Claude CLI
The model where the work already is: the terminal.
Every tool I use daily eventually gets rebuilt, because the version I build fits my hands. This one is a command-line client for Claude, written directly against the Anthropic SDK.
The brief
A chat window is a fine place to have a conversation and a terrible place to do work. Real work is a pipeline: files in the current directory, output that gets piped somewhere, a shell history that remembers what happened. Copy-pasting between a browser tab and a terminal is the tax you pay for using someone else's interface.
The build
Node and TypeScript on the official Anthropic SDK. Streaming from the first token so long answers are readable while they arrive, model selection at the call site, and the working directory as the default context. Small on purpose — it is a client, not a framework.
The status
Internal tool, in development. It exists for the same reason a carpenter builds their own jig: the shop-bought one is fine, but it is not shaped like the job.
The scoreboard
- Model access where the work already lives: the terminal
- Built on the official Anthropic SDK, streaming-first
- Deliberately small — a client, not a framework
Want a story like this on your side?
Two lines about your problem is enough to start.