AJax Code

Terminal-based AI coding agent — 100% local, no API keys needed.

AJax Code runs entirely on your machine using Ollama. No cloud, no subscriptions, no data leaving your computer.

Install

Linux / macOS

curl -fsSL https://raw.githubusercontent.com/aaravj/aj-code/main/install.sh | bash

Windows (PowerShell)

irm https://raw.githubusercontent.com/aaravj/aj-code/main/install.ps1 | iex

Manual

git clone https://github.com/aaravj/aj-code
cd aj-code
bash install.sh

Quick Start

aj-code

That's it. Ajax will detect your project and start helping.

Features

• 4-tier model routing — auto-selects the right model for task complexity
• Autonomous tool use — reads, writes, edits files; runs git, terminal, web search
• Self-healing — retries failed operations up to 5 times automatically
• Rolling context — summarises old messages so you never hit limits
• Vector memory — remembers past conversations and project decisions
• Session memory — SQLite-backed, persists across restarts

Models

Tier	Model	RAM Required	Best For
Tier 1	qwen2.5-coder:1.5b	< 6 GB	Quick fixes, simple questions
Tier 2	deepseek-coder:6.7b	6–15 GB	Features, bug fixes
Tier 3	qwen2.5-coder:14b	16–31 GB	Complex modules, architecture
Tier 4	qwen2.5-coder:32b	32 GB+	Full apps, system design

Commands

Command	Description
/help	Show all commands
/model	Show installed models
/context	Context usage %
/project	Project summary
/undo	Undo last file change
/diff	Show session changes
/commit	AI-generated git commit
/exit	Exit

Requirements

• Python 3.8+
• Ollama
• 4 GB+ RAM (8 GB recommended)
• Linux, macOS, or Windows

License

MIT © Aarav J