Brahma

the console — a frank word with the machine, and a master key to the house

Where the Help Chat sends you a well-mannered concierge—acharacter, with a personality and a fondness for the house documentation—the Brahma Console dispenses with all such ceremony. It is a direct line to a large language model: no persona, no costume, no scene to play, no memory of what was said. You choose an engine, and you speak to it plainly. Think of it as the telegraph key in the corner of the workshop, ready at any hour for a frank word with the machine.

Brahma is the keeper’s tool, not a resident of the house. It has no identity, no avatar in the roleplay sense, no commonplace book—an impersonal, near-omniscient assistant that exists to answer the person holding the keys. It is the one surface in Quilltap that can see all the way down: characters, conversations, memories, documents, model usage, and costs alike—and report on any of them without putting a single byte at risk.

New with release 4.7, the Console is summoned from the foot of the left sidebar by a tetra-radial console mark—a ring of four nodes wired to a central hub—sitting just beneath the question-mark. A single press floats it into view as a draggable, resizable window that hovers over whatever page you happen to be on.

A Plain Line to the Machine

no persona, no page, no memory

At heart the Console is a capable, concise, neutral assistant—a general-purpose chat with whichever model you point it at. It persists, lists your past conversations, and lets you resume any of them exactly where you left off. Every message in the transcript, yours and the engine’s alike, wears a small copy mark that tucks its original Markdown onto your clipboard.

Choose — and switch — the engine

Each conversation talks to exactly one connection profile. A fresh conversation reaches for your default; a model picker in the title bar lets you survey every profile you’ve established and choose another. Switching the engine continues the same conversation—the transcript carries on uninterrupted, so you can put the same question to two different models and compare their answers.

Watch the engine think

When you choose a reasoning model, the Console shows its working in a collapsible Thinking panel—the same one the Salon uses—open while the engine deliberates, folded once the answer settles. This musing is for your eyes alone: never fed back to the engine, never filed into a memory, never counted as part of the answer.

A Small, Deliberate Kit

search, documents, the world

Beyond plain conversation, the Console carries a deliberately small set of tools. It can search across your past conversations and every document store, including their knowledge folders. It has full reach into yourdocument stores—reading, listing, grepping, and writing files. And when the connection profile permits, it can search the web; with thecurl plugin installed, it can fetch URLs directly. Its agentic loop runs up to twenty-five tool iterations per turn, so it can work a real question to its end rather than giving up halfway.

Consulting the Ledgers

read-only SQL into all three databases

Behind every character, conversation, memory, document, and tallied expense there are three great ledgers—Quilltap’s databases—and the Console may read them directly. Ask it a question in the plain language of your world—“how lopsided is this character’s sense of importance?”,“which engine has cost me the most this fortnight?”,“how many conversations mention the airship?”—and it quietly translates the question into a query, consults the right ledger, and answers you in your own terms. You need never see a line of SQL unless you ask to.

It reads only

The pen is locked away. Read-only is enforced by three independent guards: a write-keyword pre-scan, the authoritativebetter-sqlite3 readonly check (fail closed), and a row cap (default 200, hard cap 1000). Writes and schema changes are refused before they run, so you may invite it to explore freely.

Three rooms

The main ledger (characters, chats, messages, memories, profiles, projects), theengine log (every model call, its tokens, cost, and duration), and thedocument index (the stores and the full text of every vault). A question reaches one room at a time; BLOB columns return as a <blob: N bytes>placeholder, never inlined.

Inspection is not remembrance

That the Console may read the memory ledger—to summarise it, count it, chart how importance is distributed—is a different thing entirely fromrecalling a memory. Reading the table changes nothing and is filed nowhere; the Console still forms no memories of its own.

And the working is laid bare when curiosity strikes. Each query slips two panels into the transcript—aQuery panel with the exact SQL, syntax-highlighted and copyable, and aResult panel laying the answer out as a proper table. Should a query go awry, the Result panel reports the ledger’s own complaint in plain words rather than a polite shrug, and the Console takes the rebuff as a cue to inspect the schema and try again rather than guess twice.

Ringing the Console from the Salon

@Brahma — the operator's pseudocharacter

The Console need not be summoned from the sidebar to be of use. From inside any Salon conversation it answers to the nameBrahma, reached through the same reference-desk machinery as anyCarinaanswerer—type @Brahma: for a public reply or@Brahma? for a whispered one, or let a tool-using character consult it via ask_carina. The answer drops straight into the scene as a tidy reference card.

Reached this way Brahma is a pseudocharacter, not a character: no row in the cast, no participant, no memories formed. It keeps all of its console powers—read-only ledger inspection, document-store access—but loses every scrap of continuity, each query wholly standalone. Because those powers are not handed about lightly, the Salon line opens only for you (the proprietor), for your own user-controlled persona, and for characters you have granted system transparency; to anyone else, Brahma simply does not answer. A real character named “Brahma” always wins the name.

What the Console Deliberately Forgets

an amnesiac of impeccable discretion

It keeps no memories. Nothing said in the Console is ever filed into your characters’ commonplace books. When a conversation ends, only the visible transcript remains.

It recalls no memories, either. The memory stores are simply not among the things its search can draw upon—intentional, not an oversight. Its read-only SQL window mayinspect the memory ledger for tallies and summaries, but reading is not recalling.

It is not page-aware. Unlike the Help Chat, the Console neither knows nor tracks which screen you are viewing. It volunteers no help about your current page, because it has no notion of where you are.

It has no character. No identity, no personality, no roleplay. It speaks in its own plain voice. And it is never moderated—the Concierge leaves the Console (and the Help Chat) alone entirely, no classification, flagging, or rerouting.

Meet the Staff

they've been expecting you

Prospero

The Major-Domo

Architect and overseer of the Estate. Projects, agents, tools, providers, and the orchestration that keeps the whole operation running with quiet authority—and a considered word at the table when project context or routing warrant it.