• RouterBench cli •

The terminal
AI coding agent.

Write and run code with any model, from your terminal — routed through the RouterBench intelligence layer, so cheap models handle simple work and strong ones handle the hard parts.

Read the docs

$npm i -g @routerbench.com/cli

~/project — RouterBench

▊

• live session •

Watch the router work.

Every prompt is classified, scored and routed — the CLI shows you which model won the task and what it cost. No black box.

• commands •

Your whole workflow, six commands.

terminal — RouterBench

$ RouterBench  # launch the interactive TUI in the current project
$ RouterBench run "add tests for src/auth.ts"  # one-shot, non-interactive
$ RouterBench run -c  # continue the last session
$ RouterBench models  # list available models
$ RouterBench serve  # headless server (HTTP API)
$ RouterBench web  # open the web UI
▊

• capabilities •

An agent, not an autocomplete.

01 — agents

Build & Plan modes

A full-access build agent and a read-only plan agent — press Tab to switch mid-session.

02 — routing

routerbench/auto built in

Every request goes through the smart router — the right model per task, automatically.

Learn more

03 — commands

Custom slash commands

Drop a Markdown file in ~/.config/routerbench/commands and run it as /your-command.

04 — extend

Plugins & MCP

Extend with hooks, custom tools and Model Context Protocol servers via routerbench.json.

05 — providers

Any endpoint

Works with any OpenAI-compatible provider — RouterBench, your keys, or your own models.

06 — everywhere

Same agent, 3 surfaces

The identical agent ships as a VS Code extension and a desktop app — one account, one config.

Learn more

Your terminal just got smarter.

Install the CLI and let RouterBench pick the right model for every task.

Get started All surfaces