Key concepts

This page explains the objects you work with in Annota AI and how they relate. Read it once before you start, then use the Glossary for quick definitions.

Organization and people

An organization holds everything else: people, datasets, knowledge bases, models, skills, plan, and credits. You can belong to several organizations and switch between them from the user menu at the bottom of the sidebar. Only you and the people you invite see an organization’s name.

Each person in an organization is a member with one role:

Role	What they can do
Owner	Everything an admin can do, plus deleting the organization and transferring ownership.
Admin	Manages members, groups, billing, and settings, and moderates every dataset, model, and skill.
Member	Works on their own datasets, models, and skills, and on those shared with them.

A group is a named set of members. Owners and admins create groups so you can share items with several members at once. See Members and invitations and Groups.

Private until shared

Datasets, knowledge bases, models, and skills are private to the person who creates them, their owner. The owner can share an item with members, with groups, or with the whole organization, with one of two access levels: Can view or Can edit. People who can edit an item can also share it with other members and groups. A knowledge base follows the access of the document dataset it indexes. Brain conversations are always private to you.

See Sharing and access.

Annota Studio

Datasets

A dataset is a collection of files of one type. You choose the type when you create it:

Dataset type	What it holds	What you do with it
Images	Photos and other images.	Annotate objects with bounding boxes and polygons.
Documents	PDFs, scans, Office and text files, and audio recordings.	Extract text with OCR or transcription, then review it.

Audio and video datasets are shown in the app as coming soon. See Datasets.

Assets

Each file in a dataset is an asset. Every asset has a review status: To annotate, Pending review, Approved, or Rejected. Review statuses let you track progress and decide which assets go into training or into a knowledge base. See Review assets.

In a document dataset, each asset is a document. Annota extracts its text page by page so you can read, correct, and approve it. See Documents and OCR.

Labels and annotations

A label is a category you want to find in your images, such as person or car. Labels belong to a dataset and have a name and a color.

An annotation is one shape on one image, linked to a label: a bounding box or a polygon. You draw annotations yourself or let Magic Select and Magic Annotate suggest them. See Labels, Annotate images, and Automatic annotation.

Dataset versions and exports

The dataset you edit every day is the live dataset. A dataset version is an immutable copy of its ready assets, labels, and annotations at a point in time. Later edits change the live dataset, never an existing version. Versions give exports and training a stable, reproducible source. See Dataset versions.

An export builds a downloadable archive from a version or from the live dataset, in a format that matches the dataset type (for example COCO for images, Markdown for documents). Exports do not use credits, but saved archives count toward storage. See Export datasets.

Models and training runs Beta

A model groups the trained versions of one task, such as object detection. A training run trains a new model version from a ready version of an image dataset. Only approved images are used. When the run succeeds, you can download the model version. Model creation is in private beta and available only to organizations with access. See Models and Train a model.

Annota Brain

Knowledge bases and documents

A knowledge base makes a document dataset searchable by Brain. Each knowledge base indexes exactly one document dataset and stays in sync with it: when you add or change documents, Brain indexes them in the background. You can create a knowledge base by uploading files (Brain creates the dataset for you) or from an existing document dataset in Studio. You can also choose to index only approved documents. See Knowledge bases and Add documents.

Conversations and sources

A conversation (or chat) is a series of questions and answers. For each question, Brain searches the knowledge bases you selected, reads the relevant pages, and writes an answer with numbered citations. Each citation opens the source document at the cited page. When Brain answers without your documents, it says so.

Brain cites its sources but does not check every statement. Parts it could not check are marked To check. Always read the cited pages before you rely on an answer. See Ask questions, Sources and citations, and Conversations.

Skills

A skill holds instructions that Brain follows whenever a request matches the skill’s description, for example a procedure your team uses to review contracts. You can create a skill or import one in the Agent Skills format, then share it. See Skills.

Plan and credits

Each organization has one plan and one credit balance, shared by both products and by every member.

Monthly credits come with the plan, are used first, and do not roll over.
Purchased credits never expire, even on the Free plan.

Actions that use AI or processing, such as OCR, audio transcription, Magic Select, Magic Annotate, knowledge base indexing, and Brain answers, use credits. Each member also has a weekly Brain usage limit that is used before the organization credits. Plans also set capacity, such as storage and how many members you can add.

See Plans and credits and Credit usage.