisor process lifecycles and background task backoff