Key concepts
The most important terms in Custodos explained in a few sentences, from workspace and credits to pseudonymisation, security mode and sharing.
This page explains the words you will keep meeting in Custodos and in this Help Center. Each section points to the page that covers the topic in full.
Workspace
The workspace is your company's shared space in Custodos. Members, settings, models, brains and credits all belong to it. As a rule, a company has exactly one workspace. An account can belong to several, and you switch between them in the account settings. More in Administration overview.
Member and admin
Everyone in the workspace has one of two roles. A member chats, creates and uses brains and agents, and can share their own work with others. An admin also manages members, models, data protection, usage and billing. Admins see usage per person, but never the content of chats that were not shared with them. More in Members and roles.
Seat
A seat is what your company books per person. There are three tiers: Mini, Business and Max. Each seat adds a fixed number of credits per month to the shared pool, depending on its tier, and admins assign a tier to each person. Prices are on the pricing page. More in Seats and billing.
Credits
Credits are the unit of usage in Custodos. Every reply uses credits, depending on the model and on the length of the question, attachments and reply. Web searches, generated images and processing brain documents use credits too. The whole team shares one monthly pool. When it is used up, everyone keeps working on a cheaper model until the pool renews on the billing day. More in Credits and usage.
Model and Auto
A model is the AI that answers, for example Claude, GPT or Gemini. Models differ in quality, speed, credit usage and where they process data. New chats start on Auto, the default model your admin has set. You can choose any enabled model per chat. More in Choosing a model and Choosing the right model.
Region
The processing region decides where models may process your messages: Switzerland, European Union, EU + Switzerland or Global. An admin sets it for the whole workspace. Models outside the region are not offered at all. Your account settings show which region applies to you, under How your data is handled here. More in Where your data lives.
Company Brain
A brain is a shared knowledge base on one topic, for example a firm handbook with the "Fee Schedule 2026" and the "Expense Policy 2026". It holds files, web pages and pasted text. Add a brain to a chat and the model searches it, and the answer names its sources. Who may use and edit a brain is set per brain. More in Company Brain overview.
Agent
An agent handles a recurring task, such as writing quotes, with fixed instructions, its own knowledge from brains and files, and prepared conversation starters. You describe it in your own words and test it before you save it. Then you keep it to yourself or share it with your team. More in Agents overview.
Project
A project groups chats that share instructions and a brain, for example all chats about the client Holzwerk Brunner GmbH. The instructions apply to every chat in the project. Projects are personal: only you see your projects. More in Projects.
Pseudonymisation
Pseudonymisation replaces detected sensitive details such as names, addresses, IBANs or AHV numbers with placeholders like <NAME_1> before sending. The model sees only the placeholder, and Custodos restores the real value in the reply. Pseudonymise is the recommended protection mode, alongside Block and Off. More in Protecting personal data in chat and How pseudonymisation works.
Security mode
Security mode is a switch in the chat for particularly sensitive conversations. An admin offers it and chooses the model. In such a chat only that model answers, pseudonymisation is always on, the web is never searched, and images cannot be attached. Once the chat has started, the mode cannot be switched off. More in Security mode.
Web search
With web search, a model looks up current information, such as prices, news or company details, and cites its sources. The search service receives only the query the model writes, never the conversation. Whether searching is allowed is the admin's decision. How thoroughly it searches is yours, in the chat. More in Web search.
Sharing
Chats, brains and agents can be shared within the workspace, with selected people or with everyone. A shared chat is read-only for others. For a brain, you decide who may edit it and who may only use it. There are no public links: to open anything, a person must be signed in and a member of your workspace. More in Sharing and finding chats and Access and sharing.