HexaLLM

AI coding assistant for the terminal — like Claude Code or Codex, but runs on your own local models via Ollama or a HexaLLM instance.

hexallm › explain the auth flow in this repo
hexallm › add input validation to src/api/users.py
hexallm › write tests for the payment module and run them

Install

pip install hexallm

Or from a downloaded wheel:

pip install hexallm-0.13.3-py3-none-any.whl

Requirements

• Python 3.10+
• Ollama running locally (ollama serve)
• A model: ollama pull llama3.2 or ollama pull codellama

Quick start

hexallm                   # start interactive session
hexallm -m codellama:7b   # use a specific model
hexallm models            # list available models
hexallm set model codellama:7b   # save default model

Session commands

Tools

Remote daemon mode

Connect to a HexaLLM server and let the web UI dispatch tasks to your machine:

hexallm login https://your-hexallm-server   # authenticate
hexallm daemon                              # start listening

The HexaLLM Chat UI will show a "CLI" badge — select your connected machine and the AI will execute tasks locally and stream results back.

Configuration

Stored at ~/.hexallm/config.json.

hexallm config                            # show config
hexallm set model codellama:13b           # default model
hexallm set max_steps 30                  # allow longer runs
hexallm set ollama_url http://host:11434  # remote Ollama