The control room: settings, plans, and sharing

You're mid-conversation with the agent and a dialog stops you: "Usage limit reached". Or it's something smaller. The body font feels cramped, you want the interface in Korean, or a colleague asks for a copy of your workspace, graph and all. All of these end in the same place: one tabbed screen called Settings, plus a single Share dialog in the sidebar.

This lesson is the tour: from the nine settings categories to the two kinds of credits to handing a whole workspace to a colleague.

One tabbed screen

Three ways in. Press ⌘, (Ctrl+, on Windows). Or open the global search palette and type "Settings"; each category has its own palette entry, so you can jump straight to one. On the welcome screen, the gear button labeled "Open settings" is the way in.

It opens as a regular tab titled "Settings", and it's a singleton: reopen it and the existing tab comes back into focus instead of stacking duplicates. While that tab is active, the left sidebar becomes a nine-item settings menu.

Make it yours: appearance and language

The Appearance tab is small but packed. "Theme mode" is a three-way switch: System, Light, or Dark. On macOS, a "Liquid Glass" toggle turns the frosted, see-through look of the window itself on or off.

Under "Typography" sit four controls:

• "Font family": System, Geist, Helvetica, Serif, Georgia, or Monospace, each drawn in its own face.
• "Font size": 12–24 px, default 16.
• "Line height": 1.0–2.5, default 1.7.
• "Readable width": 30–80 rem, default 46.

Hover or focus any of them and a live preview tooltip pops up: a fully rendered sample document with prose, quotes, tables, math, and code. You see every setting applied before you commit. The "Reset" button in the section header lights up only when something differs from the defaults.

The General tab holds "Interface language", which offers "Follow system" plus English, 한국어, Español, Français, Deutsch, 日本語, and 中文. Next to it are "Check for updates" and an "Automatic updates" switch.

Two actions here are destructive, and both say how far they reach. "Clear history" deletes saved AI chats from this device, and "Reset all app data" wipes chats, recents, settings, and indexes. Both dialogs are clear about one thing: your Markdown files on disk are never touched.

What the AI uses, and what it may do

The AI tab opens with two pickers that answer different questions. "Default AI model" sets which model the assistant reaches for at the start of each new run, and it's the same choice as the model picker in the chat composer. The catalog lists frontier models from the leading AI labs, each shown as "Label · Provider", with a strong default already picked.

"Extraction model" is separate: a two-way preset for the AI that turns your notes into the knowledge graph. You pick "Cost-first" (the default) or "Quality-first", which reads every note more deeply and pulls out more accurate relationships.

• "Automatic knowledge-graph extraction" reads and links your notes in the background every time you save. Its parallel speed self-tunes to your usage limits.
• "Surface notes about the same idea" suggests related notes by meaning, even when they share no words.
• "Capture your decisions and questions" pulls out statements verbatim, so "what did I decide about pricing?" stays answerable later.
• "Add suggested connections automatically" applies suggested typed links on its own. Merges always wait for your OK.
• "Rebuild knowledge graph" gives you a "Rebuild" action that re-extracts every note with the latest logic, in the background.
• "Code intelligence" and "Cloud semantic code search" build a local map of your source code for the agent, with an optional cloud embedding step on top. You're asked to confirm once before any code leaves your machine, and turning it off keeps everything on-device.

The Permissions tab separates what the AI knows from what it may do. "Default permission mode" sets how new chats start, using the same four modes as the composer badge, and each chat can still change its own mode from there. The picker keeps its warning about "Full access" in plain view: use it only in workspaces you trust.

Below it are two capability switches. "Let the AI run commands" covers sandboxed shell execution: read-only commands run on their own, risky ones ask first. "Let the AI use your computer" covers the screen plus mouse and keyboard: looking runs on its own, while clicking and typing ask once per task.

Once you've clicked "always allow" on an approval card, a "Permission rules" section appears with every saved rule. Each rule has its own "Remove" button, and "Clear all rules" makes the agent start asking again everywhere.

