Three things you receive at handover.

When a Vellum system goes into production, three artefacts go with it. Each has a defined purpose. Together they let your team take operational ownership without an indefinite dependency on us.

Documentation

A written architecture for your system. What each component does, how the layers connect, where state is stored, how to reason about failures. Not auto-generated — written for the team that will operate it.

Training

Sessions with your operating team. We walk through the system live, demonstrate the inspection workflow, answer questions in context. Recorded and retained for future hires.

Guided Inspection

A defined workflow your team uses to inspect, evaluate, and adjust system behaviour as your needs evolve. The system isn't a black box — there's a published method for looking inside.

The Inspection workflow.

Operating an AI system is different from operating conventional software. Outputs are probabilistic, inputs evolve, and behaviour can drift. The Guided Inspection workflow is the four-stage method your team uses to keep the system aligned with your needs.

01

Identify

Spot outputs that don't match expectations, or workflows that need to change.

02

Inspect

Use the system's audit trail and inspection tools to understand why the output occurred.

03

Propose

Frame the adjustment — a schema change, a prompt update, a routing rule, a new approval gate.

04

Approve

Test the proposed change in a sandbox. Approve it. Deploy. Monitor.

Managed Operations — when you want backup.

Most Tier 02 and all Tier 03 engagements include a Managed Operations retainer. Your team operates the system; we provide a defined level of support behind it. Monitoring, model updates, prompt tuning, quarterly reviews, and incident response within an agreed SLA.

Managed Operations is optional for Tier 01, recommended for Tier 02, and required for Tier 03. See engagement models for details.