AI-Cortex

An agentic CLI coding assistant powered entirely by local LLMs via Ollama.

AI-Cortex gives you an agentic coding assistant that runs 100% locally — no API keys, no cloud, full privacy. It uses a local LLM (default: qwen3.5:9b) and an agentic loop that decides which tools to call, executes them, and feeds results back to the model until the task is done.

Features

• Agentic loop — model autonomously decides which tools to call and chains them together
• 6 built-in tools — read_file, write_file, run_bash, list_dir, search_files, fetch_url
• Streaming output — responses appear token by token
• Permission prompts — asks before running bash commands or writing files
• Loop detection — catches the model repeating the same action
• Error handling — retries on failures, catches malformed tool calls
• Configurable — model, host via ~/.ai-cortex/config.yaml or CLI flags

Prerequisites

• Ollama installed and running
• A local model pulled (e.g. ollama pull qwen3.5:9b)

Install

pip install ai-cortex

Or install from source:

git clone https://github.com/hasanjawad001/ai-cortex.git
cd ai-cortex
uv venv --python 3.12
source .venv/bin/activate
uv pip install -e .

Usage

# Start ai-cortex
ai-cortex

# Skip permission prompts
ai-cortex --yolo

# Use a different model
ai-cortex --model=llama3:8b

Example

you> read pyproject.toml and tell me the project name
  → read_file({"path": "pyproject.toml"})
  [project]
  name = "ai-cortex"
  ...
ai-cortex> The project name is "ai-cortex", version 0.1.0.

Configuration

AI-Cortex creates ~/.ai-cortex/config.yaml on first run:

model: qwen3.5:9b
host: http://localhost:11434

How It Works

1. You type a request
2. The LLM decides which tool(s) to call
3. AI-Cortex executes the tool and feeds the result back
4. The LLM either calls another tool or responds with a summary
5. Repeat until done (max 15 turns)

License

MIT