kodwai

Kodwai is a developer challenge platform for solving coding problems with your own AI agent on your own machine, then scoring and ranking full sessions on a public leaderboard.

AI Agent Development AI Developer Tools Programming Agent

Visit Website

Developer challenges scored through AI-agent workflows

Kodwai is a platform for developers to solve coding challenges with an AI agent on their own machine. The homepage names Claude Code, Cursor, and Codex as supported agents and presents the product as a way to test how well you direct an agent through real work, rather than how well you memorize interview puzzles.

The workflow starts from the CLI: pick a challenge, download the problem files and tests, solve the task locally, and submit the session for scoring. Kodwai then evaluates the run using Direction, Outcome, and Lift, with evidence taken from prompts, recovery, commits, and test runs before placing the result on a public leaderboard.

Core capabilities

Challenge library with filters

Browse ticket-sized coding problems across categories and difficulties, then choose a challenge that matches the kind of work you want to practice.

CLI-based challenge start

Start challenges from the terminal with the Kodwai CLI, which downloads the problem statement, starter files, and tests before initializing a git repository and timer.

Local agent workflow

Work locally on your own machine with your own AI agent instead of inside a controlled sandbox, using the editor and workflow you already have.

Session packaging and submission

Submit the full session for scoring, including your code, git history, test runs, agent transcript, and elapsed time.

Evidence-backed scoring

Receive scores for Direction, Outcome, and Lift, with per-signal evidence that shows why each axis scored the way it did.

Leaderboard and profile

Compare runs on a public leaderboard and keep a public profile tied to your scored sessions.

Practical ways to use Kodwai

Practice agent steering
Use Kodwai when you want to practice the real workflow of directing an AI agent through a coding task, including prompting, verification, and recovery when the agent goes wrong.
Evaluate prompt quality
Use it to compare how different approaches to the same problem score, especially when one prompt gets tests passing but shows weak judgment or limited verification.
Prepare for coding interviews
Use the challenge library for interview-style preparation on ticket-sized backend, frontend, security, or other shipping work without switching to whiteboard puzzles.
Track progress publicly
Use the leaderboard and public profile to track scored runs over time and share a visible record of your challenge results.

Pros and Cons

Pros

Supports a local workflow on the developer's own machine.
Accepts familiar AI agents named on the homepage, including Claude Code, Cursor, and Codex.
Captures more than final output by scoring prompts, recovery, commits, and test runs.
Provides per-axis scoring with evidence so users can see why a run received its score.
Publishes results on a leaderboard and profile for comparison over time.

Cons

The source does not provide a full integration matrix or compatibility details beyond the named agents on the homepage.
Pricing and plan details are not shown in the collected page text, aside from language suggesting the product is fully free.

FAQ

What does Kodwai do?

Kodwai is a platform for solving coding challenges with your own AI agent on your own machine. The source shows support for Claude Code, Cursor, and Codex, with challenges started from the CLI and submitted for scoring.

How do you use it?

The rendered page shows a CLI flow: start a challenge from your terminal, download the problem files and tests, solve locally, then submit the session for scoring.

Who is it for?

Kodwai is aimed at developers who want to practice directing an AI agent through real coding work, rather than solving abstract puzzles or whiteboard-style problems.

What does Kodwai score?

The source describes scores based on Direction, Outcome, and Lift, with evidence drawn from prompts, recovery, test runs, commits, and the agent transcript.

Does it require a sandbox or special environment?

The page explicitly describes the product as working on your own machine with your own agent and says there is no sandbox to fight. It does not provide integration or environment details beyond the agents named on the homepage.

Quick Facts

Category: Developer Tool
Platform: Web app with CLI workflow
Primary users: Developers practicing AI-agent coding workflows
Source domain: kodwai.com
Notable workflow: Pick challenge → run CLI → solve locally → submit → get scored