Guide 28 · Basics
Muse AI Glossary: Every Term, Plainly Explained
Agent, artifact, connector, approval card, Jolly, tokens — the whole Muse vocabulary in plain English.
By Abhit Tomar · Updated November 19, 2026 · 7 min read

Short answer
Muse AI's core vocabulary: an agent acts on your behalf, artifacts are the finished documents and pages it makes, connectors link your apps, approval cards are your review checkpoints, and Jolly is your customizable avatar.
The basics: what you're actually using
Muse AI is Meta's personal AI assistant — a chat-based helper you talk to in plain language, designed to answer questions, draft text, and increasingly take actions on your behalf. It lives in a standalone mobile app, on the web, inside WhatsApp, and through voice conversation. This site is an independent guide to using it; we are not Meta, and nothing here is official.
AI agent is the term for what Muse becomes when it stops just answering and starts doing. An agent can carry out multi-step tasks — checking out through partner stores, looking things up across sources, working across connected apps — with your permission at each sensitive step. The industry uses "agent" loosely, but the practical test is simple: if the AI is acting in the world rather than just chatting, it's behaving as an agent.
Prompt is what you type or say to the AI. It can be a question, an instruction, or a whole document with a request attached. The quality of what comes back depends heavily on the quality of the prompt: specific context, a clear task, and a stated format reliably produce better results than vague one-liners. Prompting is a skill, not a talent, and it improves fast with practice.
Hallucination is when an AI confidently states something that isn't true — a wrong date, a fabricated citation, a detail that sounds right but was never real. It happens because language models predict plausible text, not verified facts. The defense is habit: verify anything important against a second source, especially numbers, dates, and anything you'll act on. This applies to every AI assistant, not just Muse.
System prompt is the hidden instruction set that shapes how the AI behaves — its tone, its rules, what it refuses to do. You don't write it; Meta does. But it explains why the assistant sometimes declines a request or answers in a particular style. When Muse says it can't do something, that's usually the system prompt talking, not a technical limitation of the moment.
How the AI thinks: tokens, memory, and models
Token is the unit AI systems use to measure text. As a rough rule of thumb, one token is about three-quarters of a word in English, so a thousand tokens is roughly 750 words. Everything you send and everything the AI generates is counted in tokens, and Muse's usage limits are denominated in them. You don't need to count tokens yourself, but understanding the unit helps the pricing and the free tier make sense.
Context window is how much text the model can consider at once — its working memory for a conversation. Muse's frontier models support around a million tokens of context, which is roughly an entire bookshelf of text. In practice, a bigger window means you can paste longer documents, hold longer conversations, and have the AI remember earlier parts of a complex task. When an AI "forgets" something from earlier in a chat, the context window is usually why.
Muse Spark was Meta's first model in the current Muse line — a lightweight, low-latency model introduced as the starting point on the path to what Meta calls "personal superintelligence." It powered the early Muse experience: fast answers, quick drafts, everyday questions. Later checkpoints like Muse Spark 1.3 improved reliability across hundreds of internal evaluations.
Muse Wonder and Muse Wonder Full are Meta's first frontier models, released in October 2026 by Meta Superintelligence Labs. They are built specifically for agent work: long, multi-step tasks that require memory and persistence. Wonder is available through Meta's Model API and Muse Code; the larger Wonder Full is currently limited to Muse Code subscribers. Independent benchmarks place them near the top of the field on coding and agentic tasks.
Active Memory is one of the two headline capabilities of the Wonder models. It lets the model retain what matters about you across conversations — your preferences, your ongoing projects, the context you've built up — and pull the relevant details back when a new task needs them. The practical effect is that you stop re-explaining yourself every session. It is the difference between an assistant with a notepad and one with amnesia.
Active Focus is the other headline capability. It keeps the agent working toward a difficult goal instead of giving up: trying alternative approaches when blocked, asking for clarification when something is ambiguous, and checking in before taking important actions. For long tasks — the kind that take dozens of steps — this persistence is what separates a useful agent from a chatbot that stalls at the first obstacle.
Temperature is a technical setting that controls how predictable or creative a model's output is. Low temperature means safe, consistent answers; high temperature means more varied, surprising ones. You won't see this setting in the Muse app — it's a developer and API concept. It matters only to know the term exists, because "turning up the temperature" is sometimes suggested as a fix for boring outputs in tools that expose it.
