Launching today

Omi CLI
An AI coding agent in your terminal, on 13 models
33 followers
An AI coding agent in your terminal, on 13 models
33 followers
Omi can build anything, now from your terminal. Omi CLI reads your repo, shows every edit as a diff you approve, and runs tests until they pass. New in 0.5.0: parallel subagents, browser testing, background and cloud agents. Free: npm i -g @builditomi/cli


















Free Options
Launch Team

ZethosThe ledger you and your AI work from
Promoted
Hey Product Hunt! I'm Om, the maker of BuildItOmi.
Our idea has always been simple: Omi can build anything. Omi CLI brings that to your terminal.
What it does:
Reads your repo, makes a plan, and shows every edit as a diff you approve (yes / no / always)
Runs your tests and keeps going until they pass
Auto picks the model for each turn from 13 models (DeepSeek, Gemini, Gemma, gpt-oss, GLM), so everyday edits stay cheap and hard problems get a stronger model
/undo puts back every file the last task changed, even without git
One plan in USD with clear 5-hour and weekly limits, and a Free plan to try it
New in 0.5.0 (out today on npm):
Subagents that edit in parallel, each in its own git worktree, merged back three-way
Built-in browser testing: Omi opens your app in headless Chrome, clicks through it and reports console errors and failed requests
Background agents (omi bg "task") that keep working after you close the terminal
Cloud agents (omi cloud "task") that run on your repo's GitHub Actions and open a pull request
Honest numbers: on our first public benchmark (10 small tasks, judged by hidden tests the agent never saw) Omi CLI solved 10 out of 10. It is a small sample, and the full runs are on cli.builditomi.com/benchmark.
Try it: npm i -g @builditomi/cli, then run omi in any project.
I would love your feedback: which task should Omi CLI get better at next?
once /test allows npm test for the session, what happens if it later edits the test script in package.json?
13 models in one terminal agent is handy. Can I set a cheaper model for small edits and switch to a bigger one for refactors, or is it a manual switch each time?