Commercial Evaluation Path

GenAI Smart Router is evaluated as a governed enterprise gateway for customer workloads. Metrum provides an evaluation path that lets buyers validate their own clients, security requirements, reporting needs, provider access, and deployment model before choosing commercial access.

To request an evaluation, contact contact@metrum.ai.

Deployment Paths

Path	Best fit	What the evaluator receives
Metrum-managed hosted service	Teams that want the fastest evaluation with a managed endpoint	A router base URL, one or more router-issued caller tokens, allowed deployment-defined model groups, and administrator-provided report excerpts for the evaluation window.
Enterprise or on-prem deployment	Teams that need the router inside their own infrastructure	A licensed deployment package, signed JSON license file, sample config, operator docs, and support for connecting approved provider keys or private upstreams.
Private customer-cloud deployment	Teams that want cloud isolation under their own account or network controls	A deployment package and implementation plan for customer-owned cloud infrastructure, private networking, identity policy, usage database, reporting, and provider onboarding.

Metrum and the customer agree on the deployment shape, license template, provider access, data handling, workload proof points, and reporting package before the evaluation starts. Production access uses either a signed enterprise license for customer-operated deployments or a private managed plan for Metrum-operated customer deployments.

Buyer Journey

1. Request an evaluation and describe the intended clients, workloads, security requirements, provider preferences, and reporting goals.
2. Receive either a Metrum-managed endpoint and router token or a deployment package with a signed license for customer-controlled infrastructure.
3. Use /v1/models to discover the deployment-defined model groups the evaluation token can request.
4. Validate workloads through the same API shape the production client will use, such as OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, Codex CLI, Claude Code, or an SDK.
5. Inspect proof from Admin Browser Reports, Report Examples, and generated usage reports: selected providers/models, cost, savings, latency, fallback, cache, quota, and security access signals.
6. Review security posture, deployment readiness, data retention, provider onboarding evidence, rollback criteria, and any required procurement controls.
7. Choose the commercial path that fits the deployment: annual enterprise self-hosted, private managed, marketplace/private-offer, renewal, top-up, or other contracted terms as agreed in the commercial plan.

Evaluation And License Options

Commercial evaluations use a license or managed endpoint sized for the proof:

Option	Typical use	What changes at production time
eval-72h	Short hosted or partner proof	Replace with a pilot, enterprise annual, private managed, or marketplace agreement.
pilot-30d	Paid validation of workloads, reporting, private upstreams, and governance	Replace with an annual or private managed license after acceptance criteria pass.
enterprise-annual	Customer-operated production deployment	Renew or amend the license as feature, volume, retention, or deployment scope changes.
credit-pack-5m / credit-pack-25m	Prepaid volume or top-up	Replace with a new issued license when the volume envelope is exhausted or expires.
marketplace-seat	Procurement through a private cloud marketplace offer	Renew or modify through the marketplace private-offer process.

The license controls product capabilities and deployment limits; provider keys, upstream choices, model groups, and caller access remain deployment-specific. Use /v1/models to see the model groups allowed for the evaluation token. For the full commercial access map, see Choose a Deployment Path.

Evaluation Evidence To Request

• Caller-facing compatibility: chat, agent, tool-call, image/VLM, streaming, structured-output, and max-token cap behavior for the client shapes that matter.
• Access control: user, project, membership, API-key, /v1/models, quota, rate-limit, and key-rotation examples.
• Cost governance: stored request-time actual cost, source-dated baseline assumptions, savings by user/project/key/group/provider-model, and caveats for historical rows that predate cost fields.
• Performance: downstream user latency and throughput plus upstream provider/model/dialect latency, TTFB, throughput, attempts, errors, and fallbacks.
• Trust and security: provider-key isolation, metrics-admin isolation, report-admin authorization, diagnostics redaction, security access reporting, private-upstream network controls, and signed-license status.
• Retention posture: raw operational rows, daily rollups, dry-run retention status, legal holds, archived exports, and the purge workflows available in the deployment.
• Provider onboarding: direct upstream smokes, router-level smokes, workload acceptance tests, source-dated pricing, tool/modality metadata, and rollback criteria.

Support Evidence

Evaluation artifacts should use request IDs, public token IDs, anonymized report excerpts, and sanitized metadata. Keep production credentials, customer content, and deployment-private details in approved support channels.

For the technical checklist, continue with Evaluate GenAI Smart Router. For deployment proof, review Deployment Readiness, Deployment Security Assessment, and Model Group Quality Criteria.