How it works

How software gets built from a conversation, and stays right afterwards.

For the person who will be asked how it actually works. Outline, the factory behind it, and the platform they both run on.

Architecture at a glance

Four stages, one platform underneath all of them.

Intent in, software out, platform underneath. What you need goes to Outline, Outline to the factory, the factory to your application. A change is described to Outline and goes round again. All three of Outline, the factory and your application query the data platform and the knowledge platform through MCP. Intent in, software out, platform underneath.describe a change, and it goes round againWhat you needspoken, in your own wordspause and resumeOutlinefast model talks,frontier model checksmock-ups, then a specThe factoryagents plan, build, testan engineer reviewskit of reused partsYour applicationscreens and agentsin your tenantacts on live dataall three query the platform through MCPThe platform (being built in the open)Data platformLive operational data, semantic layer, row-level security.ERPs, SQL databases and legacy systems.Knowledge platformExpert interviews, documents, hybrid retrieval.A base that consolidates, finds gaps and learns. Intent in, software out, platform underneath. What you need goes to Outline, Outline to the factory, the factory to your application. A change goes round again through Outline. All three query the data and knowledge platforms through MCP. What you needspoken, in your own wordspause and resumeOutlinefast model talks,frontier model checksmock-ups, then a specThe factoryagents plan, build, testan engineer reviewskit of reused partsYour applicationscreens and agentsin your tenantacts on live dataa change goes round againall three query the platform through MCPThe platformbeing built in the openData platformLive operational data,semantic layer,row-level security. ERPs,SQL and legacy systems.Knowledge platformExpert interviews,documents, hybridretrieval. A base thatconsolidates and learns.

One spec is the source of truth throughout. Outline, the factory and the application all query the same data and knowledge platforms through MCP.

Intent to spec

Two models run the interview. One talks, one checks.

A fast model runs the conversation, so it keeps up with speech. A frontier model reads the transcript as it grows and feeds back the gaps and contradictions. Better at spotting the hole, too slow to talk.

What it draws on

What Outline draws on during the interview
SourceHow Outline uses it
Your data platformEvery entity, field or rule you name is checked against the live data.
Your knowledge baseDocuments, prior interviews and expert corrections.
Screen mock-upsDrawn as you talk, in your brand, confirmed by you.

What comes out

A specification folder any team can read and any agent can build from: testable requirements, acceptance scenarios, flows with their exceptions. You own it.

Specification format

EARS statements, Gherkin scenarios, Cockburn-style flows, a Volere-style requirement shell, ISO/IEC/IEEE 29148 as the coverage model, serialised as a Spec Kit / Kiro shaped Markdown folder.

Spec to software

Built from the approved spec, tested against it, signed off by a person.

Agents plan from the spec, build against the platform, and run its acceptance scenarios as tests. What fails goes back round. An inmydata engineer reviews what passes.

The spec is executable

Testable statements and scenarios, so the build has a definition of done from the first minute.

The platform is already there

Data, auth, access control and knowledge retrieval come from the platform, not the application.

The kit grows

Parts that pass review are reused. Early applications take days; later ones take less.

  1. 01

    Interview

    Spoken, checked against your data.

  2. 02

    Spec

    Approved by you. The source of truth.

  3. 03

    Build

    Agents build and test. An engineer reviews.

  4. 04

    Your tenant

    Screens and agents on live data, in our cloud or yours.

One spec, interview to tenant

Every build runs through gates we record, and the software shows its working. Read how we build.

Data platform

Live operational data, made queryable by software and by models.

Connects to the systems you already run, including the legacy ones most AI tooling cannot reach. A semantic layer a model can understand, sub-second queries, row-level security on every one. One MCP interface serves the interview, the build and the application.

Data platform capabilities
CapabilityWhat it means in practice
Real-time connectorsChange data capture, no batch loads.
Semantic layer"Overdue orders" means what your finance team means.
Sub-second queriesMany questions per decision.
Row-level securityApplied once, inherited by every application and agent.
Runs where you need itOur cloud, your cloud, or a Windows or Linux server on your premises.
Knowledge platform

The judgment your experts carry, captured and kept current.

Most of how a business runs is in people, not systems. The knowledge platform captures it through AI-led interviews and the documents you already have, and keeps it retrieved, not just stored. Outline builds from it. The application runs on it.

Knowledge platform capabilities
CapabilityWhat it means in practice
Expert interviewsBy voice, over as many sessions as it takes.
Document ingestionPolicies, procedures, wikis and email threads.
Hybrid retrievalThe passage that answers, not the one that matches.
A base that learnsContradictions consolidated, gaps asked about, expert corrections win.
How machine-written knowledge earns trust
How software stays right

Change is the normal case, so the pipeline is built for it.

The business, the platform and the models all change under an application. Each is a routine event with a known path, not an incident.

What happens when things change
What changesWhat happens
The businessDescribe it to Outline. Spec updated, delta built, full suite run.
The platformPatches roll out across every tenant, each application’s tests first.
The modelsTested against each application’s scenarios before it is switched.
Your schemaThe semantic layer absorbs most of it. The rest is regenerated from the spec.
Your knowledgeCorrections and new documents flow in continuously, without a release.
Why patching has to run without a person
Open source and security

Built in the open.

We are opening the data platform and the knowledge platform: run them yourself, read the code, leave with everything intact. We charge for the factory and for keeping your software right.

Security and data

Security and data
PointDetail
Access controlRow- and column-level security, inherited by everything on the platform.
Your dataYours outright. Deleted shortly after termination. DPA in the standard terms.
AuditEvery query and agent action is logged. Logs on request.
CertificationsNone yet. SOC 2 compliance is a design goal. Listed when held, not before.
ModelsYour choice of provider where allowed, billed to your own key.

Ask us the hard questions.

Bring your architect. Half an hour on your data, your security model and what the factory can’t do yet.