// Setup

What I Use

The exact hardware, AI models, dev tools, infra, and services powering 54 apps across 7 categories. Updated February 2026.

Hardware

Mac Mini M4 Pro

Always-on server. Runs OpenClaw, Docker (8 services), all 40+ automated pipelines 24/7. Zero cloud compute required for orchestration.

MacBook Pro

Primary dev machine. All coding, architecture, and local experimentation happens here before it ships to the Mac Mini.

4K Monitor

Single display. Dashboard-heavy workflow — mission-control, trade monitoring, terminal.

AI / LLMs

Claude (Anthropic)

Sonnet 4.6 for complex agentic work — coding agents, synthesis, architecture decisions. Primary reasoning layer. Tier 4 API access.

Gemini 2.5 Flash

Monitoring, post-mortems, fast synthesis, image analysis. Cost-optimized for high-frequency pipeline tasks. Powers invictus-sentinel post-mortems.

GPT-4

Model routing decisions. Used selectively when task profile better suits OpenAI's architecture.

Leonardo AI Phoenix 1.0

Cinematic hero image generation. 24.5k tokens/month. Used by blog-autopilot to generate every published article's cover image.

Dev Tools

VS Code

Primary editor. Clean configuration — no bloat. Extensions only for what's actually needed.

Claude Code

AI coding agent CLI (this is what spawned this site). Deployed via OpenClaw's coding-agent skill for implementation-heavy tasks.

GitHub + GitHub Actions CI

All repos under Invictus-Labs. 90%+ test coverage mandate enforced via CI on every PR. Non-negotiable.

Vitest

Primary test runner. PolyEdge has 1,063 automated tests at 93% coverage. Every system gets the same treatment.

Docker Desktop

8 containerized services with health checks. Manages all persistent background services on the Mac Mini.

Infra / Deploy

Vercel

Static Next.js deploys for jeremyknox.ai and other public-facing sites. GitHub → auto-deploy on merge to main.

Cloudflare Pages + Tunnels

Secure public exposure of local services with zero open ports. Other projects deployed via Cloudflare Pages.

Docker (8 services)

Containerized services with health checks, automated restarts, and volume mounts for state persistence.

SQLite

Embedded state management. trades.db (PolyEdge), indecision.db (InDecision signals), kanban state. No Postgres required.

launchd (macOS)

4 always-on daemons via macOS launchd. Survives reboots. OpenClaw, monitoring, and core infrastructure stay up.

Languages

TypeScript

Primary language for all systems. Next.js sites, PolyEdge trading engine, OpenClaw bridge, and every client-facing app.

Python

Automation, trading bots, ML pipelines. blog-autopilot, InDecision scoring engine, Leonardo image generation pipeline.

Bash

Ops scripting. Deployment hooks, health checks, pipeline orchestration glue. Used sparingly and purposefully.

Content / Creative

Leonardo AI

Hero images for every blog post. Phoenix 1.0 model. Cinematic, dark, consistent with the brand aesthetic.

Supadata

YouTube transcript extraction. Powers blog-autopilot's source gathering step — pulls transcripts from target YouTube videos for article generation.

Beehiiv

Newsletter platform. Signal Drop — automated crypto/AI intel digest. Subscriber-gated content layer.

Gamma.app

Presentation generation. OpenClaw skill auto-generates decks from topic + outline when needed.

Services / APIs

Anthropic API (Tier 4)

Shared 60-second rolling window rate limiter, 1.9M TPM ceiling. All Claude Code sessions and OpenClaw pipelines share this quota.

Perplexity

Real-time web search for OpenClaw. When the agent needs current market data or news, Perplexity is the search layer.

Discord

Async comms layer for all pipeline output. Every bot alert, trade signal, blog PR, and system event routes through Discord.

Polymarket API

Live trading venue for PolyEdge. Binary options on crypto price outcomes. $5 live bet size, fee-aware execution.