Introduction
IWE addresses the fragmentation of work context across different tools and AI services. It provides a structured environment to capture knowledge, plan work, and maintain continuity between sessions. Key components include:
- Exocortex: External memory files containing plans, context, and conclusions.
- ORZ Protocol: A ritual for every session and day (Open → Work → Close) to prevent context loss.
- Pack: A formalized knowledge base for specific domains.
- Strategist: An AI agent for automated daily/weekly planning.
Prerequisites
Per the README, prerequisites include:
- Git, Node.js, and (for the full flow) Claude Code.
- Minimum 4 GiB free RAM while Claude Code (or another agent) runs on top of IWE. Lower memory can cause tool read/write failures with a
PreToolUse hook did not respond before its timeoutmessage. - A GitHub account and
ghCLI for the fork-based quick start.
Permissions and Limitations
- License: MIT (Copyright (c) 2026 Tseren Tserenov). Forking and customization are the intended model.
- Data Handling: Three protection zones are described: local, GitHub (private repos), and platform (per-user isolation). See
docs/DATA-POLICY.mdanddocs/DATA-RESIDENCY.mdfor details. - Platform Support: macOS and Linux have first-class scenarios. On Windows, only parts are validated via Git Bash with native Python;
session-guard.shand session opening require WSL2 due to Unix lock issues. Full installation and updates on Windows remain unverified. - No Runtime Verification: This guide has not been executed or tested in a live environment.
Setup Instructions
Quick start (Git, Node.js, Claude Code already installed):
mkdir -p ~/IWE && cd ~/IWE
gh repo fork TserenTserenov/FMT-exocortex-template --clone
cd FMT-exocortex-template
bash setup.shAfter setup:
cd ~/IWE
claudeThen tell Claude: «Проведём первую стратегическую сессию» to walk through goal setting, the first plan, and environment configuration.
Full installation from a clean computer is documented in docs/SETUP-GUIDE.md (30–60 minutes). A minimal setup.sh --core variant works with any AI CLI. Readers not on macOS or not using Claude Code should refer to docs/PORTABILITY.md.
Customization
IWE is updated as a distribution. Platform updates arrive without losing personal settings.
Extensions (extensions/) add blocks to protocols:
# Добавить рефлексию в конце дня
echo "## Рефлексия дня
- Что было сложным?
- Что бы сделал иначе?
- За что похвалить?" > extensions/day-close.after.mdParameters (params.yaml) toggle protocol steps:
reflection_enabled: true # Включить рефлексию
video_check: false # Отключить проверку видео
multiplier_enabled: true # Мультипликатор IWEUpdates: bash update.sh updates the platform while preserving your extensions/, params.yaml, and edits to CLAUDE.md (3-way merge).
Repository Families
The exocortex is not one repository but a set of clearly bounded families:
| Family | Stores | Example Name |
|---|---|---|
Base (ZP/FPF/SPF/FMT-*) |
Platform principles and formats — supplied ready-made | ZP, FPF |
PACK-* |
Formalized domain knowledge (source-of-truth) | PACK-my-domain |
DS-* |
Code, plans, courses — derived from Pack | DS-strategy |
PD-* |
Personal data of a single type (no code) | PD-persona, PD-metrics |
MC-* |
Agent data (dialogues, service logs) — not read by humans day-to-day | MC-sessions |
FAQ
Q: Is an Anthropic subscription required?
A: For the full installation (Claude Code), the README recommends Claude Pro ($20/mo), with a possible move to Claude Max (~$100/mo) for unrestricted work. The minimal setup.sh --core path works with any AI CLI. See docs/SETUP-GUIDE.md.
Q: Does it work with AI tools other than Claude?
A: Yes, three agents are documented out of the box: Claude Code (full support via CLAUDE.md), Kimi Code in VS Code (reads AGENTS.md automatically), and Hermes Agent (connect Aisystant MCP). Other agents such as Cursor, Copilot, and Gemini require adaptation. See docs/PORTABILITY.md.
Complete upstream material
IWE — Intellectual Work Environment
Operating system for intellectual work. Your knowledge. Your experience. Your environment — runs on top of any AI platform.
Repository type:
Base/Formats(FMT) — a template distribution. After forking, it becomes your personal environment with AI agents.
The Problem
AI assistants can generate text, code, and answers. But most users face the same issues:
- Work context is scattered. AI services can store conversations and project materials, but decisions and plans from other tools are not always available in the next session or to another agent.
- Knowledge stays in your head. You took a course, read a book, solved a problem — but a month later, you cannot reproduce your train of thought.
- AI replaces thinking rather than enhancing it. You get an answer, but you do not become more competent. Without AI, you are back to zero.
