Skip to main content

Enterprise and Private Managed Procurement

Enterprise self-hosted and private managed paths are intended for customers that need contractual procurement, dedicated deployment planning, private upstream support, security review, or workload-specific validation.

Enterprise Self-Hosted​

The customer operates the router in their own environment and installs a signed license.json. This path is appropriate for regulated, customer-cloud, on-prem, or air-gapped deployments.

Commercial and technical planning usually covers:

  • annual or contracted license term;
  • BYOK provider keys, private upstreams, or approved Metrum-managed provider access;
  • deployment scope, instance limits, and optional deployment binding;
  • licensed features such as routing, usage reporting, admin reports, dynamic routing, external policy, PII filtering, private upstreams, and exports;
  • model-group quality criteria and acceptance tests;
  • security assessment, diagnostics redaction, metrics-admin isolation, rollback, and support escalation.

Private Managed Deployment​

In a private managed deployment, Metrum operates a dedicated deployment for one customer or a customer-specific high-availability environment. This is distinct from a shared public multitenant API-credit service.

The plan should define:

  • customer isolation and network controls;
  • endpoint handoff and caller-token administration;
  • provider-cost handling, including BYOK or pass-through where applicable;
  • usage reporting, retention, and support evidence;
  • operational acceptance tests and rollback criteria;
  • renewal, top-up, cancellation, and incident contacts.

Procurement can use direct order, quote/invoice, or marketplace private offer when available. Public docs intentionally avoid private hostnames, SSH details, raw router tokens, provider keys, signing internals, and customer-specific license payloads.

Acceptance​

Before production use, validate the same client surfaces that will be used after rollout: OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, Codex CLI, Claude Code, tool calls, image/VLM requests, streaming, and structured outputs as applicable.

Use usage reports and admin reports to confirm provider/model selection, request-time cost, latency, fallback behavior, quota enforcement, license status, and access-control results.