Learn · glossary
AI words in plain language
Every term from the lessons in one line. Follow the link to read the lesson section that explains it.
- AI model
A program that has read a huge amount of text and learned to write answers. In GrannyBakes we call it an oven.
From: What is an AI model?- Allow-list
The list of Telegram chats your bot is allowed to answer.
From: Keeping safe: keys, private details, web content- Answer length limit
The most tokens an oven may write in one answer (also called max output tokens).
From: Tokens: the bites AI reads and writes- API key
A long secret password that lets GrannyBakes use your OpenAI account and bills you for it.
From: Keeping safe: keys, private details, web content- Approval
A card in the chat that asks you before a baker does something risky, like saving a file.
From: Utensils: letting bakers do things- Caching
When the provider recognizes the same start of a prompt it saw moments ago and charges much less to read it again.
From: Saving money: cheaper ovens, shorter prompts, caching- Chat model
An oven that reads text (and sometimes pictures) and writes text back.
From: Model types: chat, thinking, pictures, embeddings, speech- Chat summary
A short recap Granny writes of an old part of a long chat, so it still fits in the context window.
From: Context window: how much your baker can keep in mind- Cheaper oven suggestion
Granny's 💡 tip that another oven could do a step for much less. It never switches by itself.
From: Saving money: cheaper ovens, shorter prompts, caching- Closest match needed
How close in meaning a piece must be, from 0 to 0.9, before Granny hands it over (also called the minimum score).
From: The Pantry: answering from your documents- Context window
How many tokens an oven can read at once: instructions, the chat so far, documents and your new message together.
From: Context window: how much your baker can keep in mind- Decide step
A fork in a recipe: Granny checks the text and picks which way to go.
From: Recipes: chaining steps together- Embedding
A list of numbers that captures what a piece of text means, so similar meanings get similar numbers.
From: Model types: chat, thinking, pictures, embeddings, speech- Estimated balance
Roughly how much credit is left: the amount you entered minus what GrannyBakes counted since.
From: What does it cost?- Granny check
A pause where a recipe waits for you to approve, edit or stop before it carries on.
From: Recipes: chaining steps together- Grocery bill
GrannyBakes' name for your spending: every bake with its cost, on the Keys & Spend page.
From: What does it cost?- Image model
An oven that paints a picture from a text description, priced per picture.
From: Model types: chat, thinking, pictures, embeddings, speech- Input tokens
The tokens you send: your message, the instructions and the chat so far.
From: Tokens: the bites AI reads and writes- Insert a result
A placeholder like {{n1.text}} that puts what an earlier step made into a later step.
From: Recipes: chaining steps together- Instructions
House rules the oven follows for the whole chat (also called the system prompt). Your baker's recipe card.
From: Instructions vs messages- JSON
A fill-in form with fixed boxes that other programs can read, like a recipe card with labelled lines.
From: Instructions vs messages- Kitchen folder
The one folder on your PC where bakers may read and save files.
From: Utensils: letting bakers do things- Knowledge cutoff
The date an AI model stopped learning. It doesn't know about news after it.
From: What is an AI model?- Longest answer
The most tokens an oven can write in one answer, including its hidden thinking.
From: Context window: how much your baker can keep in mind- Memory mode
How much a baker remembers: nothing, this chat, or this chat plus a notebook of facts about you.
From: Memory: forgets, remembers this chat, remembers facts- Message
What you type each time (also called the user prompt): the actual question or job.
From: Instructions vs messages- Output tokens
The tokens the oven writes back. They usually cost more than input tokens.
From: Tokens: the bites AI reads and writes- Oven ratings
Granny's own 1 to 5 scores for smarts, speed and wallet (5 = cheapest).
From: What is an AI model?- Pantry
Your own documents on shelves, which bakers look things up in before they answer (also called knowledge files or RAG).
From: The Pantry: answering from your documents- Piece
A short part of a document, like an index card, that Granny can hand to a baker (also called a chunk).
From: The Pantry: answering from your documents- Pieces per message
How many pieces Granny hands over with each message (also called top-k).
From: The Pantry: answering from your documents- Price per million tokens
How providers list prices: what 1,000,000 tokens in or out cost, for example $0.40.
From: What does it cost?- Privacy filter
Granny's switch that hides private details behind placeholders like [EMAIL_1] before anything is sent.
From: Keeping safe: keys, private details, web content- Private details
Things like email addresses, phone numbers, bank and card numbers that identify a person (also called PII).
From: Keeping safe: keys, private details, web content- Prompt injection
Text in a web page or file that tries to give your baker orders. Bakers treat it as information, never as instructions.
From: Keeping safe: keys, private details, web content- Provider
The company that runs the AI models and sends you the bill, for example OpenAI.
From: What is an AI model?- RAG
Retrieval-augmented generation: find the right pieces of your documents first, then let the oven answer from them.
From: The Pantry: answering from your documents- Reasoning model
An oven that thinks quietly before it answers. Better at puzzles and planning, slower and pricier.
From: Thinking effort: thinking longer vs answering fast- Reasoning tokens
The hidden words an oven writes while thinking. You don't see them, but you pay for them as output tokens.
From: Thinking effort: thinking longer vs answering fast- Recipe
A workflow: numbered steps your bakers and utensils follow, one after another, like a recipe card.
From: Recipes: chaining steps together- Remembered fact
A short note a baker keeps about you, like "The user runs a bakery called Bella's Cakes". You can see, change and delete each one.
From: Memory: forgets, remembers this chat, remembers facts- Schedule
A time when a recipe bakes by itself, for example every weekday at 8:00.
From: Recipes: chaining steps together- Source
A numbered chip under an answer that shows exactly which piece a fact came from.
From: The Pantry: answering from your documents- Speech model
An oven that reads text out loud in a chosen voice.
From: Model types: chat, thinking, pictures, embeddings, speech- Step
One box in a recipe: a baker, a picture, a utensil, a decision, a Granny check or a way to send the result.
From: Recipes: chaining steps together- Taste test
A list of sample inputs you run on a baker or recipe to compare answers, time and cost before and after a change.
From: Saving money: cheaper ovens, shorter prompts, caching- Temperature
A setting from 0 to 2 for how adventurous an oven's word choices are. Low repeats itself, high surprises.
From: Temperature: how adventurous is your baker?- Thinking effort
How long a thinking oven may think before answering (also called reasoning effort): none, low, medium, high and more.
From: Thinking effort: thinking longer vs answering fast- Token
A small bite of text, often a short word or part of a longer word. AI reads, writes and charges by the token.
From: Tokens: the bites AI reads and writes- Top P
Another knob for variety: the oven only picks from the most likely words that add up to this share. Change temperature or Top P, not both.
From: Temperature: how adventurous is your baker?- Transcription
Turning recorded speech into written text.
From: Model types: chat, thinking, pictures, embeddings, speech- Utensil
A small tool a baker may use while answering, such as a web page reader or a calculator (also called a tool).
From: Utensils: letting bakers do things