Piper — DevOps copilot where the LLM picks typed actions from a fixed catalog rather than generating shell commands

Piper is a terminal-first, LLM-driven DevOps copilot where the LLM proposes actions from a typed catalog, deterministic code validates the choice, and humans approve anything that mutates infrastructure. The LLM never executes commands directly.

Instead of generating arbitrary shell strings, the LLM emits typed tool calls (e.g., `{ "name": "system.disk_usage", "args": { "host": "staging", "path": "/" } }`), and Piper's executor translates those into actual commands and runs them locally. Actions are audited in source code (src/actions/builtin/), and secrets are stripped before logging.

The product targets two personas: lone developers maintaining shipped apps without DevOps support, and DevOps engineers repeating the same diagnostic tasks repeatedly. Piper is designed to be non-autonomous — it does not act on mutate or destructive actions without approval.

Pre-built binaries are available for macOS (Apple Silicon) and Linux x64 (~76 MB). The wizard on first launch detects local LLM servers (Ollama, LM Studio, llama.cpp, vLLM) or accepts an OpenRouter API key, and allows selection among Featherweight (~$0.10/M), Economy (~$0.44/M), Balanced (~$3/M), and Premium ($30+/M) tiers.