Licensed to be used in conjunction with basebox, only.
// admin
Policies
basebox enforces or displays organisation-wide rules: audit log level and retention, the notice in the chat, web search approval, connectors, write permissions, limits, system prompt. This page is the map: what exists, where it is set and what it does technically – so you can translate data protection and compliance requirements into concrete settings.
Every compliance question has a setting as its answer
Data protection officers ask "What do you record, for how long, and who sees it?". IT security asks "What may go outside?". Management asks "What does it cost?". Each of these questions has one or two settings in basebox as its answer.
basebox enforces most policies technically
| Policy | What it does | Where | Enforced or displayed |
|---|---|---|---|
| Audit retention | How long audit entries are kept (default 6 months, adjustable 3–36 months) | Administration → Organisation, Audit Log Level section | Enforced |
| Audit Log Level | Standard: sign-ins and changes; High: chats as well (then also stored on the server) | Administration → Organisation | Enforced |
| Notice in the chat | As soon as conversation content is recorded, all users see a clear notice | Automatic, follows the audit log level | Displayed – cannot be switched off |
| Web search per app | Off by default; in which apps users can switch it on | App settings | Enforced |
| Consent before the first search | "Enable web search" dialog with warning | Automatic | Displayed – cannot be switched off |
| Connectors | Which external systems are reachable | Administration → Connectors + per app | Enforced (egress allowlist at the MCP gateway) |
| Write tools | Whether a connector may change things in systems | Per app, off by default | Enforced |
| Limits | Upload sizes, file count, usage limit | Organisation → Limits | Enforced |
| System prompt | Style, language, behaviour rules for all chats | System prompt | Behavioural instruction to the model – not a technical block |
| App sharing | New apps are private; sharing with groups or organisation | App sharing | Enforced |
| Roles | Administrator vs. user | Users | Enforced |
Recording and retention
By default the audit log records security-relevant actions: logins, changes to users, groups, messages of the day, apps, exports – each with timestamp, user, result, resource and client address. For web search calls, organisation, user, tool, provider, result and timestamp are recorded, not the text of the query.
Two settings go beyond this:
- Audit Log Level: with High, chats are recorded as well. Otherwise chats are stored only in the users' browser database; with "High" they are also stored on the server.
- Retention period: 6 months; you can set 3 to 36 months. Details: Audit Log.
What users can do themselves – rename, delete, back up chats – is under Chat history & data.
What may go outside
On Server and in the Cloud from its launch, content does not leave the basebox environment in normal operation; in the basebox Cloud transition phase until November 2026 every model request goes to Google Vertex AI in Frankfurt (Model providers). In hybrid operation basebox runs in your network and the AI models come from another location; model requests then go there. You control three deliberate exceptions:
- Web search – queries go to Staan.ai in the EU, only the search term, without the person's identity or IP address; basebox has a DPA with Staan, and training is excluded. Off by default: you enable it per app, the person switches it on per chat and confirms before the first search. Recommendation: do not combine web search and internal connectors in the same app, i.e. a dedicated app whose only connector is web search – see Introduction to web search security.
- Connectors – access your own systems with the respective person's credentials; the MCP gateway lets through only the declared hosts. See Connector permissions.
- App with an external model (optional) – none by default; you decide whether there is one and set it up yourself. Requests in this app go to the external provider. You decide via App approval which people and groups use it; see Model providers.
On basebox Server the infrastructure comes on top: firewall, egress, optionally air-gapped operation – the Platform Operator's job, see Air-gapped environments.
Behaviour rules for the assistant
The system prompt applies to all chats of the organisation – style, language, blocked topics. It is usually followed, but it is an instruction to the language model, not a technical block. What has to be enforced you control via the sharing of apps and connectors. Users complement it with their personal personalisation; both apply together.
Implementing a requirement
Example "No patient data in external services":
- Do not enable web search in any app – or only in a dedicated app for a trained group.
- Enable connectors only for internal systems; leave write tools off.
- State in the system prompt that no personal data belongs in web searches – as an additional layer.
- Communicate the usage rule (message of the day, welcome message).
- Check the audit log regularly for web search calls.
The defaults are restrictive
Note
- Defaults are restrictive: web search off, new apps private, write tools off, connectors only with personal credentials.
- Every change to these settings is itself in the audit log.
- How to assess the settings in the context of Cloud or Server – controller/processor, storage locations, deletion – is under Security & Compliance.
Frequently asked questions
Can I hide the notice in the chat when recording is enabled? No. It is part of the transparency commitment towards users.
Are chats recorded by default? By default the audit log records actions, not conversation content. Chats only if you set the audit log level to High.
Can I block topics technically? Only indirectly – via the system prompt (behavioural instruction) and via sharing (which apps, connectors and knowledge bases exist). basebox deliberately does not offer content filtering as a security guarantee.
Who may change policies? Administrators only.
Next step: Audit Log · Security & compliance
Need help? Contact support