- No system. Plans are in notes, tasks are in your head, knowledge is in chats. Everything is disjointed.
- Time slips away. It is unclear what you worked on, what you did, and where you are heading.
Solution: An IDE, but for Thinking
IWE (Intellectual Work Environment) — an intellectual working environment.
Just as an IDE combines an editor, compiler, and debugger into a single environment for a programmer, IWE combines knowledge, planning, and AI agents into a single environment for thinking.
| IDE (for code) | IWE (for thinking) |
|---|---|
| Editor → write code | Exocortex → capture knowledge |
| Compiler → check syntax | Principles → verify decision correctness |
| Debugger → find errors | ORZ Protocols (Open→Work→Close) → identify losses of knowledge and context |
| Linter → improve quality | ArchGate → evaluate architectural decisions |
| Git → change history | Strategist → work history and planning |
Key principle: exoskeleton, not prosthetic. IWE enhances your thinking rather than replacing it. After each session, you become more competent, not just receiving a result. Details: principles-vs-skills.md.
Key IWE Terms
| Term | What it is |
|---|---|
| Exocortex | Your external memory — files with plans, context, and conclusions that Claude reads in every session |
| Pack | A formalized knowledge base for your domain — the single source-of-truth for domain knowledge |
| ORZ | Open → Work → Close — a ritual for every session and every day, preventing context loss |
| ArchGate | Structured evaluation of architectural decisions by 7 characteristics (instead of "it seems good to me") |
| Strategist | An AI agent that automatically creates daily/weekly plans and tracks progress |
| 5 repository families | Recommended structure for separating code, knowledge, and data — see below |
| Full glossary: ONTOLOGY.md |
5 Repository Families
The exocortex is not a single repository but a set with clear boundaries: code, domain knowledge, and personal data are not mixed. setup.sh creates a minimal start (your strategic repository + 3 base principle repositories); the rest is created as needed.
| Family | What it stores | Example name |
|---|---|---|
Base (ZP/FPF/SPF/FMT-*) |
Platform principles and formats — supplied ready-made | ZP, FPF |
PACK-* |
Formalized domain knowledge (source-of-truth) | PACK-my-domain |
DS-* |
Code, plans, courses — derived from Pack | DS-strategy |
PD-* |
Personal user data of one type (no code) | PD-persona, PD-metrics |
MC-* |
Agent data (dialogues, service logs) — not read by humans day-to-day | MC-sessions |
Detailed data domain map and justification: ADR-004.
Work Culture — A New Style of Interaction with AI
IWE is not a set of prompts. It is a work culture: 14 elements (protocols, skills, formats) that transform chaotic interaction with AI into a managed process. Details: DP.M.008 — IWE Operating Rules.
ORZ Protocol (Open → Work → Close)
Each session and each day goes through three stages:
- Open — Claude checks the plan, defines the task, and agrees on the approach. You do not start work "from scratch" — the AI knows the context.
- Work — during work, Claude records valuable knowledge (Capture-to-Pack). Insights are not lost.
- Close — the result is recorded, the plan is updated, and the next session will begin where you left off.
Skipping Open = unplanned work. Skipping Close = lost results.
Exocortex — External Memory
Your knowledge, principles, distinctions, plans, and context are stored in files that Claude reads in every session. This is not a "prompt" — it is an accumulated base that grows with you.
Knowledge Formalization (Pack)
What you learn does not stay in your head. Valuable knowledge is formalized into Pack — a passport for the subject area. Pack is the single source-of-truth for domain knowledge. Details: LEARNING-PATH.md.
For Whom
Every professional drowns in information: 12+ tools (Notion, Google Docs, Slack, ChatGPT, courses...), knowledge is scattered, nothing is connected. AI can take into account the project materials available to it, but decisions from different tools are difficult to keep in a single relevant working context.
