Workshop Extension
The Minimalist Agent Harness — pi.dev
UCSI University — Agentic AI Workshop
29th July 2026
Presented by Jarod Tan
Open-source, terminal-based AI agent harness — minimal, extensible, model-agnostic.
No GUI. No SaaS. Runs in your shell.
Full source. Fork it. Own it.
15+ providers. Self-hosted OK.
Created by Mario Zechner (libGDX) · Stewarded by Earendil Inc. · 79k+ ⭐ on GitHub
Born from frustration with Claude Code's feature creep. Philosophy: ship less, let users build. 4 built-in tools, everything else is an extension.
4 tools + transparent agent loop + tree sessions.
Shell commands
Read files, glob
Create/edit files
Multi-turn chat
Tree sessions: fork, checkpoint, merge — like git for agents. Sessions are files on disk. Commitable, shareable, diffable.
Pi is a reference implementation of agent harness philosophy in production.
Production agent platform built on Pi. Validates the pattern: product on harness, harness on model.
Minimal core + extension API is the architecture that scales from CLI to platform.
Self-hosted models, local-first, no SaaS. Aligns with zero-trust requirements.
Fastest-growing dev tool. Ecosystem of skills, packages, and extensions.
We'll build with Pi, but we'll use OpenCode to do it. Install it first.
No account needed. Open source. MIT licensed.
opencode.ai
Extend Pi with community packages — install from npm, use immediately.
pi-freeFree model providers — Kilo, Cline, OpenRouter, LLM7. No API keys needed.
pi-subagentsDelegate to child agents — chains, parallel review, background workers, oracle oversight.
pi-web-accessWeb search, URL fetch, GitHub clone, YouTube & PDF understanding.
Browse all packages at pi.dev/packages
The AI coding assistant I use every day — and the tool that built this workshop.
Get 55 free conversations
opencode.ai/go?ref=HSZ7VKR1DY
Terminal-native AI agent. Open source. Model agnostic. Build your own agent harness.
Resources
📖 pi.dev · 🐙 github.com/earendil-works/pi · 💬 discord.gg/3cU7Bz4UPx
Build your harness. Own your tools. Choose your models.
UCSI University — Agentic AI Workshop
29th July 2026 · Jarod Tan