Access and money: codes, tiers, and meters
Referral code is a short code shared by an existing Muse user that gives a new user bonus tokens or credits when redeemed. Codes are entered inside the official Muse app, usually during signup or in account settings. The value varies by promotion, region, and timing — the in-app redeem screen is the only source of truth for what your account actually gets. Never pay for a code, and never enter one on a third-party website.
Invite code works much like a referral code but is typically framed around joining: early access, a waitlist bypass, or a welcome bonus. In practice the two terms overlap heavily, and most "invite code" offers are referral promotions by another name. The same rules apply: redeem inside the official app, verify the terms on the redeem screen, and treat any headline number as a starting point for verification, not a promise.
Usage meter is the in-app display showing how much of your allowance you've consumed. Muse's plans are metered in tokens per week, and the meter — usually found under Settings — is your early-warning system for approaching a limit. If you find yourself checking it often, that's a sign your usage has outgrown your current tier. Glance at it occasionally even when everything feels fine.
Free tier is Muse's no-monthly-fee plan, with a weekly token allowance that renews indefinitely. It covers everyday questions, planning, drafting, and light agent use for most people. The catch nobody puts in the headline: signup requires a payment card on file and a supported region (currently the US, Canada, Japan, Singapore, Mexico, and Australia). You aren't charged unless you upgrade, but "free" still starts with handing over card details.
Power is Muse's mid paid tier at $20 a month, with a weekly token allowance several times the free tier's. It's aimed at people who regularly hit the free ceiling: freelancers, students in crunch periods, anyone using Muse as a research assistant for a few hours a week. If you only bump into the limit once in a while, it's usually cheaper to wait for the weekly reset.
Maximum is the top tier at $100 a month, with a weekly allowance an order of magnitude above Power's. This is professional-grade: small business owners automating admin, researchers running high-volume sweeps, background tasks humming continuously. Before paying, ask whether the workload is genuinely that heavy or just badly managed — a looping task can burn tokens fast.
Where you meet Muse: surfaces and connections
Voice mode lets you talk to Muse instead of typing, having a spoken conversation the way you'd explain something to a person. It's fastest for brainstorming, thinking aloud through decisions, and anything where speaking produces more natural context than typing. Availability varies by region and device. A practical tip: ask for a spoken summary at the end of a voice session, so a messy thinking process becomes a clean record.
WhatsApp integration puts Muse inside the chat app many people open dozens of times a day. You connect it once, then chat with it like any contact — asking questions, drafting messages, getting quick answers without switching apps. It's the lowest-friction way to use Muse, though the full app offers more: longer tasks, richer formatting, and features that don't fit in a chat bubble.
Connector is an integration that links Muse to an app or service you already use — your calendar, your files, a store, a productivity tool. Connectors extend what the assistant can reach and act on: with a calendar connector, it can check your schedule while planning; with a shopping connector, it can look up real products. Availability varies by region, so check the app for what's currently offered rather than assuming a specific integration exists.
Muse Code is Meta's subscription product for developers, and currently the exclusive home of Muse Wonder Full. It bundles the frontier models with coding-agent features at $50 and $200 per month tiers. If you're not a developer, the name matters mainly as the place where Meta's most capable model currently lives — and as a sign of where the company thinks the sharpest demand for frontier models is.
Using it well: habits worth building
Iteration is the habit of refining an answer with short follow-ups instead of restarting from scratch. "Make it shorter," "rewrite the Tuesday section for a rainy day," "keep everything else" — each follow-up steers the existing answer closer to what you want. Experienced users consistently say their best results came from the third or fourth exchange, never the first. Editing beats restarting, every time.
Context, task, format, constraint is a simple four-part structure for better prompts. Context is who you are and the situation; task is the one thing you want done; format is what the answer should look like; constraint is any rule it must follow, like length or tone. You don't need all four every time, but running through them takes ten seconds and reliably produces better answers than a bare question.
Red-teaming your own prompt means asking "how could this go wrong?" before you send something important. If you're asking Muse to draft a sensitive email, consider what tone could misfire. If you're asking it to plan spending, check the numbers yourself. It's a thirty-second habit that catches most of the mistakes people blame on the AI afterward.
Keep reading
01 · BasicsWhat Are ChatGPT Dots? The Complete Guide to OpenAI's Always-On AgentsOpenAI's Dots keep working after you close ChatGPT. What they are, what they can do, how to set one up, what they cost, and where the limits are — in one place.Read guide
03 · BasicsWhat Is Muse AI? A Clear Beginner's GuideUnderstand what Muse is, what it can do, and where it fits among personal AI agents.Read guide