Plans, meters, and the two kinds of credits

Everything about billing lives in the Account tab. Signed out, it offers "Continue with Google" or an email one-time code: choose "Email me a code" and a 6-digit code arrives that verifies itself the instant you finish typing it. Signed in, you get your account card, a "My plan" section, and the usage meters.

The four plans (Free, Plus, Pro, and Ultra) differ only in usage limits. There's no feature gating, so a Free account sees the same app as an Ultra one, just with a smaller allowance.

Your allowance is measured in credits. A credit is a token weighted by each model's cost, so heavier models cost more per token.

Every plan carries two rolling windows, shown as meters under "AI usage": a "5-hour limit" and a "Weekly limit", each with its own reset time and a used-versus-cap count. Two things to know:

• The bars fill with what you've used, so a full bar means you're out.
• Because the windows roll, usage frees up bit by bit as older activity ages out, and you can often keep going before the reset time shown.

The meters refresh every 20 seconds while the tab is open. Below them, "Usage credits" is your overage wallet: "Keep working past your plan's limit by drawing on prepaid credit." Flip "Use credits past the limit" on, check your "Current balance", and top up with "Buy credits".

Presets of $50, $250, or $1,000 carry 10, 20, and 30% discounts, and a custom amount runs $5–$5,000. A "Monthly spending limit" caps what the wallet can spend each month, and "Auto-recharge" refills it from your card whenever it dips below a threshold you set.

Checkout, plan changes, the customer portal, and invoices all run through Paddle in your system browser. Paid-to-paid changes happen in place, and the confirm dialog shows the exact prorated charge or credit before you commit. To step down to Free, use "Downgrade in Manage subscription" in the plan picker; it hands off to the Paddle portal.

Plan credits

Your included allowance. Usage is counted as tokens weighted by each model's cost, and heavier models cost more per token.

Rolling window

The "5-hour limit" and "Weekly limit" free up continuously as older activity ages out. There's no calendar-day reset.

Usage credits

A prepaid USD wallet for working past a plan cap. Spent only after a cap is hit and "Use credits past the limit" is on.

Extraction model

A saved cost-vs-quality preset for the model that builds your knowledge graph. It's separate from the chat model.

Share bundle

A portable .zip of a workspace. Content files are always included; graph layers like graph.nq are opt-in.

Take your workspace with you

The last stop is outside Settings: the "Share workspace" button in the Workspace sidebar. It packs your workspace into a portable bundle the recipient opens on their own machine. Your content files always come along; everything else is an opt-in checkbox, all selected by default:

• Knowledge graph ships "Entities, synonyms, relationships and topics (graph.nq)." The recipient gets your whole graph with no re-extraction and no AI cost.
• Topic map, Domain vocabulary, Relationship proposals, Merge decisions carry your curation work along, including the same-versus-distinct calls that keep split merges split.
• Agent memory is the agent's persistent notes about this workspace.
• Semantic search embeddings make search work instantly on import, but this is the largest section. Leave it off to shrink the bundle, since it rebuilds on import anyway.
• Code intelligence shows up only when the workspace actually has a code graph.

"Save bundle…" writes a local file named <name>.consilience-share.zip with no size limit. "Create link" instead uploads the bundle to a hosted, expiring download link, capped at 5 GB. The dialog is clear: anyone with the link can download it until the expiry date, and "Stop sharing this link" revokes it early.

Every active link also shows up in Settings → Account under "Shared links", each row with a red "Stop sharing" button. On the receiving end: welcome screen → "+ New workspace" → "Import shared workspace", then pick the .zip and a destination folder, and the workspace opens with its graph and curation restored.

That's the control room: one singleton tab, never more than ⌘, away, plus a Share dialog that turns a workspace into something you can hand to a colleague. With the dials set, the next lesson opens your workspace to the outside world, where live data keeps flowing into the graph on its own.