Choose your scope.
Exact file paths, including the relevant tests. No automatic repository upload. Common credential patterns are screened locally.
IMIKWON.
Capable coding.
Open to everyone.
Build with Kimi K3.1 Pro. Choose the code it sees, inspect what it changes, and keep the cost in view.
YOUR CODE. YOUR CALL.
01 Select the files
02 Generate + self-review
03 Inspect the diff
04 Apply + run your tests
$0IMIKWON software cost
Provider usage billed separately
Provider usage billed separately
A working command-line alpha. Offline and mocked checks pass. Live Kimi output and cost per completed task await validation. See the evidence
FREE TO USE✳FREE TO MODIFY✳FREE TO BUILD UPON
START SMALL. MAKE SOMETHING WORK.
One minute.
No key needed.
Try a tiny broken calculator. The offline demonstration runs a failing test, applies a supplied fixture patch, then runs the test again.
This is a predetermined demonstration of the workflow. It makes no AI call and does not measure Kimi's coding ability.
Install Python 3.10+ and Git, then run:
git clone https://github.com/Albertito-AI/IMIKWON.git cd IMIKWON python -m imikwon demo
Use python3 if that is your Python command.
BROWSER WALKTHROUGH · SCRIPTED FIXTURE
def add(a, b):
return a - bTest expects add(2, 3) = 5. The current result is -1.
REAL TASKS. DELIBERATE STEPS.
The model proposes.
You decide what ships.
Look at the change.
Kimi produces a proposal and reviews it in a second pass. You inspect a diff. Source hashes stop stale changes being applied.
Run your checks.
Apply the proposal explicitly, then run your chosen local test command. Keep its exit status and output alongside the cost record.
Self-review is one model checking its own work. Your tests and code review remain essential. Local tests run with your user permissions.
SPEND LESS BY SENDING LESS.
Small context.
Visible cost.
IMIKWON charges nothing. Your provider bills your API usage. Focused files, limited output and an optional single pass help you control how much work you request.
The calculator uses K3.1 Pro's published USD rates checked on 5 November 2026: $2 per million uncached input tokens, $0.20 cached input, $2 five-minute cache writes and $10 output.
Illustrative usage, before taxes. Reasoning also consumes output tokens. A real task may need several attempts. This is not a measured task cost or a promise of savings.
Check current provider pricesREADY FOR YOUR OWN PROJECT?
Bring a task.
Bring your own key.
Install IMIKWON from the repository into a Python virtual environment. Get a Moonshot API key and configure it locally. Choose a source file and its test, then set your cost ceiling.
The command-line tool sends those files to your chosen provider. Keep personal information and confidential code out unless you have permission to share them.
Open the setup guideimikwon propose "Fix the bug and add tests" \ --file src/calculator.py \ --file tests/test_calculator.py \ --input-ceiling 6 \ --output-price 10 --budget 1
After installation and key setup. Replace the file paths with yours. The $6 input ceiling conservatively includes the higher cache-write rate; it is not the standard input price. Local estimates cannot guarantee the provider's bill.
CLEAR TERMS. ROOM TO IMPROVE.
What you are getting.
Is it really open-source?
Yes. IMIKWON's original code and documentation use the MIT licence, including commercial reuse. Kimi is an upstream model with separate terms. We have not verified K3.1 Pro's downloadable checkpoint and specific licence, so this release uses its documented API and distributes no model weights.
Can I use it completely free?
The software and offline demo are free. Live model calls require your own provider account and are billed by that provider. There is no IMIKWON subscription, markup or free hosted inference allowance.
How much coding can this release handle?
Bounded tasks across up to 24 selected UTF-8 files, 64 KB per file and 160 KB total context. It can create and modify files. It does not execute model-selected commands, delete files, commit or push automatically. A full browser IDE and autonomous repair loop are future work.
Has the quality been independently proven?
No. The software's local and mocked tests pass, but live Kimi inference, independent security review and comparative cost or coding-quality benchmarks remain outstanding. The repository records these limits openly.
A PUBLIC PROJECT BY ALBERT TAN SIN HONMore people
More people
should get to build.
Try it. Inspect it. Adapt it. Help make the next release better with a reproducible test, a useful correction or a measured result.
Build with IMIKWON