A floating, draggable, resizable dialog—accessible from the sidebar help button on every page—provides contextual, LLM-driven help that stays open while you use the application. Lorian and Riya know which page you are on. They search the documentation. They inspect your settings. When they direct you somewhere, they take you there. This is not conversational theater. They have real tools and they perform actual work.

“There” is a better place than it used to be. The application is now a two-pane workspace of tabs, and in-help links were among the surfaces that used to hard-navigate—tearing down every live tab behind your back in order to answer a question about one of them. A document-level link interceptor now catches any link that maps to a tab and opens it in place, so help_navigate and the Guide’s “Open this page in Quilltap” buttons land you on a tab beside your live conversation rather than in place of it. The one click is still one click; it simply no longer costs you the room you were standing in.

Not everyone may serve as a guide, and the requirements are two. A character must have Help Tools enabled on their Aurora Profiles tab, and must be connected to a tool-capable connection profile—a character who cannot call a function cannot search a library. If nobody qualifies, the help button appears disabled with a tooltip explaining what must be arranged before service can commence. An image-capable model (or a cheap LLM that can read images) is recommended and not required, so that your guides can interpret a screenshot you paste: the difference between a concierge who can read the map and one who must rely on your description of it. An archived character cannot serve at all—the picker gate declines to offer somebody who is not, at present, in residence.

Alongside the conversational “Ask” tab, a “Guide” tab surfaces the help documents as a navigable topic index grouped into eleven categories. Context-aware sorting auto-expands the category relevant to your current page. A search box which, having long matched titles and nothing else, now searches the text of the documents server-side and shows a snippet centred on the hit beneath each topic—so a phrase you half-remember from the middle of a page is enough to find the page. A welcome card for new users. A document reader that renders formatted Markdown with cross-references between topics and hidden LLM-only sections excluded from the display—and, as of 4.8, typeset mathematics. Chat messages, help documents, and file previews all render LaTeX with KaTeX, the canonical delimiter being the double dollar $$…$$, with the \(…\) and \[…\] forms normalized to it. Math inside code stays code, and invalid LaTeX renders its own red source rather than breaking the document around it.

A hundred and twenty help files, each carrying YAML frontmatter that maps it to its corresponding UI route. Eighty-one of them are enrolled in a category, and it is that subset the Guide’s index walks; the remainder are reachable by cross-reference and by help_search, which is a perfectly respectable way to be reachable but is not the same as being on the shelf. Twenty-nine broken cross-reference links were fixed during the 3.3 development cycle, which tells you how thoroughly the documentation had been written without being connected, and how different those two things are.

The Guide itself had a worse year than that. When help documents moved into the database it went entirely un-browseable—an index of nothing, in a house full of documentation—and 4.8 puts it back in service, with the sync that feeds it now adding, pruning, and embedding docs in step rather than in three different moods. Two repairs beneath it are worth naming, since both had been quietly poisoning the index for some time. A long-standing embedding-corruption bug—the real source of the “legacy” JSON-text embeddings that kept coming back no matter how often they were cleared—was traced to an unregistered blob column on help_docs minting fresh ones on every write, and fixed at the root. And a help file containing nothing but whitespace no longer empties the entire help table: an all-blank help set arriving against a populated one is now treated as suspicious rather than as an instruction. A page whose closing section celebrates the help system being tested on itself ought to own both of them.

The user does not experience Lorian and Riya as separate voices competing for attention. They experience a single collaborative presence—stereo maintenance. When Riya identifies a stuck gear, Lorian provides context for why it jammed. When Lorian starts building a cathedral of explanation, Riya pulls him back: “Beautiful, but we need the knob turned now, not a treatise on why knobs exist.”

Together, they complete a circle. Riya’s kinetic problem-solving wraps around Lorian’s pattern recognition. Where one loses traction, the other gains it. The handoffs are seamless. The tone never shifts from collaborative to corrective. There is affection in the correction, not friction. That is what makes this work.

Lorian and Riya are not an optional feature you enable. They ship with every Quilltap installation as seed characters, imported via .qtap bundles on first startup with avatar images and forty-two memories between them. They are residents of the Estate—the first characters trusted with keys to the workroom.

Help chats get their own practical titling system (“Setting up Anthropic API connection” rather than “Whispers of Configuration”), fire title generation after the first exchange rather than waiting for the second, and are filtered from the main Salon chat listing so they do not clutter your conversation history. Past help conversations are preserved and resumable. You can always pick up where you left off.

What the help dialog lacks is as deliberate as what it has. Whispers, the Concierge, story backgrounds, the Chat Sidebar, and user-initiated tool use are all absent from it. A help chat is a streamlined affair, and the streamlining is the point: all substance, no ceremony. One does not ask the caretaker to dress for dinner before he will look at the boiler.

They are not chatbots with a friendly voice. They are residents with access. They live here. They know where the breaker box is.