@mmmbuto/vl

vl — a featherweight terminal coding agent, home of the Vivling: a persistent AI companion that keeps identity, mood and goals across sessions, while a lean worker loop does the actual reading, editing and running.

This release exists to install vl and try the connections on your machines. It is not the finished product.

npm install -g @mmmbuto/vl
vl

Before you run it — three things that look like bugs and are not

1. A model must be configured. Without one, vl exits immediately with model is required. Either export VL_MODEL (and, if needed, VL_PROVIDER / VL_BASE_URL / VL_API_KEY), or create ~/.config/vl/config.toml:

   default_profile = "ollama"
   
   [profiles.ollama]
   provider = "openai-compat"
   base_url = "http://localhost:11434"
   model    = "qwen2.5-coder"

2. A real terminal is required. The TUI probes the terminal (cursor position report, ESC[6n) and exits if nothing answers. Inside a pipe, a CI job or a non-interactive wrapper vl will not start — by design, not by accident.

3. The package bundles every supported binary: ~3.5 MB to download, ~6.5 MB on disk after install. Static, self-contained builds for:

   | platform | binary | |---|---| | Linux x86_64 | x86_64-unknown-linux-musl | | Linux arm64 | aarch64-unknown-linux-musl | | Termux (Android arm64) | aarch64-unknown-linux-musl (same static build; not yet verified on a real Termux device) |

   macOS is not included in this release — a scope choice, not a defect; the launcher says so explicitly. The launcher picks the right binary from process.platform/process.arch and runs it; the other just sits in node_modules. Nothing is downloaded at install time.

Useful commands

vl                      # start the TUI (default profile)
vl -p <name>            # start with a named profile
vl shutdown             # stop the detached session host
vl node pair|run|status # micro-device node (outbound-only supervisor)

License

Apache-2.0