Help
Glossary
This glossary explains the terms that come up again and again in Custodos and in this documentation. Each explanation is short and, where one exists, points to the detailed page.
Last checked on 28 September 2026
On this page
Workspace and people
| Term | Meaning |
|---|---|
| Workspace | A company's shared space in Custodos. The team, models, settings, chats and usage belong to exactly one workspace. One person can belong to several workspaces. See Creating a workspace. |
| Admin | The role that manages the workspace: members, usage, billing, models and settings. Admins see usage and cost, never the content of other people's chats that were not shared with them. See Roles. |
| Member | The role for everyone who works with Custodos without managing the workspace. Besides admin, the only role. |
| Seat | A person's access to the workspace. Every seat is billed individually and has a tier: Mini, Business or Max. The tiers differ only in credits, not in features. See Plans and seats. |
| Invitation link | A link through which several people join a workspace as members. It is valid for a limited time and a maximum number of people. See Inviting members. |
| Trial | The first 7 days of a new workspace, with 300 credits and no credit card. Company Brain and agents are locked during this time. |
Credits and billing
| Term | Meaning |
|---|---|
| Credits | The unit in which Custodos counts model usage. More demanding models and longer replies use more credits. The model picker shows the Credit usage for each model. See Credits and throttling. |
| Monthly budget | The credits a person can use per month. As a rule it equals the credits of their seat. |
| Throttling | What applies once the credits are used up: everyone keeps working on a cheaper model, the economy model. The model choice is locked until the reset and image generation pauses. Custodos charges nothing extra for this. Only when usage goes far beyond the monthly credits does Custodos stop accepting requests. |
| Usage | The page where admins see how many credits the workspace has used, by member and by model. Chat content does not appear there. See Usage. |
Knowledge, agents and templates
| Term | Meaning |
|---|---|
| Company Brain, brain | A shared knowledge collection of documents, web pages and pasted texts, with its own permissions. In the chat the model searches it and cites the passages as sources. See Company Brain overview. |
| Source | A single document, web page or text in a brain. New sources are added with Add sources. |
| Agents | Saved assistants with their own instructions, a model and, if you wish, brains and files. Agents cover recurring tasks and workflows. See Creating agents. |
| Prompts | Saved templates for a message, personal or for the whole workspace. Custodos asks for placeholders in double curly braces when you insert one. See Prompt library. |
| Projects | Personal collections of chats with shared instructions and, if you like, one brain. See Projects. |
Models
| Term | Meaning |
|---|---|
| Model picker | The selection in the message field. It only shows the models available in your workspace, each with where it is processed and its credit usage. See Models. |
| Auto | Uses the default model the admins have set for the workspace. If none is set, Custodos picks one of the enabled models. |
| Model groups | The model picker sorts models into groups: Frontier (leading US models served from data centres in the EU or Switzerland) and Open models (open weights on European infrastructure). When the region is set to Global, the group Global is added (the provider's worldwide infrastructure, with no commitment on the processing location). |
| Context window | The amount of text a model can process at once: the messages, replies and attached files of a chat together. |
Data protection and region
| Term | Meaning |
|---|---|
| Processing region | Where a workspace's prompts may be processed: Switzerland, European Union, EU + Switzerland or Global. Custodos does not offer models outside the region. See Processing region. |
| Residency | How firmly a model's processing location is backed. Enforced: the provider's endpoint itself refuses requests outside the region. Verified: Custodos can check the location independently. Contractual: the provider commits to the region, but it cannot be measured from outside. |
| Pseudonymisation | Detected personal data is replaced with placeholders before sending and restored in the reply. The model only receives the placeholders. See Protecting personal data. |
| Protection level | The workspace's preset for personal data: Standard, Strict or Custom. It decides what is replaced and what is blocked. Admins can also switch scanning off entirely; the account settings then show Protection level: Off. |
| Security mode | A switch in the chat for especially sensitive conversations. Only the model chosen by the admins works, pseudonymisation is always on, and web search and images are switched off. See Security mode. |
| Retention | The period after which Custodos deletes chats automatically. Without a period, chats stay until someone deletes them. |
| Activity log | The list of who administered members, settings and brains in the workspace. Chat content never appears there. |
| DPA | The data processing agreement. It is part of the terms of service accepted when a workspace is created. See Data processing agreement. |
Read next
Still have a question? Write to us.