Skip to content

The admin console

If you have the admin role, an Admin entry appears in the account menu, opening a separate console with its own navigation. This is where an instance is configured for everyone on it.

Overview

A dashboard of usage statistics — tokens, cost, runs (and failed runs), active users, cache hit rate and tool-call analytics — with charts per day, per model and per user. The same chart components power each user's own Usage tab.

Providers

The center of model configuration:

  • Add a provider by API key; its models are discovered automatically, and you enable the ones users may pick.
  • Tag models (frontier, coding, reasoning, fast, vision, cheap) to drive auto model selection, and set governance tags (local, eu, …) that gate what each provider is cleared for.
  • Add custom providers — self-hosted, OpenAI-compatible or keyless local endpoints.
  • Configure the guard model (reviews tool calls), the curator model (proposes memory updates), voice models (speech-to-text and text-to-speech), model pricing overrides, the organization governance floor, and view the governance audit log.

Users & groups

  • Users — manage the people on the instance.
  • Groups — teams with shared folders and workflows. Membership can be assigned manually or mapped automatically from an identity-provider group path.

Agents

Global agents — admin-managed specialist agents available to every user, each with a persona, a "when to use" hint, an optional required chat mode and required tool groups.

Integrations

Govern every integration independently of what its manifest declares: enable or disable it platform-wide, restrict which scopes may configure it (removing user forbids personal entries), and add required governance tags on top of the manifest's — admins can tighten, never loosen.

Skill catalog

Curate the skills marketplace: add Git repositories in the open SKILL.md format (for example, anthropics/skills) as catalog sources that users browse and install from with one click.

Budgets & command policy

  • Budgets — set monthly USD caps that cascade user → org → global; over the cap, new runs are declined.
  • Command policy — admin rules that run before the built-in command catalog (first match wins, in every security mode) to allow, require approval for, or deny specific programs or command patterns.

More for operators

Installing and running an instance is covered in the Self-hosting guide; the runtime guarantees are in Architecture and Frozen contracts.