Koodos installs onto a machine and gets full access to it — your files, your terminal, your projects. Ask it something from any of your devices and it answers from the one that has the answer.
Windows — PowerShell:
irm https://api.kprcli.com/dl/install.ps1 | iex
macOS / Linux:
curl -fsSL https://api.kprcli.com/dl/install.sh | bash
No key to paste. A window opens, you sign in with your email in the browser,
and the machine joins your fleet. Then it opens
localhost:6045 — your assistant, running on your own computer.
Reads, writes and edits files anywhere your account can, and runs any command you could run in a terminal. It builds, tests and checks its own work.
Pick a folder and it sees the tree, the git branch, uncommitted changes and the README. “Tell me about this repo” means the one in front of it.
Install it on the desktop, the laptop, the server. Each one appears in the panel — online, paused, or unlinked in one click.
Every model on OpenRouter, or OpenAI, Groq, or a local Ollama. Bring your own key and Koodos takes nothing, or run on ours and pay per token.
$ koodos "what's eating disk on this machine?"
run command wmic logicaldisk get size,freespace,caption
run command du -sh * | sort -h | tail -20
$ koodos "fix the failing test in cli/"
search files test_
read file cli/test_pi_launch.py
edit file cli/koodos_cli.py
run command python -m pytest cli/ -q
Or open localhost:6045 and type. Attach a screenshot, pick a model,
switch folders, watch every step it takes.
.koodos-bak copy and commands time out, but there is no confirm
prompt — treat it like a very fast pair programmer with your keyboard, and keep
your work in git.
| koodos link | connect this machine (re-run to update) |
| koodos | open the assistant |
| koodos "…" | one-shot task, right in the terminal |
| koodos devices | every machine on your account |
| koodos pause · resume · unlink | control another machine |
| koodos status | what's linked, running and answering |
The same inference Koodos uses, on its own: an OpenAI-compatible endpoint for
kx30b (qwen3-coder-30b class) served from private hardware. No prompt
logging, no subscription — prepaid credits at
$1 per 1,000,000 tokens, input and output, one price.
curl https://api.kprcli.com/v1/chat/completions \
-H "Authorization: Bearer kpr_YOURKEY" \
-d '{"model":"kx30b","messages":[{"role":"user","content":"hello"}]}'
Create keys from Settings inside the app, or start one from the terminal:
curl -X POST https://api.kprcli.com/v1/signup -d '{"email":"you@example.com"}'
100,000 tokens free to start. Every key on an account spends from the same balance, so you can make one per project. Top up with crypto — request an invoice for $1–$1000 and the credits apply once it confirms.
| One model | kx30b on the hosted API. Bring your own key for anything else. |
| Small hardware | A private rig, not a datacentre. Requests queue under load. |
| No confirm step | It acts, then tells you. Undo is .koodos-bak and git. |
| Your machine, your rules | Narrow it to one folder or switch commands off in Settings. |