Skip to content

// admin

API permissions

Overview

What an API key can do: it accesses the OpenAI-compatible and the REST API on behalf of your organisation and uses the models enabled for your organisation. This page explains the scope of a key, what you can control with it and how to limit access.

What it is for

API keys are the entry point for software: scripts, custom applications, editor integrations, coding agents. As an administrator you decide whether such access exists, how many keys there are and for whom. Creating and deleting: API keys.

Where to find the setting

Administration → API keys. The section appears only if your licence covers API use – and on basebox Server only if the Platform Operator has set AISRV_ENABLE_REST_API=true.

What a key can do

Access Scope
Chat completions POST /v1/chat/completions (OpenAI-compatible) and POST /rest/v1/chat/completions (REST)
Models GET /v1/models lists the models enabled for your organisation; exactly these can be used by the key
Reasoning Thinking modes can be controlled via the API if the model supports them
Tool calling Function calls defined by the connecting application itself
Swagger UI /v1/api/docs for testing

A key is always bound to one organisation – the one it was created in. Every request carries the key as bearer token and the X-Realm header with the organisation's realm (Cloud: your subdomain; Server: usually primary).

What a key cannot do

  • No administration. Users, groups, apps or settings cannot be managed via the API.
  • No foreign organisations. A key acts only within its organisation's realm.
  • No user identity. Requests run on behalf of the organisation, not a person. Connectors with personal credentials are therefore not available to the API.
  • No graded permissions per key. All keys of an organisation have the same scope; there are no scopes for individual models or apps.

Step by step: limiting access

Because keys cannot be restricted individually, you control via number, assignment and lifetime:

  1. One key per integration. Use descriptive names ("crm-export", "vscode-team-a"). That way you can see in the table what belongs to what and revoke selectively.
  2. Issue only to responsible people. The key is shown only once; whoever receives it is responsible for storing it securely (password manager, environment variable – never in Git).
  3. Review regularly. Check the Created column and delete keys whose integration no longer runs. Deletion takes effect immediately.
  4. Watch consumption. The Dashboard reports usage of the OpenAI-compatible API separately from the web interface. Unusual spikes are a reason to rotate a key.
  5. Limits per organisation cap total consumption – including via the API. See Settings.
  6. Rotation. Create a new key, enter it in the integration, delete the old one.

Notes

A key is a password for your organisation

Whoever holds the key can make requests and consume tokens on behalf of your organisation. Treat it accordingly and revoke it immediately on any suspicion.

Note

  • Your organisation's system prompt also applies to requests via the API.
  • An invalid key returns a precise authentication error (401); a missing licence or permission a 403. Error codes in detail: Error handling.
  • For Anthropic models, basebox adds no artificial instructions via the API.

Frequently asked questions

Can I allow a key only one specific model? No. A key can use all models enabled for the organisation. Limit the model list at organisation level – see Enabling models.

Can a user without the admin role create a key? No. Administrators only.

Can I see which requests were made with which key? The dashboard shows the organisation's total API consumption. A breakdown per key is currently not available in the interface.

Does a key expire? No, keys are valid until deleted. Plan rotation yourself.

Need help? Contact support