Manage cost

Why budgets?

AI usage costs money per request. Budgets let you cap spend — per person and company-wide — so there are no surprises and no runaway bills.

Set a budget when you create an assistant

You can give an assistant a monthly spend cap the moment you create it — see Create your first employee assistant.

Adjust budgets later

You can also set a company-wide ceiling, cap or change one person's budget, and check what anyone has spent this month, at any time. Manage this from the command line — see CLI reference → oc budget.

When someone hits their cap, their assistant pauses AI calls until the budget resets or you raise it — it never silently overspends.

Choose your AI provider

You're not tied to one vendor. OpenCrane supports two distinct ways to connect a model provider, and it's worth knowing which one you're using:

Provider keys vs. BYOK — two distinct paths

Provider keys (below) are the everyday way to add a vendor key from the CLI. BYOK (bring your own key) is a separate, org-admin-only path that provisions one raw upstream key for the whole silo directly into the model-routing layer. They are not the same mechanism — see Bring your own provider key below for when to reach for BYOK instead.

Provider keys

Add the model providers your company uses, and switch freely between them. Use Claude, GPT, or open-source models without changing anything about your assistants or skills. Budget and provider changes are recorded in the audit log.

Manage this from the command line — see CLI reference → oc providers.

Bring your own provider key (BYOK)

BYOK is a separate path for an org-admin to set one raw upstream provider key per provider for the whole silo, rather than per-assistant. The key is written into a Kubernetes Secret and registered with LiteLLM — assistants never receive it directly. Instead, each assistant gets a per-tenant LiteLLM virtual key that carries its own spend budget and model allow-list. The raw key is never returned by any read endpoint.

When you set a BYOK key for a provider, LiteLLM automatically makes that provider's models available to route assistant calls through. You can then use oc model to register specific model definitions backed by that provider credential.

BYOK is API-only today — there is no oc sub-command for it yet. See CLI reference → BYOK for the API routes and authorisation requirements.