> ## Documentation Index
> Fetch the complete documentation index at: https://engineering.datarelay.run/llms.txt
> Use this file to discover all available pages before exploring further.

# Engineering System

> A lightweight Solo AI Engineering System for turning conversation into reliable software without carrying giant prompts between tools.

# Data Relay Engineering System

**ChatGPT for thinking. Cursor for implementation. GitHub for truth. Direct server access for verification.**

This system grew from a simple problem: AI-assisted development was productive, but the surrounding workflow became expensive and hard to remember.

Large handoff prompts, stale roadmaps, repeated Cursor instructions, missing design history, full-suite tests at the wrong time, and manual copy/paste between tools all created friction.

The current system separates each responsibility and keeps only the minimum context needed for the next action.

<CardGroup cols={2}>
  <Card title="Quick Start" icon="bolt" href="/quickstart">
    Adopt the smallest useful loop first, then add optional layers only when they solve a real problem.
  </Card>

  <Card title="Canonical Engineering Standard" icon="book" href="https://github.com/datarelay-labs/engineering-system">
    Machine-consumed standards, templates, adapters, validation rules, and release governance.
  </Card>

  <Card title="Daily Workflow" icon="arrows-rotate" href="/workflow">
    Brainstorm in ChatGPT, hand implementation to Cursor, verify results, and continue without giant prompts.
  </Card>

  <Card title="Session Continuity" icon="link" href="/session-continuity">
    Repository-scoped GitHub Work Packets replace manual ChatGPT and Cursor handoffs.
  </Card>

  <Card title="Remote Code Audit" icon="terminal" href="/remote-audit">
    ChatGPT can inspect an authorized development server directly through Desktop Commander Remote MCP.
  </Card>

  <Card title="Knowledge & Source of Truth" icon="database" href="/knowledge">
    See exactly what belongs in GitHub, Work Packets, Athena, the handbook, and temporary conversations.
  </Card>
</CardGroup>

## The core idea

```mermaid theme={null}
flowchart LR
    U[Owner] <--> C[ChatGPT]
    C --> W[GitHub AI Work Packet]
    W --> R[Cursor /resume]
    R --> G[Code / Tests / Git]
    G --> CI[CI & Release Evidence]
    CI --> C
    C <--> D[Desktop Commander - Development Server]
    G --> A[Athena - Derived Knowledge]
    R --> T[Telegram - Completion Signal]
```

<Note>
  GitHub remains the source of truth. Athena, this handbook, and AI Work Packets are supporting layers with different roles.
</Note>

## Why this split works

ChatGPT is used heavily before implementation: requirements, brainstorming, architecture, design, documentation, review strategy, and the exact next action.

Cursor is used when code must actually change. This keeps exploratory conversation separate from implementation work and makes the implementation step more bounded and economical.

GitHub holds the durable state: code, tests, specs, commits, PRs, CI evidence, release evidence, and repository-scoped AI Work Packets.

Desktop Commander closes another gap: when GitHub alone is not enough, ChatGPT can inspect the authorized development Linux host directly instead of reasoning only from pasted terminal output.

## Five rules that keep it small

1. GitHub is normative; the handbook and Athena are derived.
2. ChatGPT plans and reviews; Cursor implements bounded repository work.
3. Load only the context needed for the current task.
4. Run affected deterministic checks before expensive broad qualification.
5. Keep active session state in repository-scoped Work Packets, not giant handoff prompts.

## What this system is not

It is not an enterprise process framework, a central project-management database, or a replacement for Git.

It is a small set of rules that makes a solo AI-assisted workflow repeatable, auditable, and cheap enough to use every day.

<Tip>
  If you remember only one rule: **conversation is temporary; durable engineering state belongs outside the conversation.**
</Tip>