Get started with the CLI

The command line is the most direct way to install Hara, evaluate a form, run a source file, and keep a REPL open while you work. Use this path when you want a small local setup, a scriptable workflow, or a runtime that can later be connected to an editor.

Install Hara

Linux on x86_64 and macOS on arm64 or x86_64 can use the installer:

curl -fsSL https://www.hara-lang.org/install.sh | sh -- --rust --truffle

The installer places hara and the native-image hara-truffle in ~/.local/bin. GitHub Releases is the publishing authority for the downloaded packages and checksums.

Install only the Rust runtime:

curl -fsSL https://www.hara-lang.org/install.sh | sh -- --rust

Install only the Truffle native image:

curl -fsSL https://www.hara-lang.org/install.sh | sh -- --truffle

Use HARA_INSTALL_DIR to choose another destination, or set HARA_VERSION to pin a release:

HARA_INSTALL_DIR="$HOME/bin" \
HARA_VERSION=v0.1.0 \
  curl -fsSL https://www.hara-lang.org/install.sh | sh -- --rust

Make sure the installation directory is on PATH, then check that the command is available:

hara --help

Evaluate one form

Run a complete Hara form with eval:

hara eval '(let [x 19] (+ x 23))'

Expected result:

42

This is useful for shell scripts, smoke tests, and checking a small expression without opening an interactive session. Quote the form so the shell does not interpret its parentheses, spaces, or symbols.

Try a persistent collection:

hara eval '(assoc {:name "Hara"} :ready true)'

Then change one part of the form and run it again. The quickest learning loop is predict, run, change, and explain.

Start the REPL

Start an interactive ROOT session:

hara

Start the REPL without a RESP listener:

hara --offline

The REPL is the normal place to explore a namespace one form at a time. It supports multiline forms, persistent history, slash and symbol completion, inline documentation, and listener control through /resp.

Use the REPL for experiments, but keep durable work in .hal files. The live session is feedback; the source tree remains the project record.

Run a source file

Create hello.hal:

(ns hello)

(defn greeting [name]
  (str "Hello, " name))

(greeting "Hara")

Run it:

hara run hello.hal

You can also pipe source through standard input:

cat hello.hal | hara stdin

The stdin path is useful for editor integrations and generated source because the source does not need to be written to a temporary file first.

Start a project directory

A Hara project keeps executable source and workspace data together:

my-project/
  project.edn
  workspace.edn
  src/
    app.hal

Use .hal for executable Hara source. project.edn describes project roots and requested capabilities. workspace.edn describes the files, areas, nodes, and connections used by visual workspaces.

A useful first routine is:

1. Open a terminal in the project root.
2. Start hara and leave the session running.
3. Edit a .hal file.
4. Evaluate a small form while the program remains alive.
5. Keep the useful change in source and run the file again.

Choose a runtime deliberately

The Rust CLI and Truffle runtime implement the same portable Hara language surface, but they connect to different host environments.

Use the Rust runtime for a small native CLI, server work, and the runtime core shared with WebAssembly. Use the Truffle runtime when you need the JVM provider, Java classes, Maven-based contributor tests, or the GraalVM/Truffle host.

Portable code should not depend on a host merely because it runs on that host. Files, sockets, Java reflection, browser APIs, and other host operations cross explicit provider and capability boundaries.

Connect an editor later

The CLI can remain the runtime while an editor becomes the source surface. Hara VS Code connects to the listener exposed by the running server. Emacs can send a region through hara stdin. Both approaches preserve the same project and source files.

Troubleshooting

If hara is not found, confirm the installation directory is on PATH:

printf '%s\n' "$PATH"
ls -la "$HOME/.local/bin"

If an editor cannot connect, start the ordinary hara command rather than hara --offline, then inspect the listener status in the REPL. The standard editor endpoint is 127.0.0.1:4164, although a runtime or deployment may use a different address.

If host operations return unsupported or denied, distinguish two cases: unsupported means the current host has no provider for the operation; denied means the provider exists but the session has not been granted the required capability.

Continue

Read the user guide for the language surface, the REPL reference for interactive behavior, and projects and visual workspaces for the shared project model.