Workspace Configuration
Organization, branding, models, and chat defaults
Under Settings → AI & features → Chat you set the defaults every member's new chats start with – such as the default assistant. Further baseline settings live in the organization, branding, and AI models settings.
Typical settings
- organization basics
- branding and document templates
- default model and chat defaults
- which teams and people may use which models
- default assistant for new chats
- chat feature and quick action enablement
- message action bar visibility below chat replies (hover-reveal vs. always visible)
- auto-compaction of long chats: workspace-wide on/off, plus a token threshold per chat model
Most of these settings are self-explanatory in the UI — clicking an option shows what it does. One exception with non-obvious behavior is auto-compaction of long chats, covered in more detail below.
Recommendation
Start with a minimal baseline and expand step by step.
Releasing a model to specific teams or people
A model that is switched on is available to every member at first. On the AI Models page, the button next to the switch sets who may use it:
- Whole workspace – every member (default)
- Only selected – only the teams and people you add
- All except – everyone except the teams and people you add
You can also add a person as an exception, for example "all except Production, but Max yes": an entry for a person always takes precedence over their teams. Before you save, the dialog shows how many members lose access and how many shared assistants use the model.
The same works from a team's side: on the team's page, switch individual models on or off for the team under Models. That way production, for example, gets just one cost-efficient model in a few clicks. Which models a person can use, and why, is shown under Members via the model icon in their row.
Good to know:
- Workspace admins are never restricted, and free models are always open to everyone.
- Team admins can add people to their team — anyone they add gets the models released to that team.
- If a chat, an assistant or the default model is set to a model a person may not use, another allowed model answers for them automatically. The reply is marked Access.
- Models requested through the API that the API key's owner may not use are rejected with an error.
- Following the same principle, Permissions decide who may create, share and publish assistants, projects and more.
Auto-compaction
Automatic summarization of long chats is off by default for the whole workspace. As a workspace admin, switch it on under Product configuration. Once it's on, a per-model token threshold appears on the AI Models page: it controls how full a given model's context window gets before MeinGPT compacts older messages, and each model starts at a sensible default. If a member reports that a chat used to be summarized automatically and now isn't, check whether this toggle or the model's threshold changed before assuming a platform issue. See Very long chats for what compaction looks like from a member's perspective.
Instruction for all chats in the workspace
The Predefined Organization Prompt field gives every member a shared standing instruction, for example tone of voice, company principles, or a fixed answer language. You find it under Settings → AI & features → Chat in the Defaults section, at the bottom. The page is visible to workspace admins.
The text is applied as an additional system instruction in every chat of the workspace, before an assistant's instructions and together with each person's personal instruction (see Preferences). It is plain text and does not attach a knowledge source.
If the AI should answer in a specific language across the whole workspace, say so explicitly, for example: "Always answer in English." Without such an instruction, the AI answers in the language the person writes in or explicitly asks for. The platform language only affects the interface.
Knowledge for every new chat
You can't link a data pool directly to the workspace. The Predefined Organization Prompt field in the chat settings is a plain text prompt, applied as an additional system instruction before every chat, but it can't attach a knowledge source – see the "Instructions – the system prompt" section in the assistant editor.
To give every member automatic access to a data pool, connect the data pool to an assistant (the "Knowledge – files and data pools" section) instead, and select that assistant under Settings → AI & features → Chat in the Default Assistant field. It is then preselected in every new chat – along with the data pool it has attached.
The same field shows whether a default assistant is already set. If it reads No default assistant, a new chat starts without an assistant (see Data sources in normal chats).