Cheasee-Pi: Build Your Own PI. Cheap. Easy. Secure.

Token-saving agent harness with security guardrails and a Kanban git-oriented sub-agent framework. Docker + Pi AI — autonomous Kanban pipeline, sandboxed execution, real-time feedback via git worktrees for parallel development.

Cheasee-Pi TUI — multi-agent Kanban pipeline

What is Cheasee-Pi?

Cheasee-Pi is a Pi agent harness built on the Pi coding agent — engineered to save tokens, enforce security boundaries, and drive sub-agents through a Kanban git-oriented workflow. It uses a GitHub Project board to orchestrate an autonomous multi-agent pipeline — Researcher → Architect → TestDesigner → Developer → Auditor — with tools that minimise token waste, enforce security boundaries, and streamline development inside isolated git worktrees.

All components run locally. No code leaves your machine (except LLM API calls to your provider).

Extensions overview

Extension	Purpose
Structural Analyzer	AST-aware code search via ast-grep
Ripgrep Search	Fast literal/regex code search
Supervisor	Kanban-driven multi-agent pipeline
Web Crawler	Web crawling with Scrapling + Cloudflare bypass
Web Search	DuckDuckGo search via ddgs Python lib
Context Info	Rich TUI status bar (branch, model, tokens, TPS, cache)
Session Logger	Session logging to JSONL with Markdown reports
Session Advice	Post-session pattern analysis + feedback loop
Agent Harness	Runtime tool call validation (blocks dangerous patterns)
Caveman Protocol	Token-efficient communication
Ponytail	Lazy senior dev mode — YAGNI, stdlib-first, minimal code
Ask User	Interactive MC dialogs + CSV logging
Format on Save	Auto Prettier + ESLint after write/edit
PiIgnore	Path blocking via `.piignore` patterns
TSC Checkpoint	`/check` command: `tsc --noEmit`
Check Extensions	Extension compatibility audit
Worktree Sandbox	Worktree path enforcement
RTK	Token-saving bash rewrite — 60-90% less output per command
LSP Auditor	LSP diagnostics pre-audit for pipeline

Quick start

See the installation guide for full step-by-step instructions.

TL;DR for Linux:

git clone https://github.com/SchneiderDaniel/cheasee-pi.git
cd cheasee-pi
./cheasee-pi.sh

First run of ./cheasee-pi.sh builds the Docker image (~2 min) and drops you into the Pi TUI — follow the prompts to set your API key.

Documentation

Full documentation is at schneiderdaniel.github.io/cheasee-pi.

Section	What’s there
Installation	Prerequisites, step-by-step setup, verification
Architecture	System design, extensions vs MCP, git worktrees, pipeline
Extensions	All 19 extensions, agent definitions, published packages
Skills	5 reusable skill definitions
Methodology	Kanban pipeline, security, token efficiency, daily use
Prompts	11 prompt templates
SBOM	Software Bill of Materials
Acknowledgements	Credits and licenses

Daily workflow

Typical session

Start with pi
Select or create a GitHub issue
Run /supervisor <issue-number> to start the Kanban pipeline
Monitor progress via TUI status bar
Review results when pipeline completes

Update after dependency changes

When pi, ponytail, or any container dependency updates, rebuild the image:

./cheasee-pi.sh --rebuild

This rebuilds the Docker image with updated packages. The container restarts automatically.

Contributing

Fork the repository
Create a feature worktree: git worktree add -b my-feature ../my-feature main
cd ../my-feature and make your changes
Run tests: npm test
Push and submit a PR
Clean up: git worktree remove --force ../my-feature

Philosophy

Everyone should build their own Pi. This repo is my personal Pi agent harness. Fork it as a starting point, but the real power comes from shaping it into your own — your preferred tools, your workflows, your guardrails.

Customize ruthlessly. Make it yours.