How it works

From a client brief to a branded pack, in eight moves

Each step has a purpose, something you do, something the system does, and an output you can inspect. Nothing happens off-screen.

app / generation / online-skills-platform
Production
Generation studio
Running
3 of 8 complete · minimax
  • Technical Questionsdone
  • Product Requirements Documentdone
  • Statement of Workdone
  • Solution Architecturerunning
  • Milestones & Delivery Planqueued
  • Effort & Cost Estimatequeued
At a glance

The pipeline

  1. 01
    Requirements
    The brief, in the project
  2. 02
    Knowledge
    Client documents, indexed
  3. 03
    Discovery
    Classified, questioned, checked
  4. 04
    AI analysis
    Grounded in all of it
  5. 05
    Deliverables
    The core eight, in order
  6. 06
    Review
    Edited by a person
  7. 07
    Versions
    Nothing overwritten
  8. 08
    Export
    DOCX, PDF, Markdown
Step by step

What each stage actually does

The distinction that matters throughout: what you decide, and what the system does with that decision.

01

Start with the requirements

Everything downstream is derived from the brief, so it is held as a record rather than pasted into a prompt.

You
Create the engagement and write or paste the brief.
The system
Stores it on the project and treats it as the first input to every run.
Output
A project with requirements, a client and an owner.
02

Add project knowledge

Client material is what separates a generic document from one about this engagement.

You
Upload the brief pack, audits, vendor notes and transcripts.
The system
Extracts text, splits it into overlapping windows, embeds each window and stores the vectors with the project and workspace.
Output
Sources marked indexed, with a chunk count — or marked failed, with the reason.
03

Let it understand the engagement

A model asked to write a SOW straight from a brief is guessing at everything the brief left out. This is the pass that finds those gaps first.

You
Open Discovery, or simply start a run — generation refreshes it either way.
The system
Classifies the engagement, extracts requirements, assumptions, open questions, risks and dependencies as records, checks the external capabilities the brief leans on, and runs the specialists the classification calls for.
Output
A discovery model you can read, answer and confirm — reused until the brief or the sources change.
04

Configure the run

A run is a production decision: what to write, from what, with which engine.

You
Choose one deliverable or the whole pack, pick the provider and model, add a focus note.
The system
States what the run will read and what is missing before anything is spent.
Output
A readiness verdict, and a run plan for the pack.
05

Generate

Documents that depend on each other have to be written in that order to agree.

You
Start the run, and watch it or leave it.
The system
Walks the dependency graph, feeding declared dependencies first under one shared context budget, scores each finished document against its own type's criteria, and records usage per step.
Output
A new version of each deliverable, with a quality score, and a run you can cancel or retry.
06

Review the output

The document that goes to a client is the one a person approved.

You
Read each deliverable, in reader view or as the exported page.
The system
Shows the material each document was derived from beside it, and a consistency pass reports where the finished set disagrees with itself.
Output
A judgement: accept, edit or regenerate with an instruction.
07

Refine and version

Revisions must not destroy the record of what was sent before.

You
Edit in place, or regenerate from a chosen base version.
The system
Appends a version, records what produced it, and can diff or restore any of them.
Output
A lineage: what changed, when, and by whose hand or which engine.
08

Export the pack

The client receives a document, not a screen.

You
Export one deliverable, one version, or the whole pack.
The system
Renders DOCX, PDF or Markdown from the workspace's document format — one set of measurements shared by Word, PDF and the on-screen page — and records the export in the audit log.
Output
A branded file, or a ZIP of the whole pack behind a cover and contents, each document its own file in reading order.
Dependency order

Why the order is not negotiable

Writing the estimate before the architecture produces a number nobody can defend. The generator enforces the order the documents genuinely have.

Understand
  • 01Technical Questions

    no dependencies

  • 02Product Requirements Document

    after Technical Questions

Agree
  • 03Statement of Work

    after Technical Questions, Product Requirements Document

Plan
  • 04Solution Architecture

    after Product Requirements Document, Statement of Work

  • 05Milestones & Delivery Plan

    after Statement of Work

  • 06Effort & Cost Estimate

    after Statement of Work, Solution Architecture

  • 07Risk Register

    after Statement of Work, Solution Architecture

Present
  • 08Client Proposal

    after Statement of Work, Solution Architecture, Effort & Cost Estimate, Risk Register, Milestones & Delivery Plan

A run walks this graph rather than a list. Each document is generated with its declared dependencies fed first, under one shared context budget, so the document a deliverable is derived from cannot be squeezed out by whatever was edited most recently.

While a run is going

A long run is a job, not a page you have to watch

A pack run outlives the request that started it. Progress is written to the run itself, so you can close the tab, come back, and see exactly which documents are done, which is being written and which failed.

  • A cancelled run keeps the deliverables it already wrote.
  • A failed step can be retried without redoing the successful ones.
  • A run that stops reporting progress is marked failed rather than blocking the project forever.
  • The spending ceiling is re-checked before every document, not only at the start.
app / knowledge
Sources in this workspace
SourceTypeChunksState
Client brief.docxDOCX38Indexed
Existing platform audit.pdfPDF126Indexed
API limits — vendor note.mdMarkdown12Indexed
Kick-off transcript.txtTextProcessing
After generation

Revision is the normal case

Most documents are regenerated at least once — a scope change, a clarified answer, a different tone for a particular client. Every one of those is a version, and the previous one is still there.

app / projects / versions
Compare v3 → v4+234 −427
-Payment integration to be confirmed with the client.
+Payments remain on the existing gateway; no integration work is in scope.
Catalogue migration covers 4,200 SKUs.
+Milestone 1 carries 40% of the total effort and value.
  • v4Regeneratedminimax · 10 min ago
  • v3Hand-editedyou · 2 h ago
  • v2Restored from v1you · 4 h ago
The output

What actually reaches the client

A branded document with the workspace’s page furniture: logo where you put it, footer text, page numbers, confidentiality note, and sections starting on their own page if that is how your firm sends documents.

Rainstream Technologies
Statement of Work
Northwind Ltd · 1 September 2026
1. Objectives

The engagement replaces the current storefront with a headless build. Success is measured against checkout conversion, not delivery dates.

  • Lift checkout conversion by 15% within two quarters
  • Cut publish time from days to minutes
2. Scope
ItemIncludedNotes
Catalogue migrationYes4,200 SKUs
PaymentsNoExisting gateway retained
Rainstream Technologies | Page 3 of 12 | Commercial in confidence
Where it starts

A brief you already have.

Where the AI sits

One step, fed by the project.

Where it ends

A file you can send today.

Run it once, end to end

Ten minutes with a real brief tells you more than any walkthrough. The free plan is enough to do exactly that.