UNIXSI – Under Development

▖▖▖ ▖▄▖▖▖▄▖▄▖
▌▌▛▖▌▐ ▚▘▚ ▐ 
▙▌▌▝▌▟▖▌▌▄▌▟▖
    UNIXSI – Agent Harness

Version 3.6 – A minimalist, single‑script bash harness that turns any local Ollama model into an autonomous conversational agent with an endless reflection mode, message queue, and built‑in tool‑calling.

---

Overview

UNIXSI is a self‑contained Bash script that designed to run off one bash entirely in your terminal. It connects to a local Ollama server and lets you chat with any model. It offers a unique Reflection Loop – a self‑dialogue where the SI continuously responds to its own previous messages, simulating a conversation between a user and an assistant. The loop can be started with any initial prompt and runs indefinitely until you interrupt it.

Agent capabilities (command suggestion via <cmd>...</cmd> tags) are always active, making it useful for system administration, development, or exploring model behavior.

---

Features

• Reflection Loop – /reflect on <prompt> starts an autonomous self‑dialogue: the SI replies to itself, alternating between User (Reflect) and SI roles. Perfect for brainstorming, idea refinement, or testing model reasoning.
• Message Queue – Conversation history is stored and automatically trimmed based on message count and character limits (configurable), helping smaller models stay within context window.
• Thinking Visualizer – A live elapsed‑time spinner shows how long the model is taking to respond.
• Agent Mode Always On – The model can propose shell commands using <cmd>...</cmd> tags; you are prompted to execute, skip, or run with sudo.
• Model Switching – List available models and switch at runtime with /model.
• Debugging – Toggle debug output with /debug on/off.
• Pure Bash – No external dependencies beyond curl and ollama (which the script will install if missing).
• Non‑interactive Fallback – When /dev/tty is unavailable, falls back to first model or UNIXSI_MODEL environment variable.

---

Requirements

• Linux / macOS / WSL with Bash 4+.
• Ollama – (Local Only) installed and running (the script will attempt to install and start it if missing).
• Network – local access to Ollama (default http://127.0.0.1:11434).

Optional (for better JSON parsing): jq or python3. If neither is present, a pure‑awk fallback is used.

---

Installation

UNIXSI is a single Bash block. To install:

Copy the entire file (bash << 'UNIXSI_SCRIPT_END' ...) block and paste it into your terminal – it will run immediately.

---

Usage

Start the script; it will prompt you to select a model. Then you can interact:

• Type any message to send it to the model.
• Use slash commands (see below) to control the session.

Reflection Loop (Self‑Dialogue)

To start the endless loop with an initial prompt:

/reflect on write a short story about a robot

This starts a 1:1 loop between the SI> agent and the User (Reflect)> agent. The loop continues until you:

• Type /reflect off
• Or type any normal message (which will break the loop and become your next user input)

Agent Commands

If the model includes a command like <cmd>ls -la</cmd>, you will be prompted:

>>> Proposed command: ls -la
Run it? [y/N/s=with sudo]

• y – execute immediately
• s – execute with sudo
• N (or anything else) – skip

---

Basic Commands

Command	Description
/reflect on [prompt]	Start self‑dialogue loop with optional initial prompt
/reflect off	Stop the loop and return to normal chat
/model	List available models and switch to a different one
/history	Show the current message queue (with roles and content)
/clear_history	Clear the conversation history
/tools	Show agent capabilities (command execution)
/clear	Clear the terminal screen
/debug on/off	Toggle debug logging
/help	Display this help
exit, quit, q	Exit UNIXSI

---

Advanced Configuration

All settings can be overridden before running the script:

Variable	Default	Description
OLLAMA_HOST	http://127.0.0.1:11434	Ollama server URL
UNIXSI_MODEL	(none)	Pre‑select a model; if not found, script exits
REFLECT_MAX_ITERATIONS	100000	Safety cap for loop iterations
MAX_HISTORY_MESSAGES	30	Max number of messages kept in queue
MAX_CONTEXT_CHARS	8000	Approximate character limit for the context
UNIXSI_TEMPERATURE	0.8	Sampling temperature for generation
UNIXSI_DEBUG	0	Set to 1 to enable debug output at start

Example:

UNIXSI_MODEL=llama3.2 REFLECT_MAX_ITERATIONS=500 ./unixsi.sh

---

Examples

1. Example: Start a reflection loop on a coding task

/reflect on write a bash function to check if a port is open

The SI will generate a function, then respond to its own output, improving and refining the code.

2. Example: Use the model for system administration (agent mode)

You> list all files in /tmp that are older than 7 days
SI> You can use: find /tmp -type f -mtime +7 <cmd>find /tmp -type f -mtime +7</cmd>
>>> Proposed command: find /tmp -type f -mtime +7
Run it? [y/N/s=with sudo] y
[output...]

3. Example: Switch models during a session

/model
Available models:
   1) llama3.2
   2) qwen2.5:0.5b
Enter number, model name, or 'q' to quit: 2
Switched to model: qwen2.5:0.5b

---

Troubleshooting

Issue	Solution
Ollama not found	The script will try to install it automatically. If that fails, install manually: curl -fsSL https://ollama.com/install.sh | sh
Failed to connect to Ollama	Ensure Ollama is running (ollama serve). Check OLLAMA_HOST if using a custom address.
Reflection loop stops or repeats	Increase MAX_HISTORY_MESSAGES and MAX_CONTEXT_CHARS to retain more context, or raise UNIXSI_TEMPERATURE for more variation.
Command execution not working	You must type y or s when prompted. Commands are run in the current shell; sudo may require a password.
JSON parse errors	Install jq for more reliable parsing: sudo apt install jq (Debian/Ubuntu) or brew install jq (macOS).

---

License

webXOS 2026 © MIT Open Sourced for Testing and Development