Quickstart

This path uses Zen’s supported free hosted models so you can try a first question without an API key. Already have a model provider? Connect it instead, then use its name in place of zen below.

1. Install term-llm#

On macOS with Homebrew:

brew install samsaffron/tap/term-llm

Or use the release installer on a supported platform:

curl -fsSL https://raw.githubusercontent.com/samsaffron/term-llm/main/install.sh | sh

The shell command downloads and executes the installer script. Inspect it first if you prefer, or download a release archive. Published releases are self-contained: the browser interface is included, with no Node.js or frontend build required.

Confirm the binary is available:

term-llm version

You should see the installed version. If your shell cannot find term-llm, follow the installer’s PATH instructions or see installation.

2. Get a useful first answer#

term-llm ask --provider zen "Explain git rebase in three sentences"

You should receive a short explanation of how rebase replays commits onto a new base. Wording varies by model. This request asks a question; it does not run git rebase or change your repository.

Zen is a third-party hosted service. Your prompt is sent to it, and free model availability and limits can change. If the model is unavailable, run term-llm models --provider zen to inspect the current catalog, or choose another provider.

3. Choose your interface#

Both interfaces use the same underlying runtime and provider configuration. Try either one; run these commands separately.

Open the browser workspace#

term-llm serve web --provider zen

Open the URL printed in your terminal and follow the printed authentication instructions. Keep this terminal process running while you use the interface; press Ctrl+C to stop the server.

You get a complete workspace: saved conversations, project and worktree organization, model selection, tool activity, live file diffs, attachments, and an interactive shell. Some controls appear only when a project, attachment, or file change makes them relevant.

Explore the browser workspace. For remote access, read the authentication and deployment guidance before exposing the server. Do not disable authentication on a public listener.

Stay in your terminal#

term-llm chat --provider zen

Ask follow-up questions in a persistent terminal conversation. When you are ready to work with a repository, run this from its directory:

term-llm chat --provider zen @codebase

Try asking: “Where does this application handle configuration?” Review workspace access requests before granting them. The quality of agent work depends on the chosen model and its tool support.

4. Try a real workflow#

Choose a task that fits your work. These examples keep --provider zen explicit; substitute your preferred provider if needed.

Review staged changes without editing them:

term-llm ask --provider zen @reviewer "review the staged changes"

Run it in a Git repository with staged changes. The built-in reviewer is read-only and git-aware. Expect observations tied to the code, not automatic fixes.

Preview a targeted edit:

term-llm edit --provider zen "improve error handling" -f main.go --dry-run

Replace main.go with a file in your project. --dry-run previews the change without writing it to disk.

Choose and run a command in plain English:

term-llm exec --provider zen "list files"

You’ll see an interactive picker with suggested commands and an explanation for each. Use ↑/↓ to highlight an option, i for more information, and Enter to run the selected command. Choose “something else…” to refine your request, or Esc to cancel.

For this example, suggestions might include ls, ls -la, or ls -lh. Check the command before selecting it: Enter runs it, rather than just copying it. If you want to print the selected command instead of running it, add --print-only.

5. Make it your own#

• Save a default provider so you can omit --provider.
• Choose an agent for coding, review, research, or editing.
• Understand approval modes before giving agents broader tool access. Ordinary chat, ask, and web sessions default to Guardian-reviewed auto mode; use --approval prompt if you want human approval for unmatched actions.
• Browse workflows for jobs, MCP tools, media, and deployment.