For the last couple of weeks, as the cost of frontier models kept climbing, I started experimenting with open-source large language models.
What surprised me: for most of the tasks I'd normally hand to Claude or GPT, a local open-source model did the job just as well — and sometimes better.
The problem was the ergonomics. Ollama makes running local models genuinely easy, but I wanted a smoother terminal workflow — streaming chat that didn't feel clunky, safe file editing, and the ability to run real multi-file agent tasks without leaving my shell.
So I built Clive 🦀 — a local-first coding-assistant CLI powered by Ollama.
Everything runs on your machine. No prompts, code, or files ever leave your computer.
serve
, pull
, rm
, model recommendationsIt's written in Rust, MIT-licensed, and just hit its first release, v0.1.0
.
bash
git clone https://github.com/SedarOlmez94/clive.git
cd clive
cargo install --path .
clive ollama serve --detach
clive ollama pull qwen2.5-coder:latest
clive chat "Explain Rust's ownership model in two sentences"
clive agent "Refactor error handling and add tests" \
--files src/main.rs \
--verify "cargo check -q"