jxhn

jxhn is a local-first AI agent harness written in Rust. It provides a terminal UI for chatting with hosted or local models while running workspace-scoped tools with explicit permissions.

Install

Requirements: Rust 1.85+ on Linux, macOS, or Windows.

From a repository checkout:

cargo install --path crates/jxhn-cli

Or run without installing:

cargo run --release -p jxhn-cli

Use

Start jxhn:

jxhn

The first launch opens provider setup. Configure it again with Ctrl+P, choose a model with Ctrl+M, use Shift+Enter for a newline, and quit with Ctrl+C.

Useful commands:

jxhn doctor
jxhn auth login chatgpt
jxhn model add /path/to/model.gguf
jxhn model list

Features

• OpenAI API keys and ChatGPT browser/device OAuth
• Local Ollama and Ollama Cloud
• Managed llama.cpp runtime for local GGUF models
• Automatic provider model discovery
• Streaming Markdown chat UI
• Workspace-scoped tools and permission prompts
• SQLite-backed sessions and configuration

More details: docs/providers.md · docs/security.md