App Bundles — Architecture & Roadmap

What Are App Bundles?

App bundles are manifest-driven packages that ship functionality to all Boo customers. Each bundle contains:

• App — A backend service (typically Bun + Hono + Drizzle/Postgres + pino)
• Skills — SKILL.md files (maximum 1-2 skills per bundle with heavy use of references/ for sub-capabilities)
• Heartbeats — Scheduled tasks

Bundles follow the established pattern from artifacts, boosecret, and press bundles.

Build Process

1. Build the bundle in the boo-brain repo
2. Open a PR
3. Deploy the app on Railway
4. Test end-to-end
5. Always build for Boo Lab first, validate, then ship to all customers

Bundles in Development

LLM Wiki

An AI-powered company wiki inspired by Karpathy's LLM Wiki concept. Ingests company knowledge from web presence, Slack, docs, and other sources into a searchable, AI-queryable knowledge base.

• Storage: Markdown format with Postgres backend
• Search: pgvector for vector similarity search
• Auto-ingestion: Crawls public web presence (homepage, docs, blog, pricing) as the "public layer" on day zero
• Crawler: Agent-side skill crawls and POSTs to wiki API (not a separate service)

Content Factory

AI-powered content production system available to all customers out of the box.

• Database: Postgres day 0 (not SQLite), provisioned via bundle manifest
• API keys: Customer-owned keys for Semrush, GSC, GA4, CMS are per-tenant config. Boo's keys only for infrastructure it bills through (AI models, SERP API, image gen)
• Setup: First-class setup wizard, not an afterthought. Customer provides their own keys for tools they own.

AI Visibility Tracker (boo-aeo)

Tracks how brands appear in AI-generated search results across Google AI Mode, AI Overviews, ChatGPT, Claude, Gemini, and Perplexity.

• Service: "boo-aeo" (Answer Engine Optimization) — Bun + Hono + Drizzle/Postgres + pino
• API key model: Boo Lab owns all upstream keys. The proxy mints per-org API keys with rate limiting and usage tracking. No customer API keys required (too much friction).
• Upstream providers: OpenAI (with web search), Anthropic, Google Gemini, SERP data provider (SerpAPI or DataForSEO, not Semrush), Perplexity
• Revenue model: Could be a premium add-on with separate pricing

Architecture Principles

• Postgres over SQLite for day-0 database provisioning
• Aggressive skill grouping — 1-2 skills max per bundle
• Per-tenant configuration for customer-owned API keys
• Setup wizards are first-class screens, not afterthoughts
• Build for Boo Lab first, then generalize