AI terms in plain English
Last updated November 7, 2026
The words people use about AI, and the ones on this site, explained without the jargon. Where a term works a particular way on Monotonic, we say how.
AI model
A program that has learned patterns from large amounts of text, so it can read a question and write an answer. Chat assistants such as ChatGPT and Claude are built on AI models.
On Monotonic: Each account gets its own model, and it runs only inside Monotonic.
API
Short for application programming interface: a way for one piece of software to talk to another. Developers use APIs to add features from one service to their own apps.
On Monotonic: Your own apps can ask your model questions through Monotonic's API, using an API key.
API key
A long secret code an app uses to prove it may use an API. Keep it private, like a password.
On Monotonic: Each key answers as one audience, and you can end it at any time from Connect.
Audience
Who an AI answers for, and so which of your documents it may draw on.
On Monotonic: An audience is you alone (Me), a team, everyone in your account, or your customers. Everything you add belongs to the audience you're in, and answers draw only on what that audience may see.
Base model
The AI model you start from, before it learns anything about you. Companies release base models after training them on huge amounts of public text.
On Monotonic: Today's base model is Qwen3 4B.
Chat
Asking an AI questions in everyday language and reading its answers, like a conversation.
On Monotonic: Chat in the browser on any device. Each chat is kept until you delete it.
Citation
A pointer to where an answer came from, so you can check it yourself.
On Monotonic: Every answer lists the documents and passages it used.
Cleaning
Checking files for personal details and handling what's found before the files are used.
On Monotonic: You choose for each upload: review what's found yourself, hide it automatically (redact), or keep it as it is. Cleaning happens inside Monotonic, not on your device.
Connector
A link between two services, so information flows from one to the other without you copying it by hand.
On Monotonic: Connectors bring in your documents from Google Drive, Gmail, Microsoft 365, Notion, Amazon S3 and PostgreSQL. Claude and ChatGPT also use the word for outside tools they can call (see MCP).
Embed
Placing something from one service onto your own website, such as a video or a chat window.
On Monotonic: An embedded chat window lets visitors to your website ask your model questions. It answers only from what you've made customer-facing.
Encryption
Scrambling data so only someone with the right key can read it. Encrypted in transit means while it travels over the internet; encrypted at rest means while it's stored.
On Monotonic: Files travel over an encrypted connection and are encrypted under your account's own key as soon as they reach us.
Fine-tuning
Training a base model further on examples of your own work, so it picks up your tone, format and vocabulary. It's best for how a model writes; facts that change are better looked up with retrieval.
On Monotonic: We call it learning, and it's your choice: with learning off, nothing trains.
Frontier model
The largest, most capable AI models, built by companies such as OpenAI, Anthropic and Google. They know a lot about the world, and most people use them through those companies' own apps.
Indexing
Preparing documents so they can be searched quickly, much like the index at the back of a book.
On Monotonic: Each file is indexed after it's cleaned, and from then on answers can draw on it.
Isolation
Keeping each customer's data and model apart from everyone else's, even when they run on the same computers.
On Monotonic: Customers share hardware, never data. Each account has its own keys, search index and model, and nothing of one customer's ever trains or answers for another.
MCP
Short for Model Context Protocol: an open standard that lets AI assistants such as Claude and ChatGPT use outside tools. Claude and ChatGPT call these tools connectors.
On Monotonic: Add your model as a connector and Claude or ChatGPT can ask it questions. Its answers then go to that assistant, under that company's terms.
Memories
Things an AI keeps from earlier conversations, so you don't have to repeat yourself.
On Monotonic: Off unless you turn them on, kept only for you, and you can read, edit or delete each one.
Message
One question you send to an AI.
On Monotonic: Plans include a number of messages a month. Each question you ask counts as one, wherever you ask it.
Model size
An AI model is made of billions of numbers, called parameters, that it learned in training. The B in 4B, 9B or 14B means billions of them. Larger models usually handle long documents and harder reasoning better, but they're slower and cost more to run.
On Monotonic: Every plan runs Qwen3 4B today. Larger models come to Pro (up to 9B) and Business (up to 14B) once they pass our quality tests.
Open-weight model
A model whose learned numbers (its weights) are published, so anyone can run it on their own computers instead of renting it from the company that made it.
On Monotonic: We run open-weight models ourselves, so your questions are never sent to another AI company.
Personal details (PII)
Information that identifies a person: names, email addresses, phone numbers, ID numbers and the like. PII stands for personally identifiable information.
On Monotonic: Found when your files are cleaned. Names, contact details and ID numbers are always removed before training.
Public AI
An AI service anyone can sign up for, run by the company that built it, such as ChatGPT, Claude or Gemini. What you type into one goes to that company and is handled under its terms.
Qwen3
A family of open-weight AI models from Alibaba's Qwen team, released in several sizes.
On Monotonic: Qwen3 4B Instruct is the model every plan runs today.
Redaction
Hiding sensitive details in a document, like blacking out a line on paper.
On Monotonic: If you choose to redact, personal details are removed from the copy that's searched, and the original is deleted.
Retrieval (RAG)
Looking up the passages in your documents that best match a question and handing them to the model with it, so the answer comes from your material and can show its sources. RAG stands for retrieval-augmented generation.
Search only and learning
Two ways a model can use your material. Search only looks things up in your documents for each answer. Learning also trains the model on files you choose, so it picks up how you write.
On Monotonic: Each audience chooses. Either way, facts always come from search.
Training run
One session of training a model on new material.
On Monotonic: Plans include a number of runs a week. A run starts when you ask for one, or by itself once your material has grown enough, unless you turn that off.
Writing examples
Pieces of your own writing that a model learns your style from: replies, reports, posts.
On Monotonic: Added on Style, and used only while learning is on.