Guide

What aiti is, why it exists, and how to use it.

Document	Content
Project Origin	How aiti came to be: from other AI agents to pi-web, to publishing its own releases
Installation	Install the aiti web workbench: npm / one-line script / PowerShell / Docker — four methods
What is aiti	Positioning, brand, three main lines, and the relationship with pi/pi-web
Development Lineage	🧬 Full-chain provenance for AI changes: five-dimension lineage across sessions, plans, operations, and acceptance — attributable and revertible
Architecture	Layered architecture, injection chain, bridge chain, end-to-end data flow
Extension Services and Cross-Extension Call Specification	🔌 The ctx.services locator: how to provide/consume/fall back; global-variable red lines (required reading for extension authors)
Extension Data and Storage Specification	💾 The three-tier data model config/state/cache: decision tree, behavioral contract, migration iron rules, protection matrix (required reading for extension authors)
Lazy-loading contract	Two-stage conversion: thin entry stays resident + heavy UI materialized on demand
Quick Start (aiti-agent workbench)	Workbench quick start: install → launch → first-run config → daily entries
devtools quick start	5 minutes to "AI operates your page": inject → bridge → connect AI (debugging toolchain main line, see /devtools/)
MCP Bridge	Let any AI agent operate the page and collaborate with the user (Server / Sessions / Integrating agents / cockpit)