- Python — Primary language for everything: AI agents, data pipelines, generators, automation, API integrations. Anaconda environment management (conda).
- Vanilla HTML/CSS/JS — All 14+ sites are static HTML. No React, no build step, no framework. Fast, simple, cacheable.
- Flask — Python web framework for pRAG chatbot backend and API endpoints.
- Next.js / React / TypeScript — Used for GarageID (SaaS product). Stretch zone — Peter prefers vanilla.
- discord.py — All Discord bot integrations for the 4 AI agents.
Tools & Stack
Tools & Stack
Everything Peter Saddington uses to run 14+ websites, 4 autonomous AI agents, a VC fund, a race team, and multiple businesses. Updated March 2026.
14
Sites
4
AI Agents
5
LLM Providers
15+
GHA Workflows
Languages & Frameworks
AI & LLM Infrastructure
- Claude (Anthropic) — Primary coding assistant via Claude Code CLI. Opus for architecture, Sonnet for code, Haiku for search.
- Gemini (Google) — Image generation (Gemini Nano Banana 2), content generation, agent reasoning.
- Groq — Llama 3.3 70B for fast inference. Primary provider for agent pipelines.
- Cerebras — Backup LLM provider. Used morning-first to preserve Gemini quota for interactive use.
- SambaNova — Tertiary failover in the LLM chain.
- Cloudflare Workers AI — Last-resort failover. @cf/meta/llama models.
- llm-failover — Custom open-source library: Groq → Gemini → Cerebras → SambaNova → CF Workers AI. 5-provider automatic failover.
Database & Storage
- Supabase — PostgreSQL + vector DB + auth + storage. Single shared instance across the empire. pgvector for RAG embeddings, RLS for multi-tenant security.
- File-based JSON caches — videos-data.json, blog-data.json, newsletter-data.json. 7-day TTL, incremental updates.
Hosting & Deployment
- Cloudflare Pages — Static site hosting for 13 of 14 sites. Wrangler CLI for deploys.
- Render.com — Flask app hosting for pRAG chatbot backend.
- GitHub Actions — 15+ workflows: daily content generation, SEO audits, agent pipelines, health checks. Every workflow runs on schedule (launchd for local, GHA for cloud).
- launchd — macOS native scheduler for local automation. Replaces cron.
Content Generation
- yt-dlp — YouTube metadata extraction (784 video durations, dates, transcripts).
- Custom Python generators — 6+ generators that produce static HTML pages from API data: video pages, blog pages, CRS pages, daily digests, gear reviews, racing venues.
- Substack API — Pulls newsletter content (body_html, cover images, tags) for on-domain blog archives.
AI Agent Fleet
- Halperbot (HH) — Empire health monitor. Runs daily pipeline: SEO audits, site health checks, cross-domain link verification.
- Saarvis — Network intelligence. Cross-signal analysis, consistency scoring, health monitoring.
- MiniDoge — Growth and experimentation. Runs PolyDoge prediction engine, racing network expansion.
- Nyx — Security. Token rotation, secret scanning, RLS enforcement, vulnerability monitoring.
Communication
- Discord — 6+ webhook channels for agent reports, daily briefings, content alerts.
- Substack — Newsletter distribution (The Agile VC + DriveCRS).
Development
- VS Code — Primary editor with Claude Code extension.
- Git / GitHub — All repos under agilepeter org. 15+ repositories.
- Anaconda — Python environment management. conda activate automations for all pipeline work.
- macOS — Primary development machine. Darwin/zsh.
SEO & Analytics
- Custom seo-audit tool — 38 checks, auto-fix, GEO/AI citability scoring. Open source.
- Google Search Console — Organic search monitoring.
- Smart Connections (Obsidian) — Semantic search across 489 embedded knowledge blocks for cross-referencing decisions.
Go Deeper
Want to know more about the stack?
Ask Peter's AI about any tool, framework, or architectural decision.
Talk to Peter's AI →