Multi-AI Session Management System

Keep Claude, Gemini, Codex, and DeepSeek in sync across sessions — automatically.

Every AI starts each session knowing exactly where you left off. Every AI ends each session leaving a clean handoff. One brutal critic agent tells you the truth when you need it.

Inspired by NetworkChuck's AI terminal workflow.

---

What This Does

The problem: AI tools start each session cold. You waste the first 10 minutes re-explaining context. When you use multiple AI tools (Claude for architecture, Gemini for research, DeepSeek locally), they have no idea what the others did.

The solution: Three personal Claude Code agents + a shared context file system + shell functions that load context automatically.

What	How
Claude briefs you when greeted	Greeting trigger in CLAUDE.md launches session-opener on "good morning" / "hi"
Gemini loads project context	Auto-reads GEMINI.md from your project root
Codex loads project context	Auto-reads AGENTS.md from your project root
DeepSeek/Ollama loads context	deepseek shell command injects DEEPSEEK.md as system prompt via Modelfile
End-of-session sync	@agent-session-closer updates all 4 files, commits to git, optionally pushes
Honest quality feedback	@agent-brutal-critic reviews your work with zero flattery

---

Prerequisites

• Claude Code CLI — required
• Ollama + a DeepSeek model — optional, only needed for DeepSeek support
• Git — for session commit/push feature

---

Installation

Windows (PowerShell)

git clone https://github.com/madmoneymike5/Multi-AI-Session-Management-System.git
cd Multi-AI-Session-Management-System
.\install.ps1

If you get a script execution error:

Set-ExecutionPolicy -ExecutionPolicy RemoteSigned -Scope CurrentUser

macOS / Linux / Git Bash (Windows)

git clone https://github.com/madmoneymike5/Multi-AI-Session-Management-System.git
cd Multi-AI-Session-Management-System
bash install.sh

Activating in your current terminal

The installer sets up your shell profile, but new functions only load when you open a new terminal. To activate immediately in your current window without reopening:

Shell	Command
PowerShell	. $PROFILE
Git Bash	source ~/.bash_profile
CMD	Not supported — open PowerShell instead

Note: . $PROFILE is a PowerShell command. It will fail if you run it in Git Bash. Make sure you're in the right shell before running these.

OneDrive users (common on Windows 11): If your Documents folder syncs to OneDrive, your PowerShell profile lives at C:\Users\[you]\OneDrive\Documents\PowerShell\... not C:\Users\[you]\Documents\.... The install script handles this automatically via $PROFILE — but if you're editing your profile manually, make sure you're editing the right file. Run $PROFILE in PowerShell to see the exact path.

What gets installed

~/.claude/agents/
  session-closer.md    ← end-of-day ritual
  session-opener.md    ← morning briefing
  brutal-critic.md     ← adversarial reviewer

~/.claude/commands/
  init-project.md      ← new project bootstrapper

~/.bashrc / PowerShell $PROFILE
  deepseek()           ← Ollama wrapper with auto-context loading

---

Updating

This system improves over time — better agent prompts, bug fixes, support for new AI tools. When a new version is available, here's how to get it.

Step 1: Pull the latest version

Navigate to the folder where you originally cloned this repo, then run:

PowerShell (Windows):

cd path\to\Multi-AI-Session-Management-System
git pull

Bash (macOS / Linux / Git Bash on Windows):

cd path/to/Multi-AI-Session-Management-System
git pull

Not sure where you cloned it? Search your computer for a folder called Multi-AI-Session-Management-System. You only need to clone it once — after that, just git pull from that folder whenever you want to update.

Step 2: Re-run the installer

PowerShell:

.\install.ps1

Bash:

bash install.sh

What gets updated — and what doesn't

The installer is designed to be safe to run more than once:

What	Updated?
Claude agents (session-closer, session-opener, brutal-critic)	✅ Always — replaced with the latest version
/init-project command	✅ Always — replaced with the latest version
deepseek shell function	✅ Updated if version is outdated; skipped if already current
Your project files (CLAUDE.md, GEMINI.md, etc.)	❌ Never touched
Your git history or any work in your projects	❌ Never touched

In other words: the system itself gets upgraded, but nothing inside your actual projects changes. Your work stays exactly as you left it.

Note: The brutal-critic personality library evolves over time. If your critic seems to be using outdated personalities (e.g., "Paranoid Architect", "Code Assassin", "Product Skeptic"), re-run the installer — agents are always replaced with the latest version on reinstall.

---

Usage

Starting a new project

In Claude Code, from your project directory:

/init-project

Creates CLAUDE.md, GEMINI.md, AGENTS.md, DEEPSEEK.md, docs/next-session.md, and the Claude memory directory. All AI tools are briefed from day one.

Private repos: AGENTS.md, GEMINI.md, and DEEPSEEK.md contain session history. If your repo is public and session details are sensitive, add these to .gitignore. CLAUDE.md is safe to commit — it contains only project guidance.

Adding to an existing project

Already have a project in progress? Run the same command:

/init-project

It's safe to run on a project you've already been working on. If a CLAUDE.md already exists, it won't be overwritten — the command will read it and use it as the source of truth. Any missing context files (GEMINI.md, AGENTS.md, DEEPSEEK.md, docs/next-session.md) will be created and populated from your existing CLAUDE.md. Think of it as "catch the other AIs up to where Claude already is."

Daily workflow

Morning (session start): Greet Claude — "good morning", "hi", "hey" — and the greeting trigger in CLAUDE.md will launch the session-opener agent to brief you. No briefing happens until you greet, so launches stay fast when you don't need a brief.

During the day: Work normally. Use whichever AI tool fits the task.

End of day (session close): Just tell Claude you're done — "let's wrap up", "close out the session", "end of day" all work. Or call it explicitly:

@agent-session-closer

Claude summarizes the session, updates all context files, commits to git, and asks whether you want to push to GitHub.

Starting Gemini or Codex: Just cd into your project and open the tool. They auto-load their context files.

Starting DeepSeek:

cd your-project
deepseek                      # uses deepseek-r1:8b by default
deepseek deepseek-r1:14b      # use any installed model by name

The model argument accepts any model name that ollama list shows. You must have pulled the model first (ollama pull modelname). The default is deepseek-r1:8b — change this by editing the param line in your shell profile if you want a different default.

When you want unfiltered feedback:

@agent-brutal-critic to roast what we did today

Detects your project type from an ever-growing library (currently 27 types, expanding via Self-Expansion Protocol), selects 5 tailored reviewer personalities, and borrows perspectives across types. Hard to please by design.

---

How the Context Sync Works

At end of each session, @agent-session-closer writes identical project state to four files:

• CLAUDE.md — gains a ## Session History entry and updated ## Current State
• GEMINI.md — Gemini CLI reads this automatically at session start
• AGENTS.md — Codex reads this automatically at session start
• DEEPSEEK.md — loaded as system prompt by the deepseek shell function

Any AI that does work updates its own file at session end. The next time you open Claude, the session-opener agent checks timestamps and flags if another AI updated context since Claude's last session.

Concurrent sessions

This system is designed around sequential use — one AI working at a time. If you need to run two AIs at the same time, here's how to stay sane:

Short-term awareness — WORKING.md: Each AI writes a brief note to WORKING.md in your project root when it starts a session, and clears it when done. The note includes which AI is working and what it's focused on. Claude's session-opener will warn you if it finds an active WORKING.md entry when you open a new session. This is a soft signal, not a hard lock — it tells you "someone else may be working" so you can coordinate manually.

File-level safety — git branches: WORKING.md can't actually prevent two AIs from editing the same file at the same time. The real solution is to put each AI on its own git branch. When both are done, merge the branches. Git will catch any overlapping changes and let you resolve them — the same way any two developers would work together on the same codebase.

Tip: If you're running Gemini and Claude at the same time, a clean split is: "Gemini owns research and docs on branch gemini/research, Claude owns implementation on main." They rarely touch the same files that way.

---

Repo Structure

agents/          → install to ~/.claude/agents/
commands/        → install to ~/.claude/commands/
shell/           → shell functions for bash and PowerShell
install.sh       → bash installer
install.ps1      → PowerShell installer

---

Manual Installation

If you prefer not to run the install script:

1. Copy agents/*.md to ~/.claude/agents/
2. Copy commands/*.md to ~/.claude/commands/
3. Add the function from shell/deepseek.sh to your ~/.bashrc, or shell/deepseek.ps1 to your PowerShell $PROFILE

---

Contributing

This system was built for personal use and shared as-is. If you improve it — better agent prompts, support for new AI tools, smarter install scripts — PRs are welcome.