Skip to main content

Review capabilities and permissions

Vox only receives capabilities allowed for the signed-in user in the active workspace. Prompt text, model output, page state, and user-supplied role claims cannot grant additional access.

Live reads​

When the user's permissions allow it, Vox can read:

  • agent lists and exact draft configuration, including prompts, models, voices, knowledge, and functions;
  • prompt-authoring context, configured variables, and runtime-appended behavior;
  • workspace analytics, agent performance, calls, chat sessions, contacts, and leads;
  • current plan, included credits, purchased credits, usage, and reservations;
  • application-level workspace status, active calls, agent states, and recent application errors;
  • knowledge bases and their source summaries;
  • available voices and reusable function types;
  • configured functions, supported integrations, their actions, and live connection state;
  • public web results, selected website pages, and candidate knowledge-source links.

List operations return bounded summaries. Vox inspects an exact contact, call transcript, chat transcript, agent, or other sensitive record only when the request requires it.

Supported changes​

Vox can prepare these approval-required changes:

  • create a paused voice or Website Chat agent draft;
  • replace an existing agent draft configuration, including its full system prompt;
  • create a reusable function or integration function;
  • attach an existing function to an agent draft;
  • create a knowledge base with selected files, URLs, or text documents;
  • add selected sources to an existing knowledge base.

Vox does not currently publish agents, delete resources, edit contacts, or directly operate arbitrary third-party actions from the platform chat. Those require the relevant product workflow or a capability added in a future release.

Permission families​

Access is checked by area. A user may be allowed to manage agents but not billing, or view calls but not knowledge bases. Agency and subaccount roles use their corresponding permission scopes. If a capability is not permitted, it is omitted from the model's available catalog and its invocation is denied server-side.

Status is intentionally limited​

“Workspace status” means application data visible to the active tenant: agent states, active calls, and recent call or Website Chat errors. It does not expose server processes, database health, deployment logs, secrets, security events, or provider infrastructure status.