I built Agent Launch because I kept switching between local coding-agent CLIs and forgetting which flags each one used for prompts, working directories, modes, model choices, resume, and non-interactive runs.
Agent Launch gives me one command, agl, with a consistent interface for:
- choosing the backend agent
- passing a prompt
- setting the working directory
- choosing interactive or headless mode
- selecting fast/pro model classes
- resuming or continuing sessions
- dry-running the translated backend command
It currently supports Codex, Claude Code, Cursor Agent, Gemini CLI, OpenCode, Antigravity, and random selection across backends.
Example:
agl -a codex -C ~/Code/app --mode auto "fix the failing tests"
The goal is small: make switching between local AI coding agents boring. Same workflow, same muscle memory, backend-specific command generated underneath.
Top comments (0)