rocm-cli

rocm-cli is the ROCm AI Command Center CLI for AMD systems.

Install on Linux x86_64:

curl -fsSL https://raw.githubusercontent.com/powderluv/rocm-cli/main/install.sh | sh

The installer updates your shell profile to add the install directory to PATH when needed.

Re-running the installer replaces the previous rocm-cli binaries from the same install directory.

Install the latest nightly on Linux x86_64:

curl -fsSL https://raw.githubusercontent.com/powderluv/rocm-cli/main/install.sh | sh -s -- nightly

Current repository status:

• interactive rocm TUI plus non-interactive CLI commands for doctor, serve, engines, install, update, automations, daemon, and uninstall
• rocmd supervisor with managed service supervision and persisted automation/watcher runtime state
• first-party pytorch engine with managed venv installs, TheRock PyTorch wheel resolution, and OpenAI-compatible local serving
• TheRock SDK resolver for pip and tarball installs, plus managed runtime update checks
• vendored Codex TUI scaffold with a packaged rocm-codex binary for experimental interactive launch

Planned product shape:

• rocm chat-first TUI with deeper inline tool execution and approvals
• TheRock-managed runtime installs with pip venvs by default and tarballs as an explicit option
• native automations and watchers with broader contained sandbox execution
• pluggable serving engines including pytorch, llama.cpp, vllm, sglang, and atom

Workspace layout:

• apps/rocm: main CLI and future TUI
• apps/rocmd: local supervisor and daemon entrypoint
• crates/rocm-core: app paths, host summary, shared defaults
• crates/rocm-engine-protocol: shared engine request and response types
• engines/pytorch: first-party PyTorch serving engine

Planning docs:

• plans/rocm-cli-implementation-plan.md
• plans/rocm-cli-pytorch-engine-spec.md

This is an early implementation, not a production release.