Relay

Relay is an open-source AI coding agent that runs in your terminal. It connects to LLMs through OpenRouter, reads and edits your code with a built-in tool set, runs shell commands, and streams everything through a full-screen TUI.

Functionality

Interface

Tools

Agent

Getting started

# 1. Install Relay
pip install relay-code

# 2. Log in (opens your browser to authorize with OpenRouter)
relay login
relay login --paste   # or paste an API key directly

# 3. Run it
relay                 # interactive mode
relay "your prompt"   # single-shot mode
relay --cwd /path     # run against a different working directory

# Remove the saved API key
relay logout

Project instructions (AGENTS.md)

Drop an AGENTS.md at the root of your repo and Relay loads it as developer instructions at startup, use it for build and test commands, code style, or anything else the agent should know before touching your code

Relay walks up from the working directory to the repository root, so running relay inside a subdirectory still picks up the root file. Every AGENTS.md found along the way is included, ordered outermost first.

# AGENTS.md

## Commands
- Test: `pytest`
- Lint: `ruff check .`

## Style
- Type hints on all public functions.

The scope of an AGENTS.md file is the directory tree it sits in, and the more deeply nested file wins on conflicts. Files below the working directory are read on demand as the agent works in those subdirectories. Instructions you give directly in a prompt always take precedence over AGENTS.md.

Roadmap

Currently being worked on:

• Slash commands: in-session commands (e.g. /approval to switch approval mode) instead of only config-file settings.
• Approval flow: prompting before file edits and shell commands is in place; still to come is switching the mode mid-session and extending confirmations to sub-agents.
• Session management: saving, resuming, and switching between sessions.

License

GNU GENERAL PUBLIC LICENSE v3.0