The control plane

The AI-native software factory.

One governed control plane for the entire SDLC — requirements, blueprints, work orders, build and feedback connected as a single traceable graph, so every change is proven before it ships.

Book a scoping call →See it in motion

lumaq Overview — requirements, blueprints and work orders at a glance

How the platform works

Five surfaces — your SDLC, end to end.

The last feeds the first: a closed loop, with continuous verification across all five.

01 · Requirements

Every document keeps its version history, anchored comments, and links to its source artifacts. The agent drafts, reviews, and can reconstruct requirements from an undocumented codebase — but it proposes; you accept.

A feature requirement open in lumaq — Manual review and appeals, with its overview, its review workspace rules, and each numbered requirement underneath, listed beside the rest of the project's documents.

02 · Blueprints

Container, component and feature, with diagrams. Each blueprint traces to the requirement it realizes — full requirement-to-code traceability.

A blueprint open in lumaq — a traceability and governance ledger, with its purpose, the queries it has to answer, and its integrity controls, beside the other blueprints in the project.

03 · Work orders

Scope, the requirements it implements quoted verbatim, the blueprints it touches, and end-to-end acceptance tests — everything an agent or a new engineer needs, without asking. Grouped into phases, with dependencies and one-click export.

A work order open in lumaq — its summary, what is in and out of scope, and the requirement it implements quoted with its acceptance criteria, with the requirements and blueprints it touches listed alongside and two of them flagged as moved.

04 · Build

Model-agnostic: connect any coding agent your engineers already use — Claude Code, Cursor, Devin — over MCP, and work orders arrive with full context; or run the lumaq worker on your own machines. The pull request opens under your git identity, in your repository.

A verification work order open in lumaq — its test matrix case by case, the evidence each result has to carry, and the panel where a codebase is connected and the work order is handed to a coding agent.

05 · Feedback

Your systems post real user feedback through a write-only API — no account needed for reporters — grouped into themes with quotes as evidence. A theme becomes the next requirement, and the chain from a report to the commit that answered it is a single query.

The feedback inbox in lumaq — reports grouped into what is still to process and what has been processed, each with the address it came from and the date it arrived.

The control model

Agentic, not autonomous — the control points, defined.

Verification runs continuously — it informs the decision, never blocks it. Scheduled integrity checks flag drift between code, blueprints and requirements as findings, never changes.

Control points

Rule 01

Agents don't write. They propose.

Every agent change lands on a proposal shelf — accept all, reject all, or one at a time. Automations propose too; they never approve.

if (actor.type === 'agent') {
  return proposal.create(change); // never a direct write
}

Rule 02

Agents can't close their own work.

The author can't approve their own work — no self-merge, ever.

if (workOrder.author === actor.id) {
  throw Error('cannot approve own work');
}

Rule 03

A single approval, with a separation of duties.

One human gate; the approver is never the author. Two roles, model policy set once at the top.

function approve(workOrder) {
  require(reviewer !== workOrder.author);
  release(workOrder); // one human gate
}

Exampleshelved

Channel: MCP · agent identity
Refused: A direct write to REQ-032
Proposed: 12 changes on the shelf
Review: Accept all · reject all · one at a time

Examplerefused

Author: a.tucker approves own work — refused
Approver: m.demir, named on this project
Record: Append-only · cannot be edited away
Access: Two roles · model policy set at the top

The record

What the last three engagements actually produced.

Reports rendered straight from the dashboard — git, tests and agent sessions — for three legacy estates rebuilt with every external contract preserved.

Open the reports →

Working with us

How an engagement runs — three steps.

Discovery scan · claims-core

Claims Coremapped ✓

Billing Servicemapped ✓

Field Ops APIscanning…

Scope · timeline · compliancelocked

Discover

The platform maps your systems and reconstructs what they do; scope, timeline and compliance context are locked as success criteria before the build starts.

Delivery queue · phase 2 / 4

Requirements ✓→Blueprints→Work orders · 12→Build

Acceptance testsattached ✓

WO-114on the shelf

WO-108approved

Deliver

The pipeline runs and is fully visible: work orders move through phases, agents pick them up, acceptance tests and suggestions land on the shelf. You approve at the review gate as work reaches it.

Handover package · exported

specification.pdf✓

blueprints✓

audit-trail.log✓

repo · your-org/claims-coreowned

Operate & own

Handover with the specification, blueprints and full audit trail intact. It runs in your environment, owned by you — ready to extend with us or without us.

Step through the run, or scope an engagement.

Send us the system you need modernized or built — you get scope, a delivery approach, and a costed plan against your constraints.

Book a scoping call →