Overview
This system is a minimalist, filesystem-based context management tool built specifically for an AI personal assistant. Because the primary interface is a conversational Telegram bot, traditional GUI-based task management is discarded. Instead, the AI leverages a flat, hierarchical context routing strategy using Markdown files, ensuring low-latency responses without the overhead of heavy vector databases.
The system strictly prioritizes execution, prioritization, and scheduling over passive knowledge storage.
Folder Structure
The structure is optimized for instant AI read/write operations. There is no Inbox folder. The AI processes, categorizes, and routes raw Telegram input instantaneously to the correct file.
/Hipi Assistant
├── CONTEXT.md # Always-loaded global state + memory (Priorities, Schedule, Active routing, Ongoing)
├── /10_areas # Ongoing responsibilities and standards
│ ├── infrastructure_maint.md
│ ├── freelance_operations.md
│ └── ...
├── /20_projects # Active pipelines (L1 active memory files)
│ ├── ai_website_builder.md
│ ├── local_llm_context.md
│ └── ...
├── /30_archive # Completed/Inactive projects (Excluded from AI context window)
└── /history # Daily conversation-history backups (system-managed)Implementation: L0 and L1 Routing
To maintain speed and token efficiency, the assistant does not read the entire Drive folder. It uses hierarchical state management:
L0 — Global State (Global Awareness)
CONTEXT.md is bound to the AI's default context window. It contains who you are, the daily schedule, prioritized next actions, and a directory of active projects. The AI reads this to answer questions like "What is my priority today?" This file is always in context — every turn.
L1 — Project Files (Active Memory)
Every file in /20_projects/ acts as an L1 metadata structure. When a Telegram message explicitly pertains to a specific project, the AI dynamically fetches that specific L1 file to understand local context, roadblocks, and next steps, updates the Markdown file, and unloads it. L1 files are not loaded into the context window by default — only on demand.
Zero-Inbox Routing
Because the AI possesses instantaneous processing capabilities, human bottlenecks in triage are eliminated. When a thought, task, or link is sent via Telegram, the AI instantly clarifies and organizes it. It determines if the input is a calendar event, a new next action, or a project update, and writes it directly to CONTEXT.md or the respective L1 project file. There is no Inbox to process later.