The Problem
Most AI tooling is fragmented—separate chat clients, browser automation, scraping, and agent systems that don't share context or memory. Setting up a personal AI infrastructure means stitching together incompatible tools and losing state between sessions. This repo attempts to solve that by packaging AI capabilities as installable "Packs" that share a common runtime and memory system.
What This Does
PersonalAIInfrastructure (PAI) is a structured collection of AI skill packs, organized under Packs/ with 181 files. Each pack follows a consistent pattern: a SKILL.md defining behavior, Workflows/ for task sequences, Tools/ for TypeScript implementations, and INSTALL.md/VERIFY.md for setup. The pai-core-install pack contains the system architecture documentation (PAISYSTEMARCHITECTURE.md, MEMORYSYSTEM.md) and core utilities like FeatureRegistry.ts and SkillSearch.ts.
The repo includes 23 packs covering browser automation (pai-browser-skill), multi-agent orchestration (pai-agents-skill with 11 agent personas), a structured problem-solving algorithm (pai-algorithm-skill), and specialized tools like annual report fetching and BrightData scraping. The Bundles/ directory contains an official installer that orchestrates pack installation.
How To Use It
Setup: Run the official installer from Bundles/Official/install.ts. The .env.example file indicates environment variables are required, though the specific ones aren't documented in the analysis. Configuration: Copy .env.example to .env and populate values. The pai-brightdata-skill and pai-browser-skill packs likely need API credentials, but the exact variable names aren't visible in the file listing. Running it: Each pack has its own entry point. For browser automation, the main entry is Packs/pai-browser-skill/src/skills/Browser/index.ts. The system appears designed to be invoked by an AI assistant (Claude) reading the SKILL.md files, not as a standalone CLI.
Install via the official bundle
bun install.ts
Or install individual packs by following their INSTALL.md
Example: Packs/pai-browser-skill/INSTALL.md
The README emphasizes an "AI-first install" where you point your AI assistant at the repository and let it configure itself.
Real-World Use
A technical user installs PAI, then asks their AI assistant to "research competitor pricing for our product line." The assistant loads pai-agents-skill, spawns a ClaudeResearcher agent, uses pai-browser-skill to navigate pricing pages, extracts data via the Extract.md workflow, and saves findings to the shared memory system. The pai-council-skill can then run a structured debate between multiple agent personas to validate the analysis before presenting results.
Code Health & Issues
Low - Non-reproducible builds - Packs/pai-agents-skill/src/skills/Agents/Tools/package.json declares dependencies without a lockfile. Any npm install could produce different results. Med - Inconsistent packaging - Only two packs have package.json files. Most packs are pure Markdown/TypeScript with no dependency manifest, making versioning and dependency management unclear. Low - CI is minimal - Only .github/workflows/claude.yml and claude-code-review.yml exist. No test or build pipeline for the TypeScript code. Low - Sparse test coverage - 4 test files across 181 files, all in the browser skill. The agent system and algorithm packs have no automated tests. Low - Heavily documentation-dependent - 141 Markdown files vs 25 TypeScript files. The system's behavior is defined more by prose than code, which risks drift between docs and implementation.
The Bottom Line
This is a thoughtfully structured AI orchestration framework with a clear architecture, but it's documentation-heavy and light on tested code. It's best suited for AI-savvy individuals who want a customizable personal assistant stack and are comfortable debugging TypeScript. Teams needing production-grade reliability should wait for more test coverage and lockfile hygiene.