IWE is for those who want to change this:
- Entrepreneurs and Managers — strategizing, making decisions, managing projects. IWE provides a system: from weekly planning to formalizing domain knowledge
- Engineers and Developers — working with code and architecture. IWE preserves context between sessions; the AI knows your codebase, technical debt, roadmap
- Researchers and Analysts — studying, synthesizing, publishing. IWE turns scattered notes into a structured knowledge base that grows with you
- Anyone doing intellectual work — who wants symbiosis with AI, not dependence on it. An exoskeleton for thinking, not a prosthetic
Use Cases
Work Projects
| Scenario | What happens | Details |
|---|---|---|
| Product Development | Claude knows the architecture, technical debt, and roadmap. Each session is a continuation, not starting from scratch | SC.013, SC.015 |
| Documentation Management | Knowledge is recorded in Pack during work. No need to "write docs later" — they are written while working | SC.004, SC.014 |
| Project Coordination | WeekPlan, DayPlan, registry of work products — Strategist helps plan and track progress | SC.001, SC.002 |
| Review and Refactoring | ArchGate evaluates solutions against 7 characteristics. Not just "it seems good to me," but a structured assessment | SC.015 |
Personal Development
| Scenario | What happens | Details |
|---|---|---|
| Taking a Course | Claude helps record key ideas, asks questions to verify understanding, connects new material with what you already know | SC.003 |
| Writing Articles | Creative pipeline: note → draft → outline → publication. Each artifact is tracked | SC.005 |
| Strategizing | Weekly session: review of the previous week, planning for the new one, alignment with goals. Strategist prepares a draft — you make decisions | SC.011 |
| Building a Knowledge Base | Your Pack grows. In six months, you have a formalized domain knowledge base, not just a collection of notes | SC.014 |
Full catalog of 15 scenarios: USE-CASES.md
How it looks in practice
- Morning — Strategist creates a plan: Telegram notification + DayPlan file in the repository
- Open VS Code →
claude→ Claude knows what's in the plan and suggests starting with priorities - Work — Claude records knowledge as you go (Capture-to-Pack)
- Close session — results are recorded, plan updated
- Monday — Strategist prepares a draft weekly plan, you discuss it during the strategizing session
Machine Requirements
Minimum 4 GiB of free RAM while Claude Code (or another agent) runs on top of IWE. With less memory — especially on shared servers with multiple users — file read/write tools may periodically fail with messages like PreToolUse hook did not respond before its timeout. This message indicates insufficient host process memory, not template hook failure (issue #461) — bash scripts/iwe-audit.sh shows current available memory level in the "Available Memory" section.
Getting Started
Quick Start (Git, Node.js, Claude Code already installed): QUICK-START.md -- 15 minutes until first session.
Full Installation from a clean computer: SETUP-GUIDE.md -- 30-60 minutes including all dependency installations. Not on macOS or not using Claude Code? Read PORTABILITY.md — instructions for Kimi Code, Hermes Agent, and others.
A different agent or LLM? IWE is not tied to Claude. If your agent can see files in the repo folder and edit files, it will work. How to connect → PORTABILITY.md.
mkdir -p ~/IWE && cd ~/IWE
gh repo fork TserenTserenov/FMT-exocortex-template --clone
cd FMT-exocortex-template
bash setup.shAfter installation:
cd ~/IWE
claudeThen tell Claude: «Проведём первую стратегическую сессию» to walk through goal setting, the first plan, and environment configuration.
Customization
IWE is updated as a distribution. Platform updates arrive without losing personal settings.
Extensions (extensions/) add blocks to protocols:
# Добавить рефлексию в конце дня
echo "## Рефлексия дня
- Что было сложным?
- Что бы сделал иначе?
- За что похвалить?" > extensions/day-close.after.mdParameters (params.yaml) — enable/disable protocol steps:
reflection_enabled: true # Включить рефлексию
video_check: false # Отключить проверку видео
multiplier_enabled: true # Мультипликатор IWEUpdates — bash update.sh updates the platform while preserving your extensions/, params.yaml, and edits to CLAUDE.md (3-way merge).
More details: extensions/README.md
Documentation
| Document | Contents |
|---|---|
| Onboarding Guide | Start here if you are new to IWE. What it is, why it exists, and its components -- without technical jargon |
| Quick Start | 15 minutes from git clone to your first session. For those who already have Git and Claude Code |
| SETUP-GUIDE.md | Step-by-step installation from a clean computer. Requirements, modes (core/full), verification |
| LEARNING-PATH.md | The IWE learning path: architecture, principles, protocols, Pack, roles |
| DATA-POLICY.md | Data policy: what is collected, where it is stored, how to delete |
| DATA-RESIDENCY.md | Residency principle: data you bring into IWE from outside (health, calendar, work hours) -- where it can and cannot go |
| IWE-HELP.md | Quick reference and FAQ |
| principles-vs-skills.md | Why principles matter more than skills: the generative hierarchy |
| ONTOLOGY.md | Ontology: all IWE terms and abbreviations |
| CHANGELOG.md | History of template changes |
Two documents cover related topics:
DATA-POLICY.mdcovers data the platform collects about you;DATA-RESIDENCY.mdcovers data you yourself bring into IWE from outside.
FAQ
Q: Is an Anthropic subscription required?
A: For full installation (Claude Code), Claude Pro ($20/month) is recommended. If needed, you can upgrade to Claude Max (~$100/month) for unrestricted work. For minimal installation (setup.sh --core), it works with any AI CLI. More details: SETUP-GUIDE.md.
Q: Does it work with other AIs (not Claude)? A: Yes, three agents are supported out of the box:
- Claude Code — full support: reads
CLAUDE.md, all skills and hooks work. - Kimi Code (VS Code) — automatically reads
AGENTS.mdwhen opening the repo. Customization:extensions/orAGENTS-agent-blocks.md. Skills (/day-open, etc.) via Claude Code. - Hermes Agent — connect Aisystant MCP through Hermes settings, and it will receive instructions automatically.
For other agents (Cursor, Copilot, Gemini), adaptation is required. More details: PORTABILITY.md.
The minimal installation (setup.sh --core) works without binding to a specific agent.
Q: Does it work on Linux/Windows?
A: Standard scenarios work on macOS and Linux; the Strategist automation uses launchd and systemd respectively, with the cloud variant using GitHub Actions. On Windows, individual parts have been verified via Git Bash with native Python (Strategist lock, #1030), but session-guard.sh and session opening require WSL2: in native Git Bash, they fail beforehand due to Unix fcntl locking (#1032). Full installation and updating of IWE on Windows remain unverified — more details: SETUP-GUIDE.md § Windows.
Q: What if the computer is turned off or sleeping — will automation stop?
A: Cloud Scheduler (GitHub Actions) runs in the cloud even when your computer is turned off. For local agents: scripts automatically prevent sleep during operation (macOS: caffeinate, Linux: systemd-inhibit). For laptops, it is recommended to configure automatic wake-up and disable idle sleep — see SETUP-GUIDE.md.
Q: What is a Pack? A: A formalized domain of knowledge — the single source-of-truth for domain knowledge. Details: LEARNING-PATH.md.
Q: Are my data safe? A: Three protection zones: local, GitHub (private repos), platform (per-user isolation). Details: DATA-POLICY.md.
Q: How does IWE differ from Obsidian / Notion / Logseq?
A: Obsidian is a note storage tool. IWE is a work environment with protocols, AI agents, and knowledge formalization. For notes, you can open a separate governance repository (DS-strategy) in Obsidian as a vault. The root of IWE (~/IWE) is not supported as an Obsidian vault because it may contain very large Markdown files (e.g., FPF/FPF-Spec.md), which can cause Obsidian to display a white screen. The entire workspace can be safely viewed via VS Code.
Q: Do I need to know programming? A: No. The template is a ready-made configuration. Installation is done via setup.sh. Work is conducted through Claude Code using natural language.
Q: Can I use it without the Strategist? A: Yes. Claude Code + CLAUDE.md + memory/ work fully. The Strategist automates planning. Without it, you plan manually.
Q: How do I set up the strategy day?
A: In memory/day-rhythm-config.yaml, change strategy_day: sunday to your desired day. Details: LEARNING-PATH.md.
Q: Did cloning end up in ~ instead of ~/IWE?
A: All installation commands must be executed in the same terminal session. If you opened a new one, it starts from ~. Delete the folder from ~ and repeat with cd ~/IWE. Details: SETUP-GUIDE.md.
Full FAQ with detailed answers: DP.IWE.002 §11. Practical handbook: LEARNING-PATH §11.
IWE Community
IWE is an environment you build alone. But you develop it together.
The IWE Community is a closed chat of practitioners who work within the same system: ORZ, Pack, exocortex. It is a place where they discuss not "how to prompt better," but how to seriously build intellectual work.
What happens there
- Work product reviews — participants show real Packs, plans, retrospectives. They receive feedback from people who understand what "closing without recording results" means.
- Experience with installing and customizing IWE — what broke, how it was fixed, which extensions proved useful.
- Discussion of methods — ORZ fractal, ArchGate, Capture-to-Pack in practice: what works, where theory diverges from reality.
- Links and findings — tools, patterns, SOTA that fit into the IWE philosophy.
Why this matters
Studying the system alone is possible. But most questions arise at the application stage: "How to formalize this domain of knowledge?", "Am I using ORZ correctly?", "Who has experience with this tool?"
In the community, these questions get answered by people who have already gone through this.
Free channels
- GitHub Discussions — questions, ideas, show your setup
- Issues — bug reports and feature requests
Closed Community (Telegram)
Deep practice, reviews of working products, direct support. Access via the seminar «IWE for Practitioners» (5000₽) in the bot @aist_me_bot.
Contributing
See CONTRIBUTING.md — how to help the project.
For IWE team developers (T4+ level): single entry point — Developer Getting Started Guide. In 10 minutes you will understand the development pipeline (6 stations, double output) and complete your first task.
License
MIT
Source and license
Source · License · 415d61a98cd2dc471480b3ad01ece6648d814181
MIT License
Copyright (c) 2026 Tseren Tserenov
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.