System Record
Artemis Evolution Console
How this AI-enabled execution platform was created, evolved, and safely maintained. A living system record for how Artemis is created, tested, deployed, and improved through AI-assisted implementation loops.
Position
An execution bridge, not a chatbot
Artemis is a public-safe AI implementation platform. It is not a demo gallery and not a chat interface — it is the bridge between intent and shipped software: ideas enter as prompts, and verified, reviewed, deployed pages come out the other side.
Origin
From local experiments to a governed platform
Artemis did not start as a platform. It evolved — from local experiments and standalone HTML tools, through AI-generated prototypes and route-based demos, into public-safe Vercel deployments. Each stage kept what worked, rebuilt what didn't, and moved everything private behind the boundary.
Local experiments
Disposable prototypes proved ideas fast, with zero exposure.
Standalone HTML tools
Single-file apps established the tool patterns still in use.
Route-based demos
Prototypes became addressable pages inside one application.
Public-safe rebuilds
The best work was rebuilt natively, sanitized, and published.
Architecture
The whole machine, in eight layers
Every part of the system has one job and a clear boundary. Brand presence, product platform, source of truth, framework, deployment, AI builders, controlled APIs, and the future authenticated tier.
AGOraXAI / Squarespace
External presence
Media, traffic, brand storytelling, and public-facing identity. The front door that routes attention toward the Artemis platform.
Artemis / Vercel
Product platform
The software product itself — this site. Every route, lab, tool, and demo lives here as a deployable, versioned application.
GitHub
Source of truth
Version history, branch isolation, review, and recovery. Any state of the platform can be inspected, compared, or restored.
Next.js / React / TypeScript
Application framework
Typed, route-based application architecture with Tailwind and a shadcn-style component system for a consistent executive visual language.
Vercel deployment
Preview & production
Branch pushes generate preview deployments; merges to the production branch deploy live. Rollback and revert are first-class operations.
AI build tools
Implementation layer
ChatGPT, Codex, Claude Code, Kimi, Gemini, and 9Router-style model routing — orchestrated as builders inside a governed workflow.
APIs
Controlled connections
External data and service connections are added deliberately, scoped narrowly, and never expose credentials on public pages.
Future data & auth layer
Supabase or equivalent
A planned authenticated tier: private workspaces, databases, and audit logs — kept fully separate from the public-safe surface.
Builder Matrix
Every AI tool has a defined role
Artemis is built by a roster of AI tools working inside one governed workflow. No tool free-lances: each has a role, and every output passes the same gates before it ships.
ChatGPT
StrategistStrategy, architecture, critique, prompt design, and documentation. Shapes what should be built before code is written.
Codex
ImplementerImplementation and direct code changes — turning specified intent into working diffs.
Claude Code
Repo engineerRefactoring, testing, page building, and repository-level work — multi-file changes verified with typecheck, lint, and build.
Kimi
Variant generatorAlternate implementations and design variants — parallel takes on the same brief so the best version can be selected.
Vercel
Deployment railPreview deployments for every branch, production deploys on merge, and instant rollback when something regresses.
GitHub
Memory & recoveryBranch history, pull-request review, and recovery. Nothing is lost; every change has an author, a diff, and a reason.
9Router
Model routingA local AI routing / proxy concept — directing each task to the model best suited for it, under one workflow.
Operating Loop
Minimum founder involvement, by design
The founder supplies intent and final approval. Everything in between — implementation, verification, preview — is handled by the loop. This is what makes the platform maintainable with minimal hands-on time.
- 1
Idea
A need or improvement is identified.
- 2
Prompt
The intent is written as a precise, scoped brief.
- 3
Branch
Work starts on an isolated Git branch — never on production.
- 4
AI Build
An AI builder implements the change end to end.
- 5
Typecheck / Lint / Build
Automated gates must pass before anything ships.
- 6
Vercel Preview
A live preview URL is generated for the branch.
- 7
Human Review
A person inspects the preview and the diff.
- 8
Production Merge
Approved work merges and deploys to production.
- 9
Evolution Log
The change is recorded in docs/evolution/ for the next builder.
Vercel's Git workflow underwrites the loop: branch pushes and pull requests generate preview deployments, merges to the production branch deploy live, and rollbacks and reverts are part of the standard deployment flow.
Safety Boundary
What never crosses onto a public page
Artemis maintains a hard boundary between the public-safe surface and everything private. Public pages must use synthetic, sanitized, or explicitly approved data only. The following never appear here, in any form:
The boundary, enforced
- No API keys
- No .env.local contents
- No secrets or tokens of any kind
- No SSNs, EINs, or bank data
- No raw ERP data
- No private local file paths
- No client-specific records unless sanitized or approved
Private apps, credentials, raw client files, and sensitive operational data live behind the boundary — never in this repository, never on this domain.
Timeline
Six phases of evolution
Each phase changed what Artemis is — not just what it contains. Phases 0 through 4 are history; Phase 5 is the roadmap.
Local experiments
Standalone HTML tools, AI-generated prototypes, and local-only apps — fast, disposable, and private by default.
AGOraXAI context
The Squarespace brand presence takes shape and the Artemis subdomain direction is set: a dedicated product platform beside the media brand.
Vercel + Next.js foundation
The platform moves to a typed, route-based Next.js application with Git-backed deployment — the foundation everything now builds on.
Labs & proof library
Route-based demos, labs, and a proof library turn one-off experiments into a navigable, growing body of work.
Public-safe rebuilds
Standalone tools and demos are progressively rebuilt as sanitized, native pages — synthetic data only, nothing private exposed.
Authenticated platform
The future tier: private workspaces, APIs, a database, auth, and audit logs — a governed platform behind the public surface.
Handover Protocol
How any AI builder should continue Artemis
This is the operating method any builder — human or AI — must follow to extend the platform. It is what keeps a system built by many tools coherent, recoverable, and safe.
Never edit production blindly
Production is the output of the loop, not the workspace. All changes start from intent, on a branch.
Work on a branch or isolated route
New work lives on its own Git branch and, where appropriate, its own route — so nothing existing is disturbed.
Preserve existing routes
The homepage, Labs, Tools, and every published page must keep working exactly as before your change.
Do not expose secrets
No keys, tokens, environment variables, private paths, or client data may appear in code, content, or commits.
Run typecheck, lint, and build
All three gates must pass locally before a change is proposed. A red build never ships.
Return a Vercel preview link
Every proposed change comes with a live preview deployment so a human can inspect it in the real environment.
Document changes in docs/evolution/
Each change gets a short evolution log entry: what changed, why, and what the next builder should know.
Only merge after review
Human review is the final gate. Approval — not automation — is what promotes work to production.
Continue
See the platform the loop produces
Explore the labs, tools, and demos that this system record describes — every one of them shipped through the loop on this page.