AISH ­– inline AI agent for devops

AISH is the world's first context-aware inline devops sh agent. It is local-first but works also with any commercial OpenAI compatible LLM. Delegate shell tasks to aish who lives in the terminal. You can do in-line chatting with aish. Just mention @aish and send a request and @aish will complete it for you.

This demo was created using Devstral-2505 24B Q4 running on vllama which is world's fastest drop-in replacement for ollama.

🚀 Features

• In-line Chatting: Tag @aish anywhere in your terminal session for instant assistance, and it works on any bash terminal including SSH
• Context Awareness: aish sees your bash session without initiating agentic sessions and can reference previous commands and outputs
• Interactive Command Execution: Approve, deny, or comment on proposed commands before execution
• Memory System: Persistent memory that learns your preferences and environment details across sessions
• Smart Model Selection: Automatic model discovery from your LLM endpoint during setup
• Rich Terminal UI: Beautiful, colorized output with panels and syntax highlighting
• Local-First Design: Works with local LLM runners (Ollama, vLLama) or commercial OpenAI compatible APIs
• Bash Session Logging: Optional real-time session recording for review and sharing
• Question Detection: Automatically detects when aish asks questions and provides interactive responses

📦 Installation

Prerequisites:

• Python 3.12+
• A running OpenAI-compatible LLM endpoint. aish is designed with a local-first approach, compatible with local LLM runners like vLLaMA and Ollama. It also works with OpenAI and any other OpenAI-compatible API.

Quick Install:

git clone https://github.com/erkkimon/aish ~/Software/aish
cd ~/Software/aish
bash setup.sh

The setup script will:

1. Create a Python virtual environment automatically
2. Install all required dependencies
3. Set up the @aish command in your ~/.bashrc
4. Discover available models from your endpoint and let you choose interactively
5. Create a user configuration file at ~/.aish/config.yaml

After installation, restart your terminal or run source ~/.bashrc to start using @aish.

🎯 Usage

Basic Usage

To run the agent, use the @aish alias with your command as arguments:

@aish show me all Python files in this directory

Context-Aware In-line Chatting

aish is context-aware and can see your recent terminal history. This enables powerful in-line conversations:

Example 1: Analyzing file permissions

$ ls -la
total 24
drwxr-xr-x  5 user user 4096 Dec 14 10:23 .
drwxr-xr-x 18 user user 4096 Dec 14 09:15 ..
-rw-r--r--  1 user user  220 Dec 14 10:20 config.yaml
-rwxr-xr-x  1 user user  450 Dec 14 10:15 setup.sh
-rw-r--r--  1 user user 1200 Dec 14 10:10 README.md

$ @aish do those file permissions look okay?
🤖 Plan: I'll check the file permissions from your ls output and analyze if they look correct.
🛠️ Command: echo "Analyzing permissions..."
Execute? (Y/n/c):

Example 2: Working with command output

$ find . -name "*.log" -size +10M
./logs/app.log
./logs/debug.log

$ @aish compress those large log files
🤖 Plan: I'll compress the large log files you found to save disk space.
🛠️ Command: gzip ./logs/app.log ./logs/debug.log
Execute? (Y/n/c): y
✓ Command executed successfully

Example 3: Debugging after errors

$ python script.py
Traceback (most recent call last):
  File "script.py", line 15, in <module>
    process_data()
  File "script.py", line 8, in process_data
    result = 10 / 0
ZeroDivisionError: division by zero

$ @aish fix that error
🤖 Plan: I can see the ZeroDivisionError in your script. Let me examine the file and fix the division by zero issue.
🛠️ Command: sed -i 's/10 \/ 0/10 \/ 1/g' script.py
Execute? (Y/n/c): c
💬 Comment: Actually, check what the correct divisor should be
🤖 Plan: You're right, let me first examine the script to understand the context...

Interactive Controls

When aish proposes a command, you can:

• y or Enter: Execute the command
• n: Stop the agent
• c: Provide a comment to revise the agent's plan

Memory Management

aish has persistent memory that stores system state - durable facts about your environment:

$ @aish remember that I prefer vim over nano
🧠 Memory Update: Added preference
✓ Memory updated successfully

$ @aish open that config file
🛠️ Command: vim config.yaml   # aish remembers your preference!

The memory stores current facts, not history:

• User preferences: "prefers vim, uses 4-space indentation"
• System configuration: "nginx installed, serves /var/www/html"
• Project conventions: "uses Python 3.11 with poetry"
• Custom setups: "backup script at ~/scripts/backup.sh"

Interactive Programs

aish properly handles interactive terminal programs. When you ask to open files in vim, less, htop, etc., they run with full terminal access:

$ @aish open the log file in vim
🛠️ Command: vim /var/log/app.log
Execute? (Y/n/c): y
# vim opens normally with full interactivity

Supported interactive programs: vim, nano, less, htop, man, ssh, tmux, and more.

⚙️ Configuration

aish is configured via a config.yaml file. The setup script automatically creates one for you at ~/.aish/config.yaml.

Configuration options:

• model: The name of the model to use at your endpoint
• endpoint_url: The URL for the chat completions API
• api_key: Optional API key if your service requires authentication

During setup, aish will:

1. Connect to your endpoint URL
2. Fetch available models automatically
3. Present an interactive menu to select your preferred model
4. Configure everything automatically

Supported Providers

• Local LLMs: Ollama, vLLaMA, LocalAI, and other OpenAI-compatible servers
• Commercial APIs: OpenAI, Kimi Code API, and other OpenAI-compatible providers

For Kimi Code API, use:

• Endpoint: https://api.kimi.com/coding/v1
• Model: kimi-for-coding
• API key from your Kimi membership page

📝 Bash Session Logging (Optional)

AISH includes optional bash session logging that creates a real-time carbon copy of your terminal session. This enables aish to be truly context-aware.

To enable bash session logging:

1. Add this line to your ~/.bashrc:

   source ~/Software/aish/bashrc_extension_aish.sh

2. Restart your terminal

The logging automatically creates session files in ~/.local/share/bash_sessions/ and cleans up files older than 7 days.

Note: The logging functionality is completely separate from the @aish command and can be used independently.

🔄 Updating aish

To update aish to the latest version:

cd ~/Software/aish
bash setup.sh  # This will update paths and configuration

Or manually:

cd ~/Software/aish
git pull
pip install -r requirements.txt

🛠️ How It Works

• setup.sh - Automated installation script that sets up everything including model discovery
• aish.py - Main AI assistant with rich terminal UI and memory management
• bashrc_extension_aish.sh - BASH session logging for context awareness
• Dynamic path resolution - Works from any installation directory

The system is designed to be completely portable and self-configuring.