Henri 🐕

This is an LLM coding assistant. Mostly an experiment to see if I could vibe code a vibe coding tool, and it turns out you can. Named after my Golden Retriever after hearing that AI coding agents were much like letting a Golden Retriever code.

There is no way this tool can keep up with the progress of Claude Code or OpenCode, but sometimes I just prefer its simplicity.

Henri remains simple. It executes one task at a time serially, providing verbose output so you can see exactly what's going on. It uses no sub-agents or background tasks. In some ways, I prefer this to what the more advanced agents are doing these days.

Features

Multi Provider / Multi Model

Built-in support for the following providers:

• GitHub Copilot
• Anthropic Claude Pro/Max
• OpenAI ChatGPT Pro
• Google Antigravity
• OpenCode Zen
• OpenAI Compatible APIs (like Z.ai)
• OpenRouter

Sandboxing

Henri has three security modes:

• Read-Write (RW): Tools and shell commands can write only inside the current directory (and children), plus /tmp. This is the default.
• Read-Only (RO): Tools and shell commands cannot write files at all.
• YOLO: No sandbox restrictions—tools and shell can write anywhere.

Switch modes with /read-write, /read-only, /yolo, or Ctrl+X to cycle through them. Start in read-only mode with henri cli --read-only.

Sandboxing uses Linux Landlock; on unsupported systems, restrictions are best-effort.

Installation

Currently, the only way to install is with Cargo:

cargo install --locked henri

This will install to ~/.cargo/bin. Ensure that directory is in your PATH, or move the binary after installation:

mv ~/.cargo/bin/henri ~/.local/bin/henri

Running

henri

On first start, you must configure a provider/model with the "/provider" command.

Adding a Provider

Henri supports multiple AI providers. Add one with:

henri provider add

To remove a provider:

henri provider remove

Configuration is stored in ~/.config/henri/config.toml.