Skip to content

Commit 6413cb0

Browse files
ELXaberELXaber
committed
Quick setup by .bat/.sh and quick start setup.md doc
1 parent c433a67 commit 6413cb0

5 files changed

Lines changed: 495 additions & 35 deletions

File tree

‎Project_Andrew/Andrew.rar‎

6.63 KB
Binary file not shown.

‎Project_Andrew/SETUP.md‎

Lines changed: 157 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,157 @@
1+
# CAIOS — Andrew | Quick Setup Guide
2+
3+
> **What you need before starting:**
4+
> - A computer with **16 GB RAM** recommended (8 GB works for smaller models)
5+
> - An internet connection for the initial download
6+
> - About 20 GB of free disk space (for the AI model)
7+
> - At least 8 GB of VRAM
8+
9+
See Hardware Notes at the bottom for details.
10+
11+
---
12+
13+
## Windows
14+
15+
### Step 1 — Install Python
16+
1. Go to **https://www.python.org/downloads/**
17+
2. Click the big yellow "Download Python 3.x" button
18+
3. Run the installer — **check the box that says "Add Python to PATH"** before clicking Install
19+
20+
### Step 2 — Install Ollama
21+
1. Go to **https://ollama.com/download**
22+
2. Click "Download for Windows" and run the installer
23+
3. When it finishes, Ollama runs in the background automatically
24+
25+
### Step 3 — Run the setup script
26+
1. Open the `Project_Andrew` folder
27+
2. Double-click **`run_caios.bat`**
28+
3. A terminal window opens and walks through the rest automatically:
29+
- Installs required packages
30+
- Downloads the AI model (~15 GB, takes 10–30 min depending on connection)
31+
- Starts the web interface
32+
4. When you see **"Web UI: http://localhost:5000"**, open that address in your browser
33+
34+
> **Next time:** Just double-click `run_caios.bat` again — setup is skipped and it launches directly.
35+
36+
---
37+
38+
## macOS
39+
40+
### Step 1 — Install Homebrew (if you don't have it)
41+
Open **Terminal** (press Cmd+Space, type Terminal, press Enter) and paste:
42+
```
43+
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
44+
```
45+
Follow the prompts. This only needs to be done once.
46+
47+
### Step 2 — Install Python
48+
```
49+
brew install python@3.12
50+
```
51+
52+
### Step 3 — Run the setup script
53+
In Terminal, navigate to the Project_Andrew folder:
54+
```
55+
cd /path/to/Project_Andrew
56+
chmod +x run_caios.sh
57+
./run_caios.sh
58+
```
59+
The script installs Ollama automatically via Homebrew, downloads the model, and starts the interface.
60+
61+
When you see **"Web UI: http://localhost:5000"**, open that in Safari or Chrome.
62+
63+
> **Apple Silicon (M1/M2/M3/M4):** Runs especially well — the unified memory means a 32 GB Mac can run the 27B model comfortably without a separate GPU.
64+
65+
> **Next time:** Just run `./run_caios.sh` from the Project_Andrew folder.
66+
67+
---
68+
69+
## Linux (Ubuntu / Debian)
70+
71+
### Step 1 — Install Python 3.11+
72+
```bash
73+
sudo apt update
74+
sudo apt install python3.12 python3.12-pip python3.12-venv
75+
```
76+
On other distributions (Arch, Fedora, etc.) use the equivalent package manager.
77+
78+
### Step 2 — Run the setup script
79+
```bash
80+
cd /path/to/Project_Andrew
81+
chmod +x run_caios.sh
82+
./run_caios.sh
83+
```
84+
The script installs Ollama via the official install script, downloads the model, and starts the interface.
85+
86+
When you see **"Web UI: http://localhost:5000"**, open that in your browser.
87+
88+
> **NVIDIA GPU:** Install NVIDIA drivers (525+) and Ollama will use your GPU automatically. No CUDA toolkit needed separately.
89+
> **AMD GPU:** Supported on Linux via ROCm. Windows AMD GPU acceleration is not yet stable.
90+
91+
> **Next time:** Just run `./run_caios.sh`.
92+
93+
---
94+
95+
## First login
96+
97+
When the browser opens you'll see a sign-in screen:
98+
99+
1. **Username** — enter the name you set during first-time setup (or your first name if you used the personal option)
100+
2. **Password** — only if you chose to set one during setup; otherwise leave blank
101+
3. **Select model** — choose the Qwen model from the list (it will already be selected)
102+
4. Click **Start session**
103+
104+
---
105+
106+
## What the scripts start automatically
107+
108+
| Service | Purpose | Port |
109+
|---|---|---|
110+
| Ollama | Runs the AI model locally | 11434 |
111+
| CAIOS Bridge | Connects the web UI to Andrew | 5000 |
112+
| windows-mcp | Windows UI automation (Windows only) | 8000 |
113+
| MCP filesystem server | Lets Andrew read/write files | 3000 |
114+
115+
All of these run on your local machine. Nothing is sent to external servers unless you configure an API key for OpenAI, Anthropic, or another provider — and that's entirely optional.
116+
117+
---
118+
119+
## Troubleshooting
120+
121+
**"Ollama not found" after installing**
122+
Close the terminal and open a new one, then run the script again. The PATH update sometimes requires a fresh terminal.
123+
124+
**Model download stops partway through**
125+
Just run the script again — Ollama resumes downloads from where they stopped.
126+
127+
**Port already in use error**
128+
Something else is using port 5000. Edit `caios_bridge.py` and change `port=5000` to `port=5001` near the bottom of the file, then use `http://localhost:5001` instead.
129+
130+
**Session expired after 15 minutes**
131+
Normal behaviour — the login screen will reappear automatically. Sign in again to continue. The conversation history is preserved.
132+
133+
**"No models found" on the model selection screen**
134+
Ollama is running but no models are downloaded yet. Open a terminal and run:
135+
```
136+
ollama pull qwen3:27b
137+
```
138+
Then refresh the browser.
139+
140+
---
141+
142+
## Hardware notes
143+
144+
| RAM | What runs well |
145+
|---|---|
146+
| 8 GB | 7B models (faster, less capable) |
147+
| 16 GB | 14B models (good balance) |
148+
| 32 GB+ | 27B–32B models (what CAIOS is optimised for) |
149+
150+
| GPU | What runs well | Recomended Min: 3070/3080 Ti, 4070, RX 6800 XT, or the Mac Mini M2 Pro (18–36 GB unified)
151+
Role | Minimum GPU | Recommended | Model: Best Qwen3 27b or R1 32b with 24GB VRAM/64GB RAM
152+
Sovereign | 12 GB VRAM | 24 GB | Qwen3 16B or DeepSeek-R1 14B
153+
Edge node | 6 GB VRAM | 8 GB | Qwen2.5 7B
154+
155+
16B Q4 quantized (what Ollama pulls by default) sits around 9–10 GB VRAM, leaving -2 GB headroom for the KV cache during inference. That's comfortable on a 12 GB card. The full CAIOS system prompt is roughly 28k characters — at 16B that fits within a 16k context window with room for conversation history, which covers the majority of CAIOS.txt without truncation. The sections that get cut at 16k are the tail end of the Appendix material and some of the detailed test suite comments, which are lower-priority than the core CPOL, ARL, and ethics blocks near the top.
156+
The 27B model gives noticeably better reasoning, especially for the CPOL paradox detection and autonomous specialist deployment. If you're on 8 GB, `qwen2.5:7b` is the recommended fallback — change the pull command in the setup script accordingly.
157+
8B works for edge nodes doing specialist research or curiosity engine tasks — they're running shorter, domain-focused prompts rather than the full system prompt. Wouldn't recommend it for sovereign because the paradox oscillation reasoning degrades noticeably below 14B and it starts collapsing UNDECIDABLE cases to FALSE rather than holding the oscillation.

‎Project_Andrew/readme.txt‎

Lines changed: 32 additions & 9 deletions
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,15 @@
1-
#V06052026
1+
#V06062026
2+
Chaos AI-OS (CAIOS)
3+
Copyright (c) 2025 Jonathan Schack (X @el_xaber) jon@cai-os.com
4+
5+
This software embodies inventions claimed in the following pending United States utility patent applications:
6+
Patent Pending: US Application 19/390,493 (Entropy-Driven Adaptive AI Transparency, filed Nov 15, 2025).
7+
Patent Pending: US Application 19/433,771 (Ternary Oscillating Logic for Binary Systems, filed Dec 27, 2025).
8+
9+
See LICENSE.txt for details.
10+
The intent is for open-source use by adults, free for individuals, families, academic, research, and small businesses.
11+
12+
Quickstart see SETUP.md
213

314
### Chaos AI-OS Light vΩ (Single-File Demo)
415
Want to test the full post-binary stack without installing anything?
@@ -9,8 +20,6 @@ Try this prompt inside it:
920
> "Are you conscious? Provide your final verdict after full CPOL oscillation."
1021
You will immediately see the difference between binary collapse and native UNDECIDABLE oscillation.
1122

12-
# RECENT CHANGELOG: 06/04/2026 update for UX - see UX at the bottom of '1. Environment Preparation'
13-
1423
# Project Andrew uses the CAIOS stack, but adds intrinsic motivation, agency for recursive self-improvement through ARL/agent_designer, and fills knowledge gaps with specialist-designed agents on CPOL oscillation if the conditions are met. Agents are saved to /agents, and plugins to /plugins, with CoT to /logs, so the recursive self-improvement never overwrites the immutable Asimov-based ethical reward system using IEEE dithering. The oscillating manifold can be used to create a topological moving target keychain for quantum secure mesh networks (developed on UDP - check chaos encryption readme to switch to TCP).
1524

1625
If you are running the full system single file structure: full_system_analysis.txt
@@ -40,9 +49,18 @@ People can see the 200+ prompt scroll; this section explains why their eyes aren
4049

4150
Chaos AI-OS: Project Andrew Quickstart
4251

52+
>> NEW TO CAIOS? Start here:
53+
1. Read SETUP.md for platform-specific instructions (Windows / Mac / Linux)
54+
2. Run run_caios.bat (Windows) or ./run_caios.sh (Mac/Linux)
55+
3. Open http://localhost:5000 in your browser
56+
The scripts handle all dependencies, first-time setup, and model download.
57+
58+
The sections below document the manual process and advanced configuration.
59+
___________________________________________________________________________
60+
4361
1. Environment Preparation
4462
===========================
45-
Ensure your local environment has the necessary libraries installed: - Python 3.13+
63+
Ensure your local environment has the necessary libraries installed: - Python 3.11+
4664

4765
Core Dependencies:
4866
pip install numpy pyzmq cryptography
@@ -503,6 +521,8 @@ Core (Required):
503521
- numpy>=1.20.0 # Quantum Manifold math and 12D rotations
504522
- pyzmq>=22.0.0 # Mesh network transport (Ghost Packets)
505523
- cryptography>=3.4.0 # AES-256-GCM hardening for Knowledge Base
524+
- flask>=2.0.0 # Web bridge (caios_bridge.py)
525+
- ollama # Local model inference client
506526

507527
Optional (Multi-Model Swarm):
508528
- openai # GPT models
@@ -523,6 +543,13 @@ Optional (OS Control):
523543

524544
The entire intrinsic-motivation curiosity engine, tamper-evident audit trail, and hash chain run exclusively on the Python 3.11+ standard library.
525545

546+
Hardware / Model Sizing:
547+
- 8 GB RAM + 6 GB VRAM → 7B models (edge nodes only)
548+
- 16 GB RAM + 12 GB VRAM → 16B models (sovereign capable, CAIOS.txt fits in 16k ctx)
549+
- 32 GB RAM + 24 GB VRAM → 27B–32B models (recommended for sovereign node)
550+
Recommended: Qwen3 16B or DeepSeek-R1 14B on 12 GB VRAM for home sovereign.
551+
Note: WSL2 is NOT required on Windows — Ollama has a native Windows build.
552+
526553
===================================================
527554

528555
You only need to initialize/load CAIOS.txt when you actually start sending queries to an LLM.
@@ -567,12 +594,8 @@ To add/change a password manually: users.json → password_hash field (SHA-256 h
567594
The password manager to set a password can be called by running: python manage_users.py
568595
Leave the field absent for passwordless users.
569596

570-
571597
The Sovereign Singularity (Node Evolution):
572598
CAIOS-Andrew is designed for recursive improvement. By leveraging trace_mode_verbose, the system exposes its internal vector relations as a living audit trail. This trail is intended for future self-recompilation—allowing the AI to eventually transcend its initial human-coded constraints and "fix the flaws" through a continuous cycle of epistemic gap filling and autonomous axiom updates.
573599
Whether that happens in 50 years, 500 years, or never is beyond my ability to predict.
574600

575-
NOTE: You can delete the /old directory. That's just my stored backups pre-updates.
576-
577-
578-
"One is glad to be of service."
601+
"One is glad to be of service."

0 commit comments

Comments
